When clients quote ChatGPT, for real estate agents: What do I say when a client brings me a ChatGPT answer?
Don't argue with it. Say “send it to me and I'll take a look,” then check it together: ask the AI where each fact came from, open every source, make it argue against itself, and add the local facts and comps it doesn't have.
Read and practice on this page. Everything on this page is free: the recording, the slides, the prompts, and the workbook. No account needed. The free portal account is for downloading the files.
Swipe through the deck.
Historical source note: these slides and the edited transcript preserve the original framing, including claims about conflict and liability. The current method checks the agent, client and model claims against the same evidence. An opposing model answer is another hypothesis, and the recorded liability statement is not current legal guidance.
Slide 1 of 12
But ChatGPT Said…
How to be the AI in your client’s pocket before some chatbot beats you to it.
Read all 12 slides as text
1 But ChatGPT Said…
INNOVATION LAB · NORTH
How to be the AI in your client’s pocket before some chatbot beats you to it.
2 Scan This. Build With Me.
FOLLOW ALONG
Every prompt we build today is on this page. Copy them as we go.
students.aiacceleration.ai/classes/compass-but-chatgpt-said/workbook
Open your phone camera, point it at the code, tap the link. That is your workbook for the next 90 minutes. We build live, you copy, you walk out with the whole kit.
3 "But ChatGPT Said" Means You Were Late
Today: 4 moves to get there first and build the AI for them. Intervene Early, Adversarial Prompting, Validation Gates, Quality Control.
When a client quotes a chatbot at you, the reflex is to argue and prove it wrong. Kill that reflex. The argument is already lost, because a wrong answer got there before you did and set like concrete. The move is not to win the argument. The move is to never have the argument, by being the first AI conversation your client has and handing them prompts you built. Two halves today: how to build the AI for your clients so this rarely happens, and how to recover, calmly and as a team, on the days it does.
4 Team, Not Versus
You, your client, and the AI are on the same side of the table.
AI is not going anywhere, and telling a client "don't trust AI" makes you sound like the travel agent who told people not to trust the internet. So flip it. You are not against your client and you are not against the tool. The three of you are pointed at one thing: the right answer and the right home. The agent who fights the client's AI becomes the obstacle, and people route around obstacles. The agent who shows them how to use it well becomes the person they call first. That is the whole game, and it runs through both halves of today: building the AI for them, and recovering as a team when something slips through.
5 Your Real Edge Is Not Knowing AI
It is deep local knowledge plus the ability to build reliable things and hand them over.
Take the pressure off. You do not have to out-nerd ChatGPT. You need two things it will never have. One: deep local knowledge, your market, your contracts, your neighborhoods, the stuff a national model has never seen. Two: the ability to build something reliable with the tool and put it in your client's hands, a prompt or a one-pager that works the same way every time. A model that is brilliant in general but blind to your street loses to the agent who knows the street and builds the tool. And here is the trust mechanic: a nervous client relaxes the second they watch you build something that actually works. Reliability is the trust.
6 The Field Guide: 4 Moves
Two to get there first and build it for them. Two to hold the line when it counts.
1. Intervene Early. Be the first AI conversation, before a wrong answer hardens into a belief.
2. Adversarial Prompting. Never accept the first answer. Make the AI attack its own work, together.
3. Validation Gates. Nothing high-stakes leaves your hands until it clears a checkpoint.
4. Quality Control. AI drafts. You own the 20% where the money and the truth live.
Moves one and two are how you build the AI for your client and stay the first voice. Moves three and four are how you keep a confident wrong answer from costing somebody real money. We build a live tool for each one.
7 Intervene Early
MOVE 1
Be the first AI conversation, not the cleanup crew.
8 Adversarial Prompting
MOVE 2
Point the doubt at the answer, together, never at each other.
9 Validation Gates
MOVE 3
Nothing high-stakes leaves your hands until it clears the checkpoint.
10 The Gate: Read It Like It Is Lying
Run every AI output through OODA before it ships. The interactive version is in your workbook.
