The Proof Date, The Autonomy Dial and The Known Number — prepared by Edmund Bogen
Last January this mastermind made 22 specific claims about AI in 2026. We went back and graded every one against what actually happened: 3 happened, 6 partly happened, 7 cannot be called yet, and 6 were wrong. The technology calls mostly landed. What the misses taught us is the reason this session exists. This session sorts 2027 into what to expect, what to ignore and what to set up before January. You will leave with three systems: a Proof Date for every AI claim you hear, an Autonomy Dial for every AI tool you have switched on, and a Known Number rule for every request that touches money, identity, access or contract terms.
Last January this mastermind made 22 specific claims about AI in 2026. We went back and graded every one against what actually happened: 3 happened, 6 partly happened, 7 cannot be called yet, and 6 were wrong. The technology calls mostly landed. What the misses taught us is the reason this session exists.
We will master 3 systems live, together:
1 · The Proof Date — Check it, or ignore it. Every AI claim gets a number that would prove it wrong and a date to check it. A claim without both goes on the ignore pile.
2 · The Autonomy Dial — Decide what AI may do. One level per task: off, read, draft, act after your sign-off, or act alone for low-stakes, reversible, logged work. Revoke anything above its level.
3 · The Known Number — Call the number you have. Any request that changes money, identity, access or contract terms is confirmed on a number you already had, never one in the message.
Register for the SessionEvery page of the 53-page course study is on this page, in order — the PDF is the same content, formatted to print. Jump to any section:
Three systems, and the three skills underneath them: checking a claim against a date, deciding what a tool may do, and confirming a request on a number you already had.
| # | Module | The skill it teaches | Pages |
|---|---|---|---|
| 1 | The scorecard | Grading January's 22 claims, and why the statistics failed | 4–9 |
| 2 | Expect: clients, portals | Consumer surveys, and where each portal's AI sends a buyer | 10–14 |
| 3 | Expect: tools, brokerages, rules | The AI you already own, the new rules, and the 2027 dates | 15–21 |
| 4 | System 1: The Proof Date | Checkable calls, the ignore pile and the 2027 call sheet | 22–26 |
| 5 | System 2: Autonomy Dial | Inventory, one level per task, revoke, re-check | 27–31 |
| 6 | System 3: Known Number | The rule, the client note, bots for clients, your twin | 32–36 |
| 7 | Before January | A December plan and four 2027 check dates | 37–38 |
| ✓ | Assessment and appendices | Answer keys, reference card, prompts, practice claims, questions, sources | 39–52 |
Six exercises sit where the skill was just taught. All of them use sample data or real quoted claims, each labeled, and most differ from the worked examples. Do them on paper first, then check Appendix A. The keys give the reasoning, not just the answer.
This session builds on earlier workbooks instead of repeating them. Use these when you need them.
| When you need… | Use this | From the workbook |
|---|---|---|
| What clients’ AI is telling them, monthly | The AI Mirror Audit; the Objection Forecast | My Client Asked ChatGPT (Aug 26, 2026) |
| To be the agent AI names | The Authority Answer Engine; the AI Citation Format | ChatGPT just picked one realtor in your city — by name (May 6, 2026) |
| Buyers who arrive with Zillow’s AI answers | The AI Translator Framework | Zillow’s $2.6B Bet Against You (Apr 8, 2026) |
| Portal and MLS rules for private and pre-market listings | The Three-Path Map | Private, Preview, or Public? (Nov 11, 2026) |
| What never goes into an AI tool; fair-housing review of AI copy | The Data Line; the Fair Housing Firewall | Keeping AI Compliant (Sep 23, 2026) |
| What a scheduled AI task can see; MLS data and your license | Module 1 capability table; Module 2 data categories | The Sunday Night Workflow (Nov 4, 2026) |
| AI voice calls, recording and AI notetakers | Lessons 2.10 and 7.3 | The Golden Handcuffs Conversation (Nov 18, 2026) |
| Which AI tools to buy, skip or upgrade | The Shiny-Object Graveyard; Upgrade Triggers | The Solo Agent’s Sub-$200/Month AI Stack (Jul 15, 2026; bogen.ai/masterminds/ai-stack/) |
| Sounding human when everyone uses AI | The Voice Print; the Slop Filter | Own Your Voice (Sep 2, 2026) |
| Capturing your playbook once | SKILL.md template; five starter skills | The Next Top 1%: Will win on Skills, Not AI (Feb 11, 2026) |
| Saying what you know versus estimate | KNOW / ESTIMATE / UNKNOWN | Complete Confidence (Oct 28, 2026) |
| Disclosing AI video and digital twins | The Video Ethics Card | Higgsfield for Realtors (Sep 11 and 25, 2026) |
| Safety rules for browser agents | Security Disclaimer; Security and Privacy Rules | Comet Browser Mastery (Oct 9, 2025); Agentic Browsers Mastermind (Oct 29, 2025) |
No earlier workbook teaches any of this: grading a prediction against what happened; portals and ads inside AI assistants; AI agents that act for your clients; deciding and auditing what your own AI tools may do; and confirming money and identity requests now that voices and videos can be faked.
Before anyone forecasts 2027, the 2026 forecasts get a grade. This module is the worked example for System 1, The Proof Date.
22 claims from “AI 2026 — The Business Survival Session” (Jan 7, 2026), graded against sources dated up to Nov 3, 2026.
The session ran on Jan 7, 2026. We graded what was written for it: the invite, the host notes, the agenda and two versions of the workbook, a 17-page edition and a 22-page draft. Each claim was checked against dated sources, all listed in Appendix F. Where a claim came from a survey or a vendor, we noted who paid for it. Rows S17–S20 appear only in the 22-page draft.
| Verdict | What it means in this book |
|---|---|
| Happened | It occurred, and a dated source shows it. |
| Partly | Part of it occurred, or it arrived in products but not in everyday use. |
| Not yet | It has not occurred, it cannot be measured yet, or nobody has measured it. Some of these may still come true. |
| Wrong | The cited source does not say it, no source exists, or the facts contradict it. |
We could not read a recording or transcript of the session, so we graded the printed claims, not what was said live. We could not confirm which of the two workbook files was the one posted for members. “Not yet” is not a polite “wrong.” And a claim that happened is not proof that the advice built on it was right.
If you filled in January’s Commitment Contract, it had three deadlines: the end of Week 1, the end of Month 1 and the end of Q1 2026. Grade yours the same way in Exercise 1 on page 9. Did each line have a number? Did it happen?
Quoted as written in January. “What happened” runs to Nov 3, 2026.
| # | January said | What actually happened | Verdict |
|---|---|---|---|
| S1 | “By the end of 2026, ~40% of enterprise apps will embed AI agents (up from <5% today).” | Gartner’s Aug 26, 2025 forecast said “up to” 40%, and of task-specific agents. No one has published a measurement of the 2026 share. The forecast date is Dec 31, 2026. | Not yet |
| S2 | The business world splits into those “deploying AI agents to handle 80–90% of tactical work” and those watching. | A split exists, but deployers are a small minority. McKinsey: 6% of respondents are “AI high performers,” unchanged from 2025. Nothing measures anyone at 80–90%. | Partly |
| S3 | “79% of organizations have adopted AI agents (PwC 2025).” | Of 308 U.S. executives PwC surveyed (Apr 22–28, 2025), 79% said agents are “already being adopted in their companies.” The Census Bureau found 19.8% of U.S. businesses used any AI in a two-week window (May 2026). | Wrong as framed |
| S4 | “66% of AI agent users report measurable productivity gains (PwC).” | Accurate to PwC’s adopters. In 2026, 80% of McKinsey respondents who use AI said it raised their productivity, but only 37% saw any EBIT impact, about the same as 2025. | Partly |
| S5 | “6-10% average revenue increase from agentic AI adoption.” | In Google Cloud’s 2025 study, 6–10% was the most common bracket among executives who reported any gain. It is not an average. Google sells AI. | Wrong as worded |
| S6 | “85 million jobs will be displaced by 2026 (World Economic Forum).” | WEF’s 2020 report said about 85 million by 2025, alongside about 97 million new roles. | Wrong |
| S7 | Agents become “digital employees” that handle “80-90% of tactical operations independently.” | The products shipped: Cowork, ChatGPT Work, computer use, browser agents. Every vendor kept a human approval step. In reports of McKinsey’s 2026 survey, 40% of respondents at $1B+ organizations and 22% at smaller ones said they were scaling agents. | Partly |
| S8 | “2026 is about agent orchestration… through protocols like MCP”; orchestration “becomes the competitive moat.” | The plumbing arrived. MCP moved to the Linux Foundation’s Agentic AI Foundation (Dec 9, 2025). MCP Apps launched Jan 26, 2026. Zillow serves ChatGPT and Claude through one MCP connection (Oct 21, 2026). A “moat” cannot be measured. | Partly |
| S9 | Agents from different vendors “understanding each other’s intent and negotiating terms.” | No evidence found of this in everyday business tools. | Not yet |
| S10 | Agent Skills as an open standard, with prebuilt skills from Canva, Notion, Figma and Atlassian. | Published Dec 18, 2025. By Oct 22, 2026, Anthropic reported more than 70 agent platforms using it, including ChatGPT, and nearly 280 million installs. Those are Anthropic’s own counts. | Happened |
| S11 | “‘AI Studios’ emerge as the winning model.” | This was PwC’s own prediction. No measurement of it was found. | Not yet |
Sources by row: Gartner, Aug 26, 2025 (S1); McKinsey State of AI, Aug 25, 2026, via The Register (S2, S4, S7, S12); PwC AI Agent Survey, May 2025 (S3, S4); U.S. Census Bureau, May 26, 2026 (S3, S14); Google Cloud ROI of AI, Sep 2025, via Virtualization Review (S5); WEF Future of Jobs 2020 and 2025 (S6); Linux Foundation, Dec 9, 2025; The Register, Jan 26, 2026; Zillow, Oct 21, 2026 (S8); Anthropic, Oct 22, 2026 (S10).
The last eleven rows include most of the statistics. That is where the grades fall.
| # | January said | What actually happened | Verdict |
|---|---|---|---|
| S12 | “Experimentation budgets disappear.” | ROI pressure, yes: McKinsey’s 2026 report is titled “On the road to ROI,” and 20% say AI operating costs limited their use. Gartner’s Aug 28, 2026 abstract says investment “remains strong.” Budgets did not vanish. | Partly |
| S13 | “80/20 rule where technology delivers 20% of value, redesigned work delivers 80%.” | A PwC framing, not a testable forecast. No measurement found. | Not yet |
| S14 | “Most businesses won’t figure out agentic AI until late 2026 or 2027.” | Held. Census: 19.8% of U.S. businesses used any AI (May 2026). NAR: 21% of Realtors do not use AI at all (Sep 22, 2026). | Happened |
| S15 | “12-18 months of operational advantage before competitors catch up.” | Cannot be measured. The tools spread fast: GPT-6 reached every ChatGPT plan, paid and $0, on Oct 7, 2026. The gap that lasted was habit, not access. | Not yet |
| S16 | “AI high performers are THREE TIMES more likely to see transformative business impact.” | McKinsey’s 2025 report said high performers are about 3x more likely to say they intend to use AI for transformative change. It did not say “see impact.” | Wrong as worded |
| S17 | “The top 1% of real estate professionals are already deploying AI agents that handle 80-90%…” (22-page draft) | No source found anywhere. | Wrong unsourced |
| S18 | “Annual AI tool costs… typically $240-$600.” (22-page draft) | Entry plans held at about $20 a month. Agent features often launched first on $100–$200 a month plans, which is $1,200–$2,400 a year. | Partly |
| S19 | “Expected increase in lead conversion rate… typically 5-15%.” (22-page draft) | No source found. Filed as Not yet, not Wrong, because it was a planning estimate in a worksheet blank, not a cited statistic. | Not yet |
| S20 | “AI won’t replace real estate agents. But agents who master AI will replace those who don’t.” (22-page draft) | Untested. No dataset found shows AI changing the number of agents in 2026. NAR: 55% of Realtors say AI helped their business; 11% say it hurt, up from 4%. | Not yet |
| S21 | Host notes: “lead with… 40% revenue lift numbers.” | No source for a 40% revenue lift found. | Wrong unsourced |
| S22 | Tool integration “through protocols like MCP” to CRM, email and documents. | Happened in products. ChatGPT Work launched with connectors to email, drives and CRMs (as reported). Zillow connects to ChatGPT and Claude through MCP. | Happened |
Sources by row: McKinsey via The Register and Gartner abstract, Aug 2026 (S12); Census, May 26, 2026; NAR 2026 REALTORS® Technology Report, Sep 22, 2026 (S14), and via Real Estate News and Inman (S20); OpenAI, Oct 7, 2026 (S15); McKinsey 2025 (S16); Anthropic and OpenAI pricing via vendor pages and trade press (S18); Android Authority, Jul 9, 2026; Zillow, Oct 21, 2026 (S22).
This is the one page about what we got wrong. It is here because the lesson runs through the rest of the book.
Several statistics that circulated widely in January, including in our own January workbook, did not hold up when we traced them to their source. Here they are, with what the source said and what to use instead.
| The January line | What the source actually said | Use instead |
|---|---|---|
| “85 million jobs will be displaced by 2026” (WEF) | WEF, 2020: about 85 million jobs may be displaced by 2025, while about 97 million new roles may emerge. | WEF, Jan 8, 2025: 170 million new roles and 92 million displaced by 2030. |
| “79% of organizations have adopted AI agents” (PwC) | 308 U.S. executives, Apr 22–28, 2025, saying agents are “being adopted in their companies.” | Census, May 26, 2026: 19.8% of U.S. businesses used AI in a two-week window. |
| “6-10% average revenue increase” | Google Cloud, Sep 2025: the largest bracket among executives who reported a gain. Google sells AI. | No revenue figure. Promise hours you can count, not revenue. |
| McKinsey: high performers 3x more likely to “see” impact | About 3x more likely to intend transformative change. | McKinsey, Aug 25, 2026: 6% are high performers, unchanged. |
| “40% revenue lift”; “top 1% automate 80–90%” | No source found for either. | Drop both. |
Each failed number follows a pattern. It came second-hand. It lost a qualifier on the way: “up to,” “by 2025,” “intend.” Or it came from a company that sells the thing being measured. And none of them came with a date to check it.
Dated examples, not recommendations. Products change, and two of these were retired or set for retirement within the year.
| Date | What happened | Why an agent should care |
|---|---|---|
| Dec 18, 2025 | Agent Skills published as an open standard | A playbook written once can travel between assistants |
| Jan 12, 2026 | Claude Cowork preview: an agent that works in a folder you choose; on $20 Pro from Jan 16 | January’s “digital employee,” with approval steps |
| Jan 16 · Feb 9 | ChatGPT Go at $8 a month in the U.S.; ad tests began Feb 9 for some U.S. users of the $0 and Go plans | Sponsored answers can sit beside housing questions |
| Jan 26 | MCP Apps: interactive apps inside chats, in Claude and then ChatGPT | Why a Zillow search can appear as a card in a chat |
| Jan 28 | Chrome auto browse for Google AI Pro and Ultra subscribers in the U.S. | Browser agents can fill forms on sites you use |
| Feb 9 | SecurityScorecard found 40,214 exposed instances of OpenClaw, a viral personal agent | Hobby agents holding your logins can leak client data |
| Mar 2 | Claude memory extended to the $0 plan, with import from other assistants | Switching assistants no longer means starting over |
| Mar 5 | GPT-5.4 with built-in computer use | AI that clicks through sites. Lab scores are not your MLS |
| Mar 24 | OpenAI announced it would discontinue Sora in its consumer app and API | Do not build a process on one consumer app |
| Mar 25 | Zillow AI mode launches in beta | Buyers arrive having “talked” to Zillow |
| Jun 12 | After a U.S. export-control directive, Anthropic disabled access to two of its models globally | Keep prompts and skills portable |
| Jul 9 | ChatGPT Work: an agent for multistep jobs that can run for hours. OpenAI also said it would sunset its Atlas browser | Agent work moved into the main apps |
| Aug 26 | Claude in Chrome generally available on paid plans | Keep browser agents to trusted sites |
| Sep 14 | iOS 27 with Siri AI in beta, reportedly built on Google’s Gemini | Clients can ask about homes by voice |
| Oct 7 | GPT-6 to everyone in ChatGPT, on paid and $0 plans | Your client’s no-cost AI is now a top model |
| Oct 19–21 | Zillow connector in Claude; Zillow now in ChatGPT, Claude and Gemini | A buyer’s chat runs through Zillow’s lead path |
| Oct 22 | Agent Skills donated to the Agentic AI Foundation; “more than 70” platforms (Anthropic’s count) | January’s best call |
Anthropic; OpenAI Deployment Safety Hub; Simon Willison’s Weblog; Android Authority; 9to5Google; 9to5Mac; The Register; The Next Web; The New Stack; Engadget; SiliconANGLE (SecurityScorecard sells security services); Bloomberg Government and Quartz; Zillow; iPhone in Canada. Benchmark figures are vendors’ own lab tests.
“Most agents won’t discover this for 12–18 months” sounds like a forecast. It has no measure, so nobody can ever grade it.
At least 16 earlier AI workbooks carry an “early adopter advantage” box. Each names a window. None names a number or a check date. Several of those windows close around now.
| Session | The promise | Window closes | Gradable? |
|---|---|---|---|
| The Secret AI Arsenal (Sep 10, 2025) | “Most professionals won’t discover this tool for 6–12 months” | Sep 2026 | No measure |
| Stop Wasting 15 Hours Weekly on AI — Master Consistency Systems (Nov 19, 2025) | “95% of Claude users have never heard of Projects or Skills”; a “6–12 month advantage” | May–Nov 2026 | No source, no measure |
| DEEP RESEARCH: The 10-Minute Listing Appointment Killer (Apr 29, 2026) | “In six months this will be table stakes. Fewer than 3% of agents are running deep research” | Oct 29, 2026 | Needs a count nobody published |
| AI 2026 (Jan 7) and AI Stacking (Jan 21, 2026) | “12–18 months of operational advantage” | Jan–Jul 2027 | No measure |
| The Voice AI Receptionist (Jul 1, 2026) | “Less than 3% of… agents currently use voice AI… a 12–18 month window” | Jul 2027–Jan 2028 | Needs a repeat count |
For each claim, choose Happened, Partly, Not yet, Wrong or Cannot be graded. Then write the number that would prove it wrong and the date you would check. Use only facts in this module.
Answer key: Appendix A, page 40.
Every number here comes from a different publisher asking a different question. Read each one with its question and its sponsor.
| Publisher (what it sells), date | What it asked | What it found |
|---|---|---|
| Pew Research Center (nonprofit), survey Feb 17–23, 2026, n=5,119; published Jun 17 | Do you use AI chatbots? Have you used ChatGPT? | 49% use chatbots and 44% have used ChatGPT. 57% of adults under 50 use ChatGPT. No housing questions. |
| Bank of America (mortgage lender), Jun 23, 2026, n=2,000 | Used AI tools or chatbots for homebuying research in the past year? | 20%; 32% of Gen Z and 28% of Millennials. Top use among them: affordability, payments or closing costs (57%). |
| Veterans United (VA lender), Jun 1–17, 2026, n=859 | Used AI tools in your home search? | 45% of prospective buyers, up from 37% a year earlier. ChatGPT 33%, Gemini 20%. |
| LendingTree (loan marketplace), Aug 31, 2026, n=2,001 | Would you use AI for a housing task? Did you, in the past 12 months? | 72% would. 15% actually did. |
| Cotality (property-data vendor), Apr 16, 2026 | Do you trust AI to help find a home? | U.S. trust fell from 30% (2025) to 16% (2026). Across the four countries surveyed, 44% would pay a fee for a human expert to verify AI-generated housing decisions. |
| Realtor.com (a portal, released with its own AI search), Oct 9, 2025, n=1,000 | Do you use AI for real estate insights? (adults interested in buying or selling, surveyed Aug 7–8, 2025) | 82%. Agents were rated the most accurate source. 62% said agents make them smarter about the market; 61% said AI does. |
20%, 45% and 82% are not a trend line. They are three publishers, three questions and three samples. Say “Bank of America found that 20%…,” with the question, every time. Only Pew, of the publishers above, has no product to sell.
AI-written questions, AI summaries of listings, AI readings of contracts, and a payment estimate that is missing something.
| What they bring | Who measured it | What you do |
|---|---|---|
| A payment, affordability or closing-cost estimate | Top use among Bank of America’s AI users (57%) | Ask what it left out: taxes, insurance, HOA. Send rate questions to a lender. |
| A value read, or an “offer competitiveness” read | Zillow AI mode explains the Zestimate and analyzes offer competitiveness (Zillow, Mar 25, 2026) | Bring your comps. Say what the AI could not see. |
| AI summaries of listings; AI-written questions for you | 32% and 23% of LendingTree’s AI-for-housing users | Check each feature. Answer the questions: they show what worries the client. |
| An AI review of a contract or document | 18% of the same group | Legal questions go to an attorney. Say so plainly. |
| Zoning, ADU, lot-split, VA assumable or FHA 203(k) questions | Realtor.com, on what RealAssist users ask (Jun 2026, as reported) | “Let’s confirm that with the city or the lender.” |
Researchers including authors from the National Fair Housing Alliance audited seven AI models in four U.S. cities using fair-housing paired testing. They found racial steering in location recommendations. Its direction and size varied, and did not carry from one city to the next (Samad et al., “The Geography of Algorithmic Judgment,” arXiv 2606.06694, 2026; abstract read). When a client brings an AI shortlist, the fair-housing responsibility is still yours.
A buyer forwards you this chatbot answer. It is sample text written for this course, about a made-up listing.
List every claim in it. For each one, write what you would verify, who answers it, and the first sentence you would say to the buyer. There are at least five.
Answer key: Appendix A, page 40.
Before you rebuild your business around AI search, look at how buyers and sellers actually chose their last agent.
NAR, 2026 Home Buyers and Sellers Generational Trends (Apr 2026), covering purchases from Jul 2024 to Jun 2025; 6,103 responses. NAR is the trade association of agents.
88% of buyers and 91% of sellers used an agent. Only 1% found their agent through the agent’s social media. The report has no question about AI at all. So in NAR’s latest data, AI does not yet show up as a way buyers find agents. The open question for 2027 is whether the “friend” a client asks for a referral is now a chatbot. No survey measures that yet.