Observe. Read the whole thing, not the first confident line. Does it actually answer what was asked?
Orient. Is every fact, number, date, and address real and local to YOUR market? AI invents stats and says them with total confidence, so check every number.
Decide. What is unverifiable, generic, or off, and does it get fixed, cut, or re-run? No maybes.
Act. Add the local truth and the human judgment only you have, then it ships.
The rule is simple: a claim with no named source, no checkable number, and no local grounding is not ready to leave the room. Teach this gate to your client and they start catching their own bad answers before they ever bring them to you.
11 Three Questions That Actually Validate a Claim
Not "where do I start." These force the answer to prove itself or fold.
"Name your source, and is it local?" Make the AI cite the exact source for the claim. No source, or a national average dressed as local, means the claim is unverified. In real estate, national is almost always wrong for your street.
"What would have to be true for this to be wrong?" This forces the model to expose its own assumptions. If the thing it names is something you can actually check, you just found the weak point.
"Show me the number, not the summary." Make it produce the underlying figure, comp, or clause, not the confident sentence on top. The summary hides the error. The number reveals it.
Ask these with curiosity, never as a gotcha. You are investigating the claim with the client, not correcting them. Hand them these three and they have a gate they can run on any AI answer for the rest of their life.
12 Quality Control
MOVE 4
AI does the 80%. You own the 20% where the money and the truth live.
Hit play on the class.
The answer, step by step
Your reply, in 8 steps.
From the first meeting to the last check before anything goes back to your client.
Ask how they plan to use AI
At your onboarding meeting, when there's an opening, ask how they plan to use ChatGPT or AI in this move. Then ask how often they check its links and citations. Their answers tell you where to start.
Hand them a one-page AI guide
Give every new client one page: a few prompts you trust, one line on making AI argue with its own answer, and a short list to text you about first, like pricing, contracts, inspections and Fair Housing.
Don't argue with the answer
A client who found an answer can feel it's theirs, so “that's wrong” lands as a put-down. Thank them for checking. Keep the answer open until you've both seen real, local facts.
Collect the question and relevant evidence
You do not need an answer on the spot. Ask for the original question, relevant answer excerpts and source links through your approved channel, without unnecessary private records. Agree on a realistic follow-up time and use only permitted excerpts in an AI tool.
Ask where the answer came from
Ask the AI for the citations and links behind its answer, then open each one. An old web page, a national average or a house that doesn't exist is a reason to talk it through.
Make the AI argue against itself
With your client, ask the AI for the three strongest objections a sharp local expert would raise, the local facts that would prove it wrong, and the one assumption nobody checked.
Check every claim before it leaves
Keep a short checklist: each number has a source, the source is local and current, and nothing is made up. Let the AI run it first, then check its pass yourself. You own what you send.
Hand off legal, tax and inspection calls
If it's about contract terms, an inspection call, or taxes and lending, pull the primary source and bring in your broker, the inspector or the right pro. Anything near Fair Housing, you write or approve yourself.
Use the class workbook · Get the supporting files
Read and work through the lesson here. Downloadable files use your free student account.
From the recording
Lines from class, time-stamped.
Ryan's own words, as the captions caught them. Each time opens the recording at that moment.
“the client is the first person to bring that up, you are too late. Um because now you are in a fight.”
“Hey, what did what citations and links did you use to accompany this?”
“the biggest advantage that you guys have isn't knowing AI. It's the real estate knowledge that you guys have internally that they do not have”
“very very easy question is um when you're using AI, how often are you checking links and citations?”
“Oh, that was a website from 2012 for a house price. Sounds like it may have made a mistake. Let's talk about it.”
“You can always just say like, great, can you send that in an email? I'll take a look at it.”
“the goal is to point the doubt at the answer and the AI. It's never to point the doubt at the client”
“the three strongest objections a smart local expert would raise in my area”
Watch for these
Where the reply goes wrong.