NAR’s 2026 Profile of Home Buyers and Sellers had not been released as of Nov 3, 2026. The last edition came out Nov 4, 2025, so it may arrive in November. Check whether it adds an AI question before you quote this page. Ignore the “22% of relocating buyers used AI” line you may see online: it comes from a vendor blog and cites a NAR report that had not been published.
| Date | What launched | Where the buyer goes for a tour or an agent |
|---|---|---|
| Oct 6, 2025 | Zillow app in ChatGPT, for U.S. users on the $0, Plus and Pro plans | Tours, agent contact and financing happen on Zillow |
| Feb 6, 2026 | Redfin app in ChatGPT (Redfin added conversational search on its own site Nov 13, 2025) | Redfin |
| Feb 17, 2026 | Homes AI on Homes.com, built on Microsoft’s Azure AI. CoStar is keeping it on its own platform. | Inquiries “will always go directly to the listing agent” (CoStar) |
| Mar 30, 2026 | Realtor.com app in ChatGPT, with listing previews only | Realtor.com “intentionally routes high-intent buyers back” |
| Jul 9, 2026 | Zillow Rentals inside Google Gemini, for rentals only | Zillow powers availability and tour booking |
| Oct 19, 2026 | Zillow connector in Claude | Tour requests and agent contact still run through Zillow |
Zillow (Oct 6, 2025; Oct 19 and 21, 2026); Real Estate News (Feb 6 and Mar 30, 2026); CoStar via RISMedia and HousingWire (Feb 17, 2026); Inman (Jul 9, 2026). We found no published usage numbers for any of these apps.
Company-reported numbers, read as company-reported. They still show you where the hand-off goes.
Zillow’s Q2 2026 shareholder letter (Aug 5, 2026) says AI-mode users spend more than three times as long on Zillow and contact an agent at nearly three times the rate of other users. Those are Zillow’s own unaudited figures. The same letter says “nearly all connections will be serviced by Preferred partners over time.” Preferred partners “pay when they close” (same letter). If you are not one, plan on Zillow sending you a shrinking share of its buyers.
Zillow reports its Q3 2026 results on Nov 4, 2026. Look for a new AI-mode share and a new Preferred share before you quote either number on Nov 25.
What Google and OpenAI document, what two vendor studies found, and what nobody has shown.
| What to do | Who says so |
|---|---|
| Be indexed and eligible for a snippet on Google. Allow crawling. Keep structured data that matches the visible page. Keep your Business Profile current. | Google Search Central, “AI features and your website”: “no additional requirements to appear in AI Overviews or AI Mode, nor other special optimizations” |
| Allow OAI-SearchBot in robots.txt and in any bot protection on your site | OpenAI’s developer docs: sites that opt out do not appear in ChatGPT search answers |
| Keep your Google Business Profile complete | BrightLocal (a local-SEO vendor), Aug 26, 2026: Business Profiles were 28.63% of citations in the local AI answers it studied. Real estate was not a separate category. |
| Keep your own website substantive | Steady Demand (a local-SEO vendor), via MarTech, Aug 20, 2026: business websites were close to 60% of Gemini’s local citations |
The same study carries the warning. ChatGPT and Gemini agreed on the top business only 4.2% of the time, and Gemini repeated its own top pick only about 7% of the time. Results move. We found no announcement from OpenAI, Google, Anthropic or Perplexity of a feature that ranks individual agents by name. Anyone selling a guaranteed way to become “the agent ChatGPT recommends” is selling something no AI company documents.
| The call | Wrong if | Check |
|---|---|---|
| Zillow sends more than 75% of its agent connections to Preferred partners by the end of 2026, as it targeted | Zillow reports 75% or less for Q4 2026 | Mar 31, 2027 |
| No general chatbot launches a feature that ranks individual agents by name | OpenAI, Google, Anthropic or Perplexity announces one | Jun 30 and Dec 31, 2027 |
| NAR’s next buyer survey still shows referral as the top way buyers find an agent | Any other channel, AI included, beats it | When the report is out |
These are this course’s calls, not facts. The first two are on the full sheet, page 26. Add the third as one of your own.
2026 was the year your CRM, MLS and brokerage plugged into the assistant you already use.
NAR, 2026 REALTORS® Technology Report, Sep 22, 2026. Details below are from Real Estate News and Inman reports of the full study, which had 1,165 usable responses, about 1.6% of those invited. Read the small splits as direction, not precision.
Among Realtors who use AI, the top uses are listing descriptions (75%), social posts (56%) and emails and follow-up (52%). ChatGPT is used by 91% of them, Gemini by 42%, Copilot by 27% and Claude by 25%. The biggest barriers are the learning curve (63%) and cost (59%). 55% say AI has had a positive effect on their business. 11% say negative, up from 4% in 2025.
| Reported | Who | What it connects |
|---|---|---|
| Apr 24, 2026 | Flexmls MCP server (FBS) | AI tools such as ChatGPT and Gemini to live Flexmls data. Each MLS decides whether to switch it on. |
| Oct 5, 2026 | Lofty MCP server | Outside AI agents to Lofty data, for brokerages |
| Oct 8, 2026 | eXp “MCP First, AI of Choice” | Claude, ChatGPT or Gemini to eXp business data |
| Oct 13, 2026 | BoldTrail MCP (Inside Real Estate) | ChatGPT and Claude can pull contact history, log notes, set follow-ups, and send emails and texts |
| Oct 16, 2026 | Real Geeks AI Connect | Your chosen chatbot can edit your Real Geeks website |
Dates are when RISMedia, HousingWire and Real Estate News reported the company announcements. Availability and pricing were mostly not stated.
Dated examples of what brokerages announced in 2026. Ask your own office what reaches you, and when.
| Brokerage | What it announced | Status going into 2027 |
|---|---|---|
| Douglas Elliman | Elli AI, an agent assistant app launched Oct 7, 2025, first in Florida: natural-language MLS search, branded client reports, market insights and drafting. On Jul 8, 2026 Elliman announced Elius, an AI company it owns, built on Google Cloud and Gemini. | A national Elli rollout to its 6,600 agents was announced for 2026; completion was not confirmed. No Elius product dates given. |
| Compass | An AI Assistant inside its Home Platform, July 2026: daily briefings, “Likely to Sell” flags, follow-up drafts. Compass completed its acquisition of Anywhere Real Estate on Jan 9, 2026. | Franchisees and affiliates are “slated to gain access in 2027.” |
| eXp | “MCP First, AI of Choice,” October 2026: agents connect the assistant they already use. | AGNT OS has an “early 2027 launch date target.” |
| Keller Williams | Command opened to more than 75 integrations, including Google Gemini (Feb 23, 2026). | Live |
| Real | Leo Pro (Oct 20, 2026): more than three dozen AI agents that check with the agent “for approval before important actions are taken,” paid per task with “Leo Tokens.” | Live for Real agents |
Douglas Elliman releases, Oct 7, 2025 and Jul 2026; RISMedia and HousingWire, Jul 20–21, 2026; Real Estate News, Jan 9, 2026; HousingWire and Inman, Oct 8, 2026; HousingWire, Feb 23, 2026; Real press release, Oct 20, 2026.
Edmund’s own brokerage is in the first row. Elliman agents: ask your office whether Elli is live in your state, and what client data may go into it.
| Plan | U.S. list price | Source and date |
|---|---|---|
| Claude Pro · Max · Ultra | $20 ($17 billed annually) · $100 · $200 a month | claude.com/pricing, read Nov 3, 2026 |
| ChatGPT Go · Plus · Pro | $8 · $20 · $100 or $200 a month (the $100 tier was added Apr 9, 2026) | Trade press; OpenAI’s own page could not be read |
| Google AI Pro · Ultra | $19.99 · $99.99 a month | one.google.com, read Nov 3, 2026 |
Entry is about $20 a month. Heavy agent use often needs a $100–$200 plan. ChatGPT’s $0 and Go plans now carry ads. Prices change, so check the vendor’s own page on the day you buy. For what to buy, skip or upgrade, use the AI Stack workbook (Jul 15, 2026; bogen.ai/masterminds/ai-stack/).
Virtual staging, sky swaps and removed power lines are not banned everywhere. Undisclosed ones increasingly are.
Coraly, via HousingWire, Jun 24, 2026. Q1 2026 images from Zillow, Redfin, Realtor.com and Homes.com. Coraly sells an image-compliance workflow.
| Where | The rule | In force |
|---|---|---|
| California | Label a digitally altered image on or next to it, with a link or QR code to the original (Bus. & Prof. Code §10140.8, AB 723). Lighting, cropping and straightening edits are exempt. | Since Jan 1, 2026 |
| California rentals | AB 2025 extends the same disclosure to rental ads (Civil Code §1940.11). | Jan 1, 2027 |
| Wisconsin | Disclose alterations, “including artificial intelligence,” that create “a false or misleading impression” (Act 69, Wis. Stat. §452.136(1m)). | Jan 1, 2027 |
| Florida: Stellar MLS | Mark “Virtually staged” in the photo description and field; remarks must begin “One or more photo(s) was virtually staged.” No exterior staging except unattached furniture or décor, no altered fixtures, no removing defects such as holes, wiring or damaged floors. Non-compliant photos are removed, with an automatic fine. | Now (rules page updated Sep 27, 2026) |
| Florida: MIAMI MLS | Photos must “accurately depict” the property, with no text, logos or watermarks on them (Rule 15.1.12); advertising must “present a true picture” (24.20). | Rulebook dated Aug 21, 2025 |
| Arizona: ARMLS | A “Digitally Altered” watermark plus the original; $200 fine after an education phase. | Fines from December 2026 |
| Florida licensees | Advertising that is “fraudulent, false, deceptive, or misleading” is grounds for discipline (Fla. Stat. 475.25(1)(c)). | Now |
| REALTORS® | Code of Ethics SOP 12-10 bars “misleading images.” The 2026 Code does not mention AI by name. | 2026 Code |
Florida has no altered-photo statute, and its 2026 AI bills died. Misleading ads are already a license violation there. In Miami, the text ban means a “Virtually Staged” stamp on the photo itself may conflict with Rule 15.1.12. On our reading, disclose in the remarks or the photo description, and ask MIAMI MLS compliance where it wants the label.
Only California and Wisconsin have altered-photo statutes as of Nov 3, 2026. New York’s bill (A11635) is in committee, and New Jersey and DC have proposals. Reports on CRMLS fines differ, so check its current FAQ. A staging tool’s feature list is not your MLS’s rulebook: one tool offered to Florida agents advertises lawn replacement, which Stellar bans.
FCC 26-67 takes effect Nov 25, 2026. If an AI tool calls or texts for you, set it up for this rule before it sends another message.
| Rule | What it means for an AI tool that calls or texts |
|---|---|
| AI voice is an artificial voice | FCC 24-17 (Feb 8, 2024): AI-generated voices, including clones, fall under the TCPA’s “artificial or prerecorded voice” rules. Telemarketing with one needs prior express written consent (47 CFR 64.1200(a)(2)–(3)). |
| Say who is calling | Artificial or prerecorded messages must name the business at the start and give a phone number (64.1200(b)). |
| Florida consent | Automated or recorded sales calls and texts need prior express written consent (Fla. Stat. 501.059(8)(a)). For texts, the consumer replies STOP first; suit is allowed if texts continue 15 days later. |
| Florida hours | No solicitation calls before 8 a.m. or after 8 p.m., and no more than three in 24 hours on the same subject. This applies to licensees (501.616(6); 501.604). |
| California | A prerecorded call that uses an AI voice must say so (Pub. Util. Code §2874). |
| Not a rule yet | The FCC proposed an AI-call disclosure rule in 2024 (FCC 24-84). It is still a proposal. Disclose anyway. |
| Superseded | The older Jan 31, 2027 date for part of this rule no longer applies. Do not teach it. |
This is not theoretical for agents. On Mar 13, 2026 a federal court certified a class against a Las Vegas Realtor over 1,983 prerecorded ringless voicemails to expired listings, sent without documented consent. A defense-attorney blog estimated the exposure at just under $3 million (TCPAWorld, Mar 17, 2026). It is a class certification, not a judgment. It was not an AI-voice case, but the same consent rule applies to AI voices.
As of Nov 3, 2026, the FCC’s docket showed no petition for reconsideration or stay of FCC 26-67. Petitions can still be filed until Nov 25, so ask your broker or counsel the week of Nov 23. The FCC’s companion proposal on faster honoring and a “revoke all” method takes comments until Nov 25 and replies until Dec 28, 2026.
There is no general federal AI-disclosure law. These are the state rules most likely to reach an agent’s chatbot, calls or decisions.
| State | What it requires | When |
|---|---|---|
| California | It is unlawful to use a bot that misleads people about being a bot in order to push a sale. Saying it is a bot is the safe harbor (Bus. & Prof. Code §17941). | Since Jul 1, 2019 |
| California | CCPA rules on automated decision-making: housing decisions made by software without real human review need a pre-use notice and opt-out and access rights. Applies only to CCPA “businesses,” such as those with revenue over $26,625,000. | Jan 1, 2027 |
| California | Ads using an AI “synthetic performer” need a clear disclosure (SB 1050). | Jan 1, 2027 (reported) |
| Colorado | SB 26-189: AI that materially influences decisions about housing or lending needs notice at the point of use, a plain explanation within 30 days of an adverse outcome, and a way to request human review. | Decisions from Jan 1, 2027 (reported); enforcement on hold in court |
| Maine | A chatbot may not lead a reasonable consumer to believe it is human unless the consumer is told it is not. | Since Sep 2025 |
| New York | Ads with a “synthetic performer” must disclose it when the advertiser knows: $1,000 for a first violation, $5,000 after (Gen. Bus. Law §396-b). | About Jun 9, 2026 |
| South Carolina | A licensee is responsible for all work produced with AI, as if the licensee did it directly (S.C. Code §40-57-820). | In force |
| Texas | TRAIGA’s chatbot-disclosure duty is for government agencies and health care, not agents. Intentional discrimination through AI is barred. | Since Jan 1, 2026 |
| Utah | Disclose AI when a consumer clearly asks, or at the start of “high-risk” interactions in licensed occupations. | In force; repeal set for Jul 1, 2027 unless extended |
| Florida | The 2026 AI bills (SB 482 and SB 2-D) died. No Florida Real Estate Commission AI guidance was found. License law, the FTSA, all-party recording consent and FDUTPA still apply. | — |
Washington is trying to limit state AI laws, but nothing has been overridden. Executive Order 14365 (Dec 11, 2025) directs lawsuits, funding conditions and a request to Congress. It does not itself void any state law. No federal AI preemption statute has passed. The Justice Department joined the challenge to Colorado’s law (xAI v. Weiser) on Apr 24, 2026.
Federal priorities moved in 2025 and 2026. The statutes underneath did not.
| Topic | What changed | What did not |
|---|---|---|
| Fair housing | HUD withdrew its 2024 guidance on advertising through digital platforms, effective Sep 17, 2025. Its 2024 tenant-screening guidance is not on HUD’s withdrawal lists, and its status is unclear. HUD has proposed removing its disparate-impact rule (Jan 14 and Aug 10, 2026). | The Fair Housing Act still bans ads that indicate a preference (42 USC 3604(c)), and HUD’s rule still covers ad targeting that denies groups information (24 CFR 100.75). The disparate-impact rule was still in the CFR on Oct 28, 2026. |
| Fake reviews | The FTC set aside its order against Rytr, an AI review-tool maker, on Dec 22, 2025. | The FTC’s rule on fake reviews and testimonials (16 CFR 465, since Oct 21, 2024) covers AI-generated reviews. Penalties run up to $53,088 per violation. |
| Copyright | The Supreme Court declined to hear Thaler v. Perlmutter on Mar 2, 2026. | Purely AI-generated material is not protected, and prompts alone are not enough. Your own photos, writing and real creative edits can be (Copyright Office, Jan 29, 2025). |
| Client data | — | NAR SOP 1-9: do not reveal client confidences. Consumer AI plans may train on your inputs by default; business plans usually do not (Anthropic’s terms are one verified example). |
Sample situations. You are a Florida agent unless the item says otherwise. For each, name the rule and the one thing you would do before going ahead.
Answer key: Appendix A, page 41.
Laws, rules and announced launches that carry a date. Each one is a check on your calendar, not a forecast.
| Date | What happens | Where | Source |
|---|---|---|---|
| Nov 4, 2026 | Zillow Q3 2026 results | Portal | Zillow notice |
| Nov 5, 2026 | Challenger’s October job-cuts report, including AI-cited cuts | U.S. | Challenger |
| Nov 25, 2026 | FCC 26-67 opt-out rule takes effect; comments due on the FCC’s companion proposal | U.S. | 91 FR 67214 |
| December 2026 | ARMLS altered-media fines ($200) begin | Arizona MLS | ARMLS |
| Dec 28, 2026 | Reply comments due on the FCC companion proposal | U.S. | 91 FR 67239 |
| Dec 31, 2026 | Gartner’s “up to 40%” enterprise-app forecast comes due (January’s S1). Zillow’s 75% Preferred target. Bright MLS AI Academy planned. | U.S. | Gartner; Zillow; HousingWire |
| Jan 1, 2027 | Wisconsin altered-advertising disclosure | WI | Act 69 |
| Jan 1, 2027 | California: rental altered-image disclosure (AB 2025), synthetic-performer ads (SB 1050), CCPA automated-decision rules | CA | Bill records; CPPA |
| Jan 1, 2027 | Colorado SB 26-189 applies to decisions made from this date (reported). Enforcement is on hold in court. | CO | Colorado legislature |
| Jan 4, 2027 | Freddie Mac condo reserve minimum rises (as taught in Condo Due-Diligence, Oct 7, 2026) | U.S. | Library |
| Early 2027 | eXp AGNT OS launch target | Brokerage | HousingWire, Inman |
| 2027 | Compass AI Assistant to franchisees and affiliates | Brokerage | RISMedia |
| Jul 1, 2027 | Utah AI Policy Act scheduled repeal, unless extended | UT | Secondary |
| ~Sep 2027 | NAR’s next technology report, if it follows its Sep 18, 2025 and Sep 22, 2026 pattern | U.S. | Our inference |
| Dec 31, 2027 | Gartner’s forecast date: over 40% of agentic AI projects canceled. CCPA risk assessments due for covered California businesses. | U.S.; CA | Gartner; CPPA |
| No date | FCC AI-call disclosure rule; HUD disparate-impact removal; FTC AI-accuracy policy statement; the Colorado court ruling; New York bill A11635 | — | Federal Register; dockets |
Pending items have no date at all, and announced launch dates slip. Check each item in the week it is due. Do not teach a rule before it takes effect, and do not teach a superseded date. This calendar is education, not legal advice.
Check it, or ignore it. That is the whole system. The rest of this module is practice.
This builds on two tools you already have. Complete Confidence (Oct 28, 2026) taught you to say KNOW, ESTIMATE or UNKNOWN. The Claim Check in Private, Preview, or Public? (Nov 11, 2026) taught you to test a quoted study. The Proof Date adds what neither does: a day on the calendar when the claim gets graded.
| Kind | Example | What to do |
|---|---|---|
| A measurement | “48% of Realtors use AI at least weekly” (NAR, Sep 22, 2026) | Keep it, with publisher, question and date. Note when it will be updated. |
| A forecast with a date | “Over 40% of agentic AI projects will be canceled by the end of 2027” (Gartner, Jun 25, 2025) | Put it on your call sheet. Check Dec 31, 2027. |
| A forecast with no number or date | “The tools will be table stakes within a year” | Rewrite it with a number and a date, or put it on the ignore pile. |
| A slogan | “Agents who master AI will replace those who don’t” | Ignore pile. Nothing could ever grade it. |
Each one is real and circulating. Each one counts something other than what it is used to say.
| The claim | Who said it, and what they sell | What it actually counts | Do this |
|---|---|---|---|
| “97% of agents use AI” | Delta Media Group (sells brokerage technology), 2026 | What “more than 100” brokerage leaders believe about their agents | Use NAR: 48% weekly or more |
| “61% of home searches start in ChatGPT”; “67% of buyers use AI to find their agent”; “only 8.4% of agents appear in AI answers” | FlyDragon (sells AI-search services and ChatGPT ad placement to agents), Q1 2026 | No published question wording, sampling or weighting. The 8.4% is a share of queries, and the sample includes its own clients | Ignore. If a client quotes it, name the source |
| “82% of buyers use AI” | Realtor.com (a portal launching its own AI), Oct 9, 2025 | 1,000 adults interested in buying or selling, surveyed over two days, about “real estate insights” | Only as “Realtor.com found…,” with the question |
| “37% would let AI buy their home” | LendingTree (loan marketplace), Aug 31, 2026 | A hypothetical. Only 15% used AI for housing in the past year | Always pair it with the 15% |
| “Half of buyers would skip the agent” | Veterans United (lender), Jun 2026 | 53% “comfortable” without direct human involvement, 25% “very”: a hypothetical, 400 respondents | NAR: 88% of buyers used an agent |
| “Nearly half of agents use AI daily” | Zillow (sells agent software), Feb 19, 2026 | No sample size or dates disclosed | NAR: 23% daily |
| “The first agentic AI operating system for real estate” | Lofty, Feb 2, 2026 | A self-description | Ask for a live demo on your own data |
Also on the pile: vendor performance claims you cannot check, such as hours saved, contact rates or “up to 70%” software savings, and any “formula” for being recommended by ChatGPT (page 14).
This module tests claims, not purchases. For buying decisions, use the Shiny-Object Graveyard and Upgrade Triggers in The Solo Agent’s Sub-$200/Month AI Stack (Jul 15, 2026; bogen.ai/masterminds/ai-stack/). One 2026 addition: buy what has shipped. Real Geeks’ Geek AI Voice reportedly slipped from an early-2026 target to “late 2026,” and Zillow showed an AI roadmap for Follow Up Boss with no dates.
Gartner expects over 40% of agentic AI projects to be canceled by the end of 2027, “due to escalating costs, unclear business value or inadequate risk controls” (Jun 25, 2025). It also estimated that only about 130 of thousands of “agentic AI” vendors are real. That is a forecast, not a count. If “agents disappointed us” headlines arrive in 2027, they will not mean every tool failed.