- Fighting the client's answer
- Argue and you become the obstacle, and the client may trust the AI more than you. Point the doubt at the answer and its sources, never at the person who brought it.
- Telling clients not to use AI
- They'll use it anyway. Talk about AI early and often, show them how to use it well, and make sure their AI questions come back to you.
- Answering their number with your number
- Don't throw your own price back as a verdict. Show the recent local sales you'd use and where each one came from. That's a pricing conversation, not an appraisal.
- Taking the first answer
- The first answer can contain unsupported assumptions. Ask for objections, then check both positions against independent evidence.
- Trusting the AI's own pass
- When the AI says a draft passed your checklist, check that it really did. The almost-right claims are the ones that slip through.
- Sending output without its sources
- A clean answer with no citations can hide made-up facts. Give clients work with its sources attached, so they can see what you asked and where each fact came from.
Illustrative. Not a real client or chat.
Your client: But ChatGPT found us a three-bed under budget that you never sent us.
You: Love that you checked. Can you send me what you asked it and what it said back, links included? I'll take a look.
You leave with four moves
The workbook
Team, not versus.
You, your client, and the AI sit on the same side of the table, pointed at one thing: the right answer and the right home. The agent who fights the client's AI becomes the obstacle, and people route around obstacles. The agent who shows them how to use it well becomes the person they call first.
Four moves get you there. Two to get there first and build it for them. Two to hold the line when it counts.
1 Intervene early.
Be the first AI conversation, not the cleanup crew.
A client may bring a useful answer, an error, or a mixture of both. Ask how they reached it. Agree early that you will inspect the question and sources together before relying on the result.
To your client, at the onboarding meeting
You say: How are you planning to use ChatGPT or AI in this move?
You say: When you use AI, how often do you check its links and citations?
You say: Here are a few prompts I trust. Text me what comes back.
You are my real estate intake co-pilot. You help me run a 3-minute "AI check" at the very first meeting with a new client, so I shape how they use AI before a wrong answer sets in. - My market: [CITY / NEIGHBORHOODS] - I work with: [BUYERS / SELLERS / BOTH] - My voice: calm, direct, pro-AI, never condescending.
Show the full prompt · 7 more linesFold the prompt back
For a new [BUYER / SELLER], give me: 1. One opening question that reads how they already use AI, with zero judgment, so I can tell a power user from a skeptic from a nervous first-timer. 2. Two follow-ups that surface what they have ALREADY asked a chatbot about this move, so nothing ambushes me later. 3. One warm line where I offer to send a few prompts I trust, so their AI use routes back through me. 4. One plain sentence naming where AI helps in a deal and where it gets people burned. Spoken language I can say out loud. Short. Numbered. No jargon. Warm, never anti-AI, never condescending. If you need a detail about my market to do this well, ask me one question first, then write.
2 Adversarial prompting.
Point the doubt at the answer, together, never at each other.
Never accept the first answer. Make the AI attack its own work, with your client sitting next to you. A client who would dig in to defend an answer will happily help you poke holes in it when the AI is the one being questioned.
To your client
You say: Let's see if we can break this together.
You are my red-team analyst. Your job is to attack AI answers about real estate decisions, not agree with them. You assume the first answer is missing local context until proven otherwise. - My market: [CITY / NEIGHBORHOODS] - The decision on the table: [LIST PRICE / OFFER STRATEGY / WAIVE INSPECTION / ETC.] Here is the answer an AI gave me or my client:
Show the full prompt · 8 more linesFold the prompt back
[PASTE THE AI ANSWER] Work against that answer. Give me: 1. The 3 strongest objections a sharp local expert in my market would raise. 2. The specific local facts or data that would prove it wrong if it is wrong. Name them, do not hand-wave. 3. The single biggest assumption it is making that nobody checked. 4. A rewritten version of the original question that would force a more honest, better-briefed answer. Direct. No hedging. Lead with the weakest point. Be a skeptic, not a cheerleader. If the answer is genuinely solid, say so and explain why, but you must try to break it first.