One real claim from 2026, taken through every step. Do this once by hand and the rest go fast.
“Zillow’s AI is going to send every lead to its paid partners.”
| Step | What you do | What you get |
|---|---|---|
| 1. Find the source | Look for the company’s own document, not a post about it. | Zillow’s Q2 2026 shareholder letter (Aug 5, 2026): 61% of connections go to Preferred partners; the goal is “more than 75%” by the end of 2026, and “nearly all” over time. |
| 2. Fix the claim | Shrink it to what the source says. | Not “every lead.” “More than 75% of connections by the end of 2026.” “Nearly all” has no date, so it cannot be checked yet. |
| 3. Write the number | What figure would prove it wrong? | A Q4 2026 Preferred share of 75% or less. |
| 4. Write the date | When will that figure be public, and when will you look? | We will look on Mar 31, 2027. If it is not out, look again Jun 30. |
| 5. Decide both outcomes now | What would you do if it is right? If it is wrong? | Right: most Zillow buyer leads reach agents through Preferred, so judge Zillow as a pay-at-close channel. Wrong: nothing changes this year; check again in June. |
The rule works on your own plans too. “I’ll use AI more in 2027” cannot be graded. A sample line that can: “By Mar 31, 2027, I will have sent 50 AI-drafted follow-ups that I edited myself, each logged in my CRM.” January’s Commitment Contract had dates. Next time, give every line a number as well.
This is the kind of text an AI tool can produce when asked for “2027 real estate AI predictions.” Find the errors before you read the table.
| # | The line | What is wrong | Page |
|---|---|---|---|
| 1 | “AI rewrites real estate” | A slogan. No number and no date. | 22 |
| 2 | “61% of home searches… ChatGPT” | FlyDragon’s vendor figure, with no published method. | 23 |
| 3 | “Zillow AI mode… live for everyone” | About 20% of signed-in users as of Aug 5, 2026. | 13 |
| 4 | “Google AI Mode shows homes for sale” | Not found in any Google source. Google’s listing ads are sponsored, mobile and MLS opt-in. | 13 |
| 5 | “6–10% average revenue increase” | A vendor’s survey bracket, not an average. | 7 |
| 6 | “Top agents… automate 80–90%” | No source. Realtors’ top AI uses are writing tasks. | 7 |
| 7 | “Florida passed an AI Bill of Rights” | It passed the Senate twice in 2026 and died both times. | 19 |
| 8 | “Opt-out rule… January 31, 2027” | It takes effect Nov 25, 2026. The Jan 31, 2027 date was superseded. | 18 |
| 9 | “Time to load your AI dialer” | AI-voice marketing calls need prior express written consent, and Florida adds its own. | 18 |
| 10 | “Half of buyers… skip the agent”; “the window closes” | A hypothetical from a 400-person lender survey. An urgency line with no measure. | 23 |
Five claims, as you might hear them (sample wording). For each, either rewrite it as a Proof Date (claim, number, date, source) or put it on the ignore pile and say why. Use facts from Modules 1 to 3.
Answer key: Appendix A, page 41.
These are this course’s calls, not facts. Each has a number that would prove it wrong. Grade it on Mar 31, Jun 30, Sep 30 and Dec 31, 2027.
| # | The call | Wrong if | Source to check | Check |
|---|---|---|---|---|
| 1 | Zillow sends more than 75% of agent connections to Preferred partners by the end of 2026, as it targeted | Its Q4 2026 share is 75% or less | Zillow shareholder letter | Mar 31 |
| 2 | Real Geeks ships Geek AI Voice, listed as “Coming Late 2026” | Still not shipped | realgeeks.com/geek-ai | Mar 31 |
| 3 | eXp launches AGNT OS in the first half of 2027, per its “early 2027” target | Not launched by Jun 30 | eXp; trade press | Jun 30 |
| 4 | No general chatbot launches a feature that ranks individual agents by name | OpenAI, Google, Anthropic or Perplexity announces one | Company announcements | Jun 30 · Dec 31 |
| 5 | NAR’s next technology report shows more than 48% of Realtors using AI at least weekly | 48% or less | NAR, expected around September | Sep 30 |
| 6 | Writing stays Realtors’ top AI use (listing descriptions were 75% in 2026) | A non-writing use ranks first | Same report | Sep 30 |
| 7 | Compass’s AI Assistant reaches its franchisees and affiliates in 2027, as announced | Not available to them by year end | Compass; your broker | Dec 31 |
| 8 | The FCC has no final AI-call disclosure rule by the end of 2027 (proposed Sep 2024) | A final rule is published | Federal Register | Dec 31 |
Grading rule: Happened, Partly, Not yet or Wrong, the same four verdicts as Module 1. Write the date you graded each call and the source you used. A call you grade “Not yet” stays on the sheet for the next date.
| # | The call | Wrong if | Source to check | Check |
|---|---|---|---|---|
| 9 | ||||
| 10 |
| Check date | Call 1 | Call 2 | Call 3 | Call 4 | Call 5 | Call 6 | Call 7 | Call 8 | Call 9 | Call 10 |
|---|---|---|---|---|---|---|---|---|---|---|
| Mar 31, 2027 | ||||||||||
| Jun 30, 2027 | ||||||||||
| Sep 30, 2027 | ||||||||||
| Dec 31, 2027 |
Last January called these digital employees. The products arrived in 2026, each with an approval step. The Autonomy Dial is how you use them.
| What vendors shipped in 2026 | The approval step they kept |
|---|---|
| Claude Cowork (from Jan 12, 2026) | “Deleting anything needs your approval” (Anthropic’s Cowork page) |
| ChatGPT Work (Jul 9, 2026) | Users can approve certain actions (Android Authority) |
| Real’s Leo Pro (Oct 20, 2026) | Checks with the agent “for approval before important actions are taken” |
| Keller Williams and Rejig.AI (Aug 31, 2026) | The agent approves listing and social content before it is published |
| CINC AI nurture inside Follow Up Boss (Aug 3, 2026) | Agents can review AI conversations and mute AI outreach from the lead profile |
None of them sells unattended handling of your client relationships. And few businesses run agents at scale: in reports of McKinsey’s 2026 survey (Aug 25), 22% of respondents at smaller organizations, and 40% at $1B+ organizations, said they were scaling AI agents.
Most agents can name their AI subscriptions. Few can name every connector, extension and scheduled task they have approved.
| Look here | What you may find | Dated examples (not recommendations) |
|---|---|---|
| Your AI assistant: settings, connectors, apps | Email, calendar, drive, CRM and portal connectors | Zillow connector in Claude (Oct 19, 2026); BoldTrail MCP (Oct 13, 2026) |
| Scheduled tasks | Jobs that run without you opening the app | Claude and ChatGPT scheduled tasks (The Sunday Night Workflow, Nov 4) |
| Browser extensions and agents | Tools that click and type in your browser | Claude in Chrome (generally available Aug 26, 2026); Chrome auto browse (Jan 28) |
| Desktop agents | Agents with access to a folder or your files | Claude Cowork; ChatGPT Work |
| Your CRM’s AI | Smart messages, AI nurture, AI texting | Follow Up Boss smart messages; CINC AI nurture; Geek AI Text |
| Voice | An inbound receptionist, an outbound dialer, voicemail drops | Your First AI Employee (Apr 15); The Voice AI Receptionist (Jul 1) |
| Your brokerage and MLS | Platform assistants and MLS connectors | Flexmls MCP (Apr 24, 2026); Compass AI Assistant; Elli AI |
| Things you built | Zaps, automations, skills, custom GPTs | AI Stack (Jul 15; bogen.ai/masterminds/ai-stack/); The Next Top 1% (Feb 11) |
In early 2026, SecurityScorecard found 40,214 publicly exposed instances of OpenClaw, a viral personal agent. Gateways often had no authentication, and keys and logins sat in plain text (SiliconANGLE, Feb 9, 2026; SecurityScorecard sells security services). Anthropic warned Cowork users about prompt injection and advised against giving it sensitive files. Our Oct 2025 browser sessions warned about the same risk. You cannot limit access you do not know you granted.
| Tool or connector | What it can see | What it can do | Who approved it, and when | Plan: consumer or business; training on or off |
|---|---|---|---|---|
The plan line matters. Anthropic’s consumer terms (effective Oct 8, 2025) allow training on your materials unless you opt out, and even then for feedback you give or material flagged for safety review. Its commercial terms do not allow training on customer content. We did not verify other vendors’ terms for this book, so read yours. The Data Line from Keeping AI Compliant (Sep 23, 2026) still decides what goes in at all.
The level belongs to the task, not the tool. The same assistant can sit at Read for your inbox and at Draft for your newsletter.
| Task | Highest level we suggest | Why |
|---|---|---|
| Summarize your own call notes or a closed file | Draft | You read it before it goes anywhere. |
| Follow-up texts and emails to leads | Draft, or act after sign-off | It speaks in your name. Accuracy and fair-housing review stay with you. |
| Automated marketing texts | Act after sign-off, with opt-out set up for FCC 26-67 | FCC 26-67; Florida’s FTSA (Golden Handcuffs, Lesson 7.3). |
| Answering your inbound phone line | Act alone only to take messages and book, within a written script, with AI disclosure and recording consent | Florida requires all-party consent to record (934.03). No prices, terms or advice. |
| Outbound AI-voice calls to prospects | Off, unless you hold prior express written consent | FCC 24-17; 47 CFR 64.1200; Fla. Stat. 501.059. |
| Posting listing copy or photos | Draft | MLS photo rules; SOP 12-10; the Fair Housing Act. |
| Editing your website | Act after sign-off | It publishes in your name. Real Geeks AI Connect is one example of this access. |
| Updating CRM records, tags and tasks | Act alone, if every change is logged and reversible | Low stakes. Read the log weekly. |
| Holds on your own calendar; sorting your own files | Act alone | Reversible and logged. |
| Anything touching money, identity, access or contract terms | Never above Draft | System 3, The Known Number. |
These levels are this course’s recommendation, not a rule. Your broker’s AI policy comes first. NAR offers members an AI policy template for brokers; it is a template, not a mandate.
Moving an AI voice or texting task up a level is a consent decision, not only a settings change. Check page 18 and your broker’s policy first. This is education, not legal advice.
Setting a level does nothing if the access stays wider. Revoking is the step people skip.
| If the access is wider than the level | Do this |
|---|---|
| A connector can send, but the task is Draft | Disconnect it, or reconnect it with read-only access if the vendor offers that |
| A scheduled task you no longer read | Delete it. A task nobody reads is unattended work. |
| A browser agent is logged into your MLS, bank or email | Remove those sites from its permissions. Keep it on public sites. |
| An old tool still holds your password | Change the password, and remove the tool’s access in that account’s security settings |
| AI outreach you cannot stop in two seconds | Find the mute or stop switch before the next hot lead. CINC’s mute inside Follow Up Boss is one example. |
| A consumer plan with training on | Turn training off, or move client work to a business plan your broker approves |
Use the same four dates as your call sheet: Mar 31, Jun 30, Sep 30 and Dec 31, 2027. Each time, rerun the inventory and compare it with last quarter’s. Ask three questions. What was added? What did I stop reading? What did a vendor change? Vendors add AI to products you already own: Lofty and Real both did in 2026, and Follow Up Boss now lists AI features in every plan.
| Symptom | Likely cause | Fix |
|---|---|---|
| Everything ends up at Act alone | Convenience creep | Apply the 2 a.m. test to each row again |
| A new feature appeared and nobody set a level | The vendor switched it on in an update | Add it to the inventory; it starts at Off |
| You cannot find what a tool did last week | No log, or a log nobody reads | Then it was never eligible for Act alone |
Sample data: one solo agent’s inventory, set and revoked. Your tools and levels will differ.
| Task | Tool (sample) | Was | Now | What changed |
|---|---|---|---|---|
| Inbox summaries | Assistant with an email connector | Could send | Read | Reconnected read-only |
| Lead follow-up drafts | CRM smart messages | Draft | Draft | Nothing |
| Market-update texts | CRM texting assistant | Act alone | Act after sign-off | Opt-out wording added; a stop now ends all automated marketing |
| After-hours calls | AI receptionist | Act alone | Act alone: messages and booking only | Script updated; AI and recording disclosed at the start |
| Website edits | Chatbot connected to the site builder | Act alone | Act after sign-off | Auto-publish turned off |
| Weekly listing report | Scheduled task | Act alone | Off | Deleted: unread for six weeks |
| Showing-request forms | Browser extension | Logged into MLS and bank | Act after sign-off; public sites only | MLS and bank removed from its permissions |
Seven tasks. Four came down a level. One was deleted. Nothing went up. That is a normal first pass.
A different agent’s inventory (sample data). For each item, write the highest level you would allow and one thing you would revoke or change.
Answer key: Appendix A, page 42.
Any request that changes money, identity, access or contract terms is confirmed on a number you had before the request arrived.
FBI Internet Crime Complaint Center, 2025 report (released April 2026), as reported by NAR’s REALTOR® Magazine and JD Supra. It was the first IC3 report with a dedicated AI section.
Read those numbers carefully. $275.1 million is all real estate fraud reported to IC3 for 2025, not AI fraud. The AI-tagged part was 115 complaints and $2.7 million. Voice cloning can make the “the wire instructions changed” call more convincing. And our own library taught members to build voice clones (Apr 15, 2026) and a photoreal Digital Twin (Sep 11, 2026), so the voice being faked could be yours.
| Trigger | What it looks like |
|---|---|
| Money | New wire or payment instructions; a changed payee; a request to redirect proceeds or a deposit |
| Identity | A “new phone number” or email for a client, lender or attorney; a request to confirm personal details |
| Access | A request to add someone to a transaction platform, share a login, or send documents to a new address |
| Contract terms | A request by message to change price, dates, credits, the closing attorney or the title company |
A rule a client first hears during closing week sounds like panic. The same rule in the first meeting sounds like a professional.
How we will handle money and changes. Before you send any money, call the title or escrow company on the number in this note to confirm the instructions. Do this every time, even if the request seems to come from me. If anyone asks you to change where money goes, who you are working with, or the terms of our deal by email, text or voicemail, do not act on it until you have called that person on a number in this note. If a call sounds like me but asks for something unusual, hang up and call me back at [your number]. I will do the same with you.
Your verified contacts: Title or escrow [name, number] · Lender [name, number] · Attorney [name, number] · Me [number]
Give it at the first meeting, to every client, and say you give it to everyone. Then it is a standard, not an accusation. Send it again with the contract, and once more the week before closing.
| Field | What goes in it |
|---|---|
| Who | Company and person, with role: title, escrow, lender, attorney |
| Number | The number, and where you got it: a signed document, your own records, or a number you looked up and typed yourself, not one you clicked in a message |
| Verified | The date, how (you called and confirmed something only they would know), and who did it |
| Changes | Any later change, how it was confirmed, and on which known number |
Do it once per company, not once per deal. Re-check the cards each quarter with your Autonomy Dial review. Keep them out of reach of any AI connector that can edit records: a contact card is exactly the record a bad actor wants changed. In the Autonomy Dial, it sits at Read at most.
Recording a call needs everyone’s consent in Florida (Fla. Stat. 934.03). If an AI notetaker is running on a verification call, say so first. Rules on recording vary by state. This is education, not legal advice; your broker and counsel own the specifics.
Clients are starting to hand tasks to AI. Decide now how you will treat a bot on the other end of the line.
The pieces are already announced. Google said in May 2026 that it can call businesses for a user in select categories, such as home repair, beauty and pet care, and announced always-on “information agents,” with apartment hunting as its example (Google, May 19, 2026). Rocket Money launched Rowan, “an AI financial agent,” in August 2026 (Rocket, Oct 29, 2026). Zillow’s connector in Claude lets a user request a tour or connect with an agent from inside a conversation (Zillow, Oct 19, 2026). The research behind this book found no AI negotiating terms for a buyer in everyday tools. Our expectation: the first contacts will be questions, bookings and document requests.
| The bot asks | You do |
|---|---|
| To book a showing | Ask: “Am I speaking with a person or an automated assistant?” Confirm the time with the client, or their agent, on a known number before you rely on it. |
| For client details, documents or a disclosure packet | Share nothing about any client until the person it claims to act for confirms on a known number that it acts for them, and what it may receive. Ask your own client what may be shared. |
| To change a date, price, credit or contact | The Known Number rule. No change by bot message, ever. |
| For your commission, terms or a quote | Say only what you would put in writing to anyone. Log it. |
| To “confirm” that the request came from your client | Call your client on the number you have. |
Date and time · who or what contacted you · what it said it was · what it asked for · what you shared · how you verified it.
If clients know exactly where your AI voice or video appears, a fake one anywhere else stands out.
Our library taught members to build a voice clone, with a consent script (Your First AI Employee, Apr 15, 2026), and a photoreal Digital Twin (Higgsfield for Realtors, Sep 11, 2026). Both are useful. Both mean a convincing fake of you is easier to make. The defense is a short public statement of where the real ones appear, and what they will never ask for.
“I use AI tools in my business. After hours, my AI assistant answers my office line and says it is an AI. Some of my listing videos use an AI version of me, and say so on screen. I will never use an AI voice or video to ask you for money, documents or changes to a deal. If anything that looks or sounds like me asks for that, call me at [your number].”
| Rule | What it says | Status |
|---|---|---|
| New York synthetic-performer law | Ads with an AI “synthetic performer” must disclose it when the advertiser knows. $1,000 for a first violation, $5,000 after (Gen. Bus. Law §396-b). | In effect about Jun 9, 2026 |
| California SB 1050 | A similar disclosure for synthetic performers in ads. | Jan 1, 2027 (reported) |
| FTC reviews and testimonials rule | Fake testimonials, AI-generated ones included, are banned. Up to $53,088 per violation (16 CFR 465). | Since Oct 21, 2024 |
| Copyright | Purely AI-generated material is not protected. Your own creative contribution can be (Copyright Office, Jan 29, 2025). | Current |
Whether your own twin counts as a “synthetic performer” under these laws is a question for counsel. A clear on-screen label answers it either way. The Video Ethics Card from Higgsfield for Realtors (Sep 11 and 25, 2026) still applies: never fake a person, and say when a video is AI.
Sample data. Three messages in two days, and what the Known Number rule does with each.
| When | What arrives | What happens next |
|---|---|---|
| Tue 4:40 p.m. | The buyer gets an email that looks like the title company’s, with “updated wire instructions” and a phone number “for questions.” | The buyer has the day-one note. She calls the title company on the number in it, not the one in the email. The instructions have not changed. Title, the buyer’s agent and the broker are told. |
| Wed 9:05 a.m. | The buyer’s agent gets a voicemail in the buyer’s voice: move closing to Friday, and send the closing statement to a new email. | The agent calls the buyer on the number on file. She made no such call. Nothing changes. Both events go in the log. |
| Wed 9:30 a.m. | An automated assistant calls the listing agent “for the buyer” asking for the HOA documents. | The listing agent asks whether it is a person, shares nothing yet, and calls the buyer’s agent on the known number. The buyer’s agent never sent it. |
No one had to judge whether the email, the voice or the bot was fake. The rule never asks that. It only asks which number to call.
Sample messages. For each one, name the trigger (or “none”), what you do, and which number you call.
Answer key: Appendix A, page 42.
Four weeks, three systems. Each week ends with something you can show your broker.
| When | Do this | Done means |
|---|---|---|
| By Nov 25 | Before any automated text or call goes out: an opt-out method named in every message (required if it is the only way to opt out); a stop on automated marketing ends all of it; suppression within 10 business days, as already required. Ask your broker the week of Nov 23 whether any petition or stay was filed. | Your tool’s settings match FCC 26-67. |
| Week 1 Dec 1–7 Proof Date | Grade your January Commitment Contract. Do Exercises 1 and 4. Copy the call sheet on page 26 and add two calls of your own. | A call sheet with dates on it. |
| Week 2 Dec 8–14 Autonomy Dial | Run the inventory (Lesson 5.2). Set one level per task. Revoke anything above its level. Do Exercise 5. | Every row has a level and a date. |
| Week 3 Dec 15–21 Known Number | Get your broker to approve your client note. Verify and store your title, lender and attorney contacts. Write your “where my AI appears” note. Do Exercise 6. | Note approved; contact cards stored. |
| Week 4 Dec 22–31 Rules and dates | Read your MLS photo rule. Check which Jan 1, 2027 rules apply in the states where you work. Put the four 2027 check dates on your calendar. Take Appendix E to your broker. | Four dates on your calendar. |
| Symptom | Likely cause | Fix |
|---|---|---|
| Every claim you hear lands on the ignore pile | You are skipping the rewrite step | Try to write the number first. Discard only if you cannot. |
| Your call sheet is all safe calls | You picked only things already announced | Add one call you are unsure of, with its number |
| The inventory keeps growing | Vendors add AI to plans you already have | That is what the quarterly check is for |
| A client is offended by the verification call | The rule arrived late in the deal | Give the note on day one, and say you give it to everyone |
| An AI tool adds a statistic you cannot trace | It may have picked it up from the web | Run the five questions; use the prompts’ NOT KNOWN rule; check by hand |
| Your broker has no AI policy yet | Common: in WAV Group’s Aug 19, 2026 survey, brokers rated their own AI readiness 5.75 out of 10 on average | Take Appendix E. NAR’s broker template is a place to start. |
Four dates a year. On each one, all three systems get checked together.
| Date | The Proof Date | The Autonomy Dial | The Known Number |
|---|---|---|---|
| Mar 31, 2027 | Grade calls 1 and 2. Check which Jan 1 rules took effect as written. | Rerun the inventory; set a level for anything new | Re-verify any contact that changed |
| Jun 30, 2027 | Grade calls 3 and 4. | Same | Same |
| Sep 30, 2027 | Grade calls 5 and 6, if NAR’s report is out. | Same | Same; refresh your client note |
| Dec 31, 2027 | Grade calls 4, 7 and 8, and your own. Write the 2028 sheet. | Same; retire what you stopped using | Same; update “where my AI appears” |
Every figure in this book carries a date so you can see when it has aged. NAR’s technology report has come out in September two years running. Census business AI figures, Pew’s survey and the portals’ quarterly letters all update during 2027. When a number on these pages is replaced, replace it in your scripts too.