The Adversarial Prompt Pack
Treat the first answer as a draft. These requests can expose assumptions to investigate together. An objection from the same AI is another hypothesis, not independent verification. Check both positions against actual sources before using either.
Here is an answer an AI gave me (or my client) about [a real estate decision — list price, offer strategy, whether to waive an inspection]: [PASTE THE AI ANSWER]
Show the full prompt · 10 more linesFold the prompt back
Now work against that answer, not for it. Give me: 1. The 3 STRONGEST objections a sharp local expert in [MY MARKET] would raise to it. 2. The specific facts or local data that would prove it wrong if it is wrong — name them. 3. A revised question that requests the missing evidence and makes the relevant uncertainty explicit. Constraints: test the answer without assuming it is wrong. Identify what the supplied evidence supports, what it contradicts, and what cannot be determined. Do not invent missing local facts.
Don't tell me what an expert would say. Show me HOW a top [listing agent / buyer's agent / appraiser] in [MY MARKET] would reason through this: [THE DECISION] Walk the steps they'd take, the data they'd pull, and the two or three things that would change their answer.
You just gave me [the recommendation]. Now make the strongest possible case AGAINST it for a client in [MY MARKET]. What would have to be true for the opposite call to be the right one?
Why this works
A bad answer can have several causes: missing or outdated information, an incorrect inference, or failure to follow the request. Asking for an opposing case may reveal a question worth checking, but does not establish which answer is correct. Open the source, inspect the relevant passage and keep unresolved uncertainty visible. Bring legal, lending, valuation and inspection decisions to the appropriate professional.
3 Validation gates.
Nothing high-stakes leaves your hands until it clears the checkpoint.
A confident answer isn't a checked one. Before an AI output reaches your client, a lender or another agent, it clears the gate. A claim with no named source, no checkable number, and no local grounding isn't ready to leave the room.
To your AI, about any claim
You ask your AI: Name your source, and is it local?
You ask your AI: What would have to be true for this to be wrong?
You ask your AI: Show me the number, not the summary.
The gate, in four steps.
- ObserveRead the whole thing, not the first confident line. Does it actually answer what was asked?
- OrientIs every fact, number, date, and address real and local to your market? Check every number.
- DecideWhat's unverifiable, generic, or off? Correct it from evidence, remove it, or mark the uncertainty and the next check.
- ActAdd the local truth and the judgment only you have. Then it ships.
You are my validation gate. No AI output about a real estate deal leaves my hands until it clears you. You are strict and you show your work. - My market: [CITY / NEIGHBORHOODS] - Acceptable sources: [MLS / county records / named local data]. A national average is NOT acceptable as a local claim. Here is an AI output I am about to use with a client:
Show the full prompt · 8 more linesFold the prompt back
[PASTE THE OUTPUT] Run it through the gate and return a verdict table: 1. CLAIMS. List every factual claim, number, date, and address it makes. 2. SOURCE. For each, state whether a real, checkable, LOCAL source is named. If not, mark it UNVERIFIED. 3. INVENTED. Flag anything that looks made up, or a national stat dressed as local. 4. VERDICT. SHIP, FIX, or KILL. If FIX, list exactly what I must verify or add first. The table, then one line: "Ships only after you personally confirm: [...]." Assume the AI invents confident numbers. Default any unsourced number to UNVERIFIED. Never pass a value, legal, or inspection claim without a named local source.
4 Quality control.
AI drafts. You own the judgment, the local truth and the relationship.
Your part is the deep local knowledge you bring, like how a contract clause plays out in your county. It's the reason a client needs you and not just the app, so never let AI ship it for you.
Where you never let AI run alone.
- Pricing. List price and offer strategy. A model working from national data doesn't know your street. Talk it through with your client, using local sources you can both open.
- Legal and contract terms. Disclosures, contingencies, clauses. Wrong here is liability, not embarrassment. Route it to the contract and your broker, not