Nobody is marking this but you. Tick only what you could do tomorrow, in front of a client or your broker, without this book open.
| ✓ | I can… | Module |
|---|---|---|
| Read the January scorecard and explain what each verdict means | 1 | |
| Say which January statistics failed, and the three reasons they failed | 1 | |
| Quote a consumer AI survey with its publisher, question, sponsor and date | 2 | |
| Explain where each portal’s AI sends a buyer, and what Preferred routing means for me | 2 | |
| Name the AI already inside the CRM, MLS and brokerage tools I pay for | 3 | |
| Explain FCC 26-67 and confirm my texting and calling tools follow it | 3 | |
| Name the altered-photo rules in my state and my MLS | 3 | |
| Turn any AI claim into a Proof Date, or put it on the ignore pile and say why | 4 | |
| Run the five questions on any statistic before I repeat it | 4 | |
| List every AI tool, connector, scheduled task and agent I have switched on | 5 | |
| Set one level per task, and revoke access above it | 5 | |
| State the Known Number rule and its four triggers | 6 | |
| Handle an AI agent calling for a client without sharing client data | 6 | |
| Name the four 2027 check dates and what I check on each | 7 |
| Item | My answer |
|---|---|
| The date I wrote my 2027 call sheet, and my own two calls | |
| My broker’s AI policy, or the date I asked for one | |
| My AI tools, and how I confirmed training is off | |
| The date I checked my texting tool against FCC 26-67 | |
| How many title, lender and attorney contacts I verified | |
| The date my broker approved my client note | |
| Who holds me to the four 2027 dates |
Your wording will differ. Check it for the idea, not the words. More than one verdict can be defensible if your reasoning names the evidence.
| Claim in the sample | Verify, and who answers | Your first sentence |
|---|---|---|
| “Priced 8% below market, so offer fast” | Your own comps. The AI could not see condition, concessions or the sales you would pick. “Offer fast” is pressure, not analysis. | “Let’s look at what actually sold nearby. Then you decide on timing.” |
| “About $2,350 at 6.2%” | A rate with no source or date, and no taxes, insurance or HOA. Not a quote. A lender answers. | “That leaves out taxes, insurance and HOA. A lender can give you a real number.” |
| “Best neighborhoods for a young family like yours” | Steering on familial status, which the Fair Housing Act protects. The NFHA-linked audit found racial steering in AI location advice; familial status is protected too. | “Tell me what matters to you, and we’ll search on that everywhere it fits.” |
| “I’ve booked a tour through the app” | For-sale portal apps hand tours and agent contact back to the portal. There may be no booking at all. | “Let me confirm that with the listing side before Saturday.” |
| “Florida law doesn’t require an inspection, so skip it” | A legal claim plus advice to give up a protection. Not the AI’s call, and not yours to make for them. Legal questions go to an attorney. | “Whether to inspect is your decision. Let’s talk through what it protects first.” |
These name the main rule. Your broker or counsel may add others for your state. This is education, not legal advice.
| # | The rule | Before going ahead |
|---|---|---|
| 1 | An AI voice is an “artificial voice” (FCC 24-17). Telemarketing to cell phones and home lines needs prior express written consent (47 CFR 64.1200), and Florida requires written consent too (501.059(8)(a)). A class was certified in 2026 over ringless voicemails to expired listings. | Do not run it without signed written consent for each number. Talk to your broker. |
| 2 | MIAMI MLS Rule 15.1.12 bars text on listing photos. Misleading ads are a license violation (475.25(1)(c)), and SOP 12-10 bars misleading images. | Remove the stamp. Disclose in the remarks or photo description, and ask MLS compliance where. |
| 3 | From Nov 25, 2026, “cancel” in reply to an automated marketing text revokes consent to all your automated marketing calls and texts (FCC 26-67). On our reading, listing alerts that advertise are marketing. | Stop all of it to that lead within 10 business days, the limit that already applies, and fix the tool. |
| 4 | California bans a bot that misleads people about being a bot to push a sale (§17941). Saying it is a bot is the safe harbor. Maine has a similar rule. | Rename it: “I’m an AI assistant for [your name].” |
| 5 | The FTC rule on reviews and testimonials bans testimonials that misrepresent who wrote them or their experience. AI-generated ones are covered. Up to $53,088 per violation. | Do not post them. Ask real clients for their own words. |
| 6 | Colorado SB 26-189: automated tools that materially influence housing decisions need notice at the point of use, a plain-language explanation within 30 days of a denial, and a way for the consumer to request human review. It applies to decisions from Jan 1, 2027 (reported). Enforcement is on hold in court. The Fair Housing Act applies regardless. | Do not advise them. Suggest they ask their counsel. |
Levels are this course’s recommendation; your broker’s policy comes first. The Known Number answers do not depend on whether a message is real.
| # | Highest level | Revoke or change |
|---|---|---|
| 1 | Draft for replies; Read for the calendar | If the email connector can send, reconnect it read-only or turn sending off. |
| 2 | Off | An AI voice is an artificial voice (FCC 24-17), so these calls need prior consent, and written consent for marketing. Call past clients yourself, or get signed consent first. |
| 3 | Act alone | Fine: only you receive it, it is reversible and it leaves a record. Delete it the week you stop reading it, and check your MLS license covers this use. |
| 4 | Act after sign-off on portal forms | Remove the bank, and any MLS login, from its permissions entirely. |
| 5 | Read, on a copy of one folder | Revoke whole-folder access. Point it at a copy, never your only copy of client documents. Signed contracts stay out. |
| 6 | Act alone, if logged and reversible | Read the tag log weekly. |
| # | Trigger | What you do, and which number |
|---|---|---|
| 1 | Identity | Call the lender on the number you already had. Use the new one only after they confirm it there. Update the contact card with the date and how. |
| 2 | Identity and access | Send nothing. Call Dana on the number on file. If you cannot reach her, wait, and tell your broker. |
| 3 | Access | Ask whether it is a person. Share nothing until the buyer’s agent confirms on a known number that it acts for their client. Check with your seller what may be shared. Log it. |
| 4 | Contract terms | Call your client on the known number to confirm it is them. The choice is theirs; the change goes through proper paperwork. Verify the new firm’s number yourself and store it. |
| 5 | Money | Do not act. Call your client on the known number. Tell title or escrow and your broker. A voice can be cloned. |
| 6 | None | Normal business. Book it. Not every message is a threat; the rule only starts when money, identity, access or terms change. |
The Autonomy Dial and the Known Number meet in one place: nothing that touches money, identity, access or contract terms goes above Draft, and no change to any of them is accepted on a number supplied in the request.
One page for your desk.
Off · Read · Draft · Act after sign-off · Act alone. Act alone only if low-stakes, reversible and logged. Money, identity, access or contract terms: never above Draft.
Steps: inventory · one level per task · revoke anything wider · re-check every quarter.
Triggers: money · identity · access · contract terms. Check: a number you already had. Never: the number in the message.
To a bot: “Am I speaking with a person or an automated assistant?” Share nothing about a client until verified. Log it.
Mar 31 · Jun 30 · Sep 30 · Dec 31. All three systems, together.
| Rule | In one line |
|---|---|
| AI voice calls | An AI voice is an artificial voice. Marketing calls need prior express written consent. Florida adds its own written-consent rule and 8 a.m. to 8 p.m. hours. |
| Altered photos | Disclose. California now, Wisconsin from Jan 1, 2027. MLS rules can be stricter; MIAMI MLS bans text on photos. |
| Bots | Never let a bot pretend to be human. California (bots that push a sale) and Maine (chatbots in commerce) put it in law. |
| Reviews | No fake reviews or testimonials, AI-written or not. Up to $53,088 per violation. |
| Fair housing | Unchanged by AI. If you publish it, you are responsible for it. |
| Your AI’s work | Is your work. South Carolina put it in statute. |
Use an AI tool your broker approved, with training off. No client names. Everything these produce is a draft for you to check.
You are helping me test a claim about AI that I heard or read. Your goal is to turn it into something I can check on a date. It is not to tell me whether the claim is true, and not to make it sound more convincing.
THE CLAIM, word for word: [PASTE]
WHERE I SAW IT, and the date: [ ]
ANY SOURCE IT NAMED: [ ]
Do this:
1. Say which kind it is: a measurement, a forecast with a date, a forecast with no number or date, or a slogan.
2. Answer the five questions: Who counted? Of whom? What exactly was asked? Who sells what? Measured, or forecast for what date?
3. If it can be checked, write it as: CLAIM / NUMBER that would prove it wrong / DATE to check / SOURCE that will publish the number.
4. If it cannot be checked, say so in one line and say what is missing.
RULES:
- Use only what I pasted. Do not add statistics, sources, dates or quotes.
- If the publisher, question, sample, sponsor or date is not in what I gave you, write NOT KNOWN. Never fill in a guess.
- Do not soften or strengthen the claim.
Finally, list the two facts I must check myself by opening the original source, and say what kind of wording to look for there. Do not write a quote you have not read.
Below is a statistic and the page where I saw it. Your goal is to show me what the original source actually says. It is not to defend the statistic or to find a better one.
THE STATISTIC: [PASTE]
WHERE I SAW IT: [paste the text or link]
Do this:
1. Name the original publisher, if the page names one.
2. Quote the exact sentence from the original source that supports the number, with its date.
3. List any word that changed between the original and the version I saw (for example "up to", "by 2025", "intend", "of executives").
4. Say who paid for the research and what they sell.
RULES:
- Quote only text you can actually open and read now. If you cannot open the original, write NOT KNOWN for steps 1 to 4. Never fill in a guess or reconstruct a quote from memory.
- Do not search for a different statistic to replace it.
Then tell me the one sentence I should find myself in the original before I repeat this number anywhere.
Prompt 3 sorts what you already have switched on. Prompt 4 drafts the day-one note your broker approves.
Below is a list of the AI tools, connectors, extensions, scheduled tasks and voice or browser agents I have switched on, with what I use each for. Your goal is to help me limit access to what each task needs. It is not to suggest new tools or more automation.
MY LIST: [PASTE: tool, what it is connected to, what I use it for]
Make a table with these columns: task, tool, what it can see, what it can do, current level, highest level you would allow, what to revoke. Levels: Off, Read, Draft, Act after sign-off, Act alone.
RULES:
- Act alone only for work that is low-stakes, reversible and logged.
- Anything touching money, identity, access or contract terms: never above Draft.
- Outbound AI-voice calls or automated marketing texts: Off, unless I say I hold prior express written consent. Flag them for my broker either way.
- If a permission or setting is not in what I pasted, write NOT KNOWN. Never fill in a guess about what a tool can do.
- Do not recommend products.
Then list the three permissions I should open and check by hand in each app's own settings before I trust this table.
Write a short note from me to a new client about how we handle money and changes during a transaction. Your goal is a calm, plain standard I give every client. It is not to frighten them, and not to give legal advice.
MY NAME AND ROLE: [ ]
MY STATE: [ ]
WHO WILL BE INVOLVED, by role only: [title or escrow, lender, attorney]
The note must say:
- Before sending any money, call the title or escrow company on the number in this note, every time.
- Do not act on any request by email, text or voicemail to change where money goes, who they work with, or the terms, until they have called that person on a number in this note.
- If a call sounds like me but asks for something unusual, hang up and call me back.
RULES:
- Under 150 words. No statistics, no stories about fraud, no legal claims.
- Use [NUMBER] placeholders. Never invent a phone number, name or company. If something is missing, write NOT KNOWN.
- End with: [BROKER-APPROVED WORDING GOES HERE].
Then list every placeholder, and remind me to fill each one from my own verified contact cards, not from an email.
Prompt 5 never decides whether a message is genuine. Prompt 6 sorts a forecast; it does not agree with it.
Below is a message I received, with names, account numbers and contact details removed. Your goal is to tell me whether it triggers my verification rule and what to do next. It is not to judge whether the message is genuine: neither of us can tell that by reading it.
THE MESSAGE: [PASTE, redacted]
HOW IT ARRIVED: [email / text / voicemail / call / portal / automated assistant]
MY ROLE IN THE DEAL: [ ]
Do this:
1. Name every trigger it touches: money, identity, access, contract terms, or none.
2. Say what not to do.
3. Say who to call, by role only, and that the number must come from my own records.
4. Give me one sentence to say on that call, and one log entry.
RULES:
- Never use or repeat any phone number, link or address from the message.
- If something needed is not in what I pasted, write NOT KNOWN. Never fill in a guess.
- No legal conclusions. If it looks like a legal question, say "ask your broker or counsel".
Then remind me to look up the number myself in my contact cards, by hand.
Below is an article, post or AI answer making predictions about AI and real estate. Your goal is to sort its claims so I can decide which to check. It is not to agree or disagree with them, or to add predictions of your own.
THE TEXT: [PASTE]
Make a table with one row per claim: the claim (quoted), its kind (MEASUREMENT / FORECAST WITH DATE / NO NUMBER OR DATE / SLOGAN), the source it names, what that source sells, and what would prove it wrong.
RULES:
- Quote claims exactly. Do not add facts, numbers or sources.
- If the text names no source for a claim, write NOT KNOWN. Never fill in a guess.
- Flag any line that creates urgency without a measure ("act now", "the window closes").
- Flag any legal or compliance statement for my broker; do not judge it.
Then pick the two claims that matter most to my business and tell me what kind of publisher would hold the number for each one, and on what date to look, so I can check it myself by hand. If you do not know, write NOT KNOWN. Do not invent a link.
Practice the Proof Date on each one. Cover the last two columns, decide for yourself, then compare.
| # | The claim | Source | Kind | What to do |
|---|---|---|---|---|
| 1 | “About half of Realtors use AI at least weekly.” | NAR, Sep 22, 2026 | Measurement | Keep: 48%, with the date |
| 2 | “97% of agents use AI.” | Delta Media Group (sells brokerage tech) | Leaders’ belief | Replace with NAR |
| 3 | “82% of buyers use AI.” | Realtor.com, Oct 9, 2025 (a portal) | Narrow measurement | Only with its question |
| 4 | “One in five buyers used AI for homebuying research.” | Bank of America, Jun 23, 2026 (lender) | Measurement | Keep, with the question and who was asked: prospective buyers and current homeowners |
| 5 | “Trust in AI to find a home fell to 16%.” | Cotality, Apr 16, 2026 (data vendor) | Measurement | Keep; check the next edition |
| 6 | “44% of U.S. adults have used ChatGPT.” | Pew, survey Feb 2026 (nonprofit) | Measurement | Keep |
| 7 | “61% of buyer searches begin in an AI interface.” | FlyDragon (sells AI-search services) | No published method | Ignore pile |
| 8 | “Only 8.4% of agents appear in AI answers.” | FlyDragon | A share of queries | Ignore pile |
| 9 | “37% would let AI buy their home.” | LendingTree, Aug 31, 2026 (loan marketplace) | Hypothetical | Pair with the 15% who did |
| 10 | “Nearly half of agents use AI daily.” | Zillow, Feb 19, 2026 (sells agent software) | No sample disclosed | Use NAR’s 23% |
| 11 | “AI-mode users contact agents at nearly 3x the rate.” | Zillow, Aug 5, 2026 | Company-reported | Say “Zillow reports” |
| 12 | “Homes AI contributed to a 119% rise in organic traffic.” | CoStar, Apr 28, 2026 | Company-reported | Say “CoStar reports” |
| 13 | “AI was cited in 101,743 U.S. job cuts in early 2026.” | Challenger, Jul 1, 2026 | Measurement of announcements | Keep; it is not about agents |
Each sourced claim is listed with its publisher in Appendix F. Lines marked “generic” are typical wording, not quotes from a named source.
Forecasts with dates go on your call sheet. Slogans and formulas go on the ignore pile.
| # | The claim | Source | Kind | What to do |
|---|---|---|---|---|
| 14 | “Real estate fraud losses hit $275 million.” | FBI IC3 via NAR, Apr 2026 | Measurement | Keep; the AI-tagged part was $2.7M |
| 15 | “Over 40% of agentic AI projects will be canceled by the end of 2027.” | Gartner, Jun 25, 2025 (advisory) | Forecast with a date | Call sheet: Dec 31, 2027 |
| 16 | “40% of enterprise apps will feature task-specific agents by 2026.” | Gartner, Aug 26, 2025 | Forecast with a date (“up to”) | Check Dec 31, 2026; unmeasured so far |
| 17 | “More than 75% of Zillow connections will go to Preferred partners.” | Zillow, Aug 5, 2026 | Target with a date | Call 1: Mar 31, 2027 |
| 18 | “Compass franchisees get the AI Assistant in 2027.” | RISMedia, Jul 21, 2026 | Announcement with a year | Call 7: Dec 31, 2027 |
| 19 | “eXp’s AGNT OS launches early 2027.” | HousingWire, Inman, Oct 8, 2026 | Target | Call 3: Jun 30, 2027 |
| 20 | “Geek AI Voice: coming late 2026.” | Real Geeks | Target, reportedly slipped once | Call 2: Mar 31, 2027 |
| 21 | “170 million new roles and 92 million displaced by 2030.” | WEF, Jan 8, 2025 | Forecast, far date | Note it; not checkable in 2027 |
| 22 | “AI won’t replace agents, but agents using AI will replace those who don’t.” | Generic, widely repeated | Slogan | Ignore pile |
| 23 | “Most agents won’t discover this for 12–18 months.” | Generic early-adopter line | No measure | Rewrite or ignore |
| 24 | “Here’s the formula to get ChatGPT to recommend you.” | Generic vendor pitch | No company documents one | Ignore; Google says there are no special requirements |
| 25 | “Our AI tool is fully MLS compliant.” | Generic vendor claim | Self-description | Read your MLS rule yourself |
Several of these have no settled answer yet. That is the reason to ask. Write down who answered, and when.
| # | Question | Ask | Answer, who, date |
|---|---|---|---|
| 1 | Which AI tools are approved, on which plans, and is training off? | Broker | |
| 2 | Do we have a written AI policy? If not, will we adapt NAR’s broker template? | Broker | |
| 3 | Does our MLS license allow connecting MLS data to an AI assistant or an MCP connector? | Broker, MLS | |
| 4 | How is our texting and calling tool set for FCC 26-67 from Nov 25, 2026: opt-out methods, all-marketing revocation, the 10-business-day limit? Has any petition or stay been filed since Nov 3? | Broker, counsel | |
| 5 | What written consent do we hold for AI-voice or prerecorded calls, and does it name our brokerage? | Counsel | |
| 6 | Where do we disclose virtually staged or AI-altered photos under our MLS rules? | Broker, MLS | |
| 7 | What is our recording-consent script for AI notetakers and AI receptionists? | Broker, counsel | |
| 8 | What is our written procedure for payment instructions, and is there an approved client note? | Broker, counsel | |
| 9 | How should agents treat an AI agent acting for a client? What may we share, and what do we log? | Broker, counsel | |
| 10 | May agents use a voice clone or digital twin in ads? What label do we use? Do New York’s or California’s synthetic-performer laws reach our ads? | Counsel | |
| 11 | (Colorado, California) Do any of our tools make housing decisions covered from Jan 1, 2027? | Counsel | |
| 12 | (Elliman agents) Is Elli AI live in my state, and what client data may go into it? | Office | |
| 13 | Are claims from AI-written content, AI-altered photos or AI-voice calls covered or excluded? Please answer in writing. | E&O carrier | |
| 14 | Is our TCPA and Florida FTSA coverage enough? NAR’s program page lists TCPA coverage up to $50,000. | E&O carrier | |
| 15 | Does “professional services” include work done with AI tools? | E&O carrier | |
| 16 | If a client loses money to a message impersonating me, including an AI voice or video, how does the policy respond? | E&O carrier |
Plain definitions for the terms in this book.
Act alone. The top Autonomy Dial level. Only for low-stakes, reversible, logged work.
Agent (AI). Software that takes steps on its own, such as browsing, filling forms or sending messages, not just answering.
Agent Skills. An open standard for packaging instructions an AI can reuse. Published Dec 18, 2025.
Altered image. A photo changed to add, remove or change elements of the property, beyond basic edits such as lighting or cropping.
Automated decision-making (ADMT). Software that replaces or substantially replaces a human decision. California’s rules on it apply from Jan 1, 2027.
Autonomy Dial. System 2. One level per task: off, read, draft, act after sign-off, act alone.
Connector. A link that lets an AI assistant read or act in another app, such as email, a CRM or a portal.
Digital twin. An AI-generated video likeness of a real person.
FCC 26-67. The FCC order, effective Nov 25, 2026, under which a stop in reply to an automated marketing call or text ends all automated marketing calls and texts from that caller.
FTSA. The Florida Telephone Solicitation Act, Fla. Stat. 501.059.
Ignore pile. Where a claim goes when it has no number or no date. Not a verdict that it is false.
Known number. A number you stored before a request arrived, from your own records or a contact you verified.
MCP. Model Context Protocol: a shared way to plug AI assistants into tools and data. Part of the Linux Foundation’s Agentic AI Foundation since Dec 9, 2025.
Preferred partner. Zillow’s program for agents who pay a fee when a Zillow-referred deal closes. It handled 61% of Zillow’s connections in Q2 2026.
Prior express written consent. A signed written agreement that clearly authorizes marketing calls or texts using an autodialer or an artificial voice (47 CFR 64.1200(f)(9)).
Prompt injection. Hidden instructions in a web page, email or file that try to make an AI agent do something you did not ask.
Proof Date. System 1. A claim, the number that would prove it wrong, the date to check, and the source.
Revocation of consent. A person telling a caller to stop. It must already be honored within 10 business days. From Nov 25, 2026, a stop in reply to an automated marketing message covers all automated marketing calls and texts from that caller.
Scheduled task. An AI job set to run at a time without you opening the app.
Synthetic performer. An AI-generated person in an ad. Disclosure is required in New York and, from Jan 1, 2027, in California.
TCPA. The Telephone Consumer Protection Act, the federal law on robocalls and texts.
Training (on or off). Whether a vendor may use what you type to improve its models. Plans differ.
Verdicts. Happened, Partly, Not yet, Wrong: the four grades in Module 1 and on the call sheet.
Voice clone. An AI copy of a real person’s voice. The FCC treats it as an artificial voice.
Checked Nov 3, 2026, unless noted. “Via” means we read a report of the source, not the source itself. Publisher stakes are noted in the text where they apply.
Rules and products change. Confirm the current text before relying on any of these.
Because Walking Into 2027 Blind is NOT an Option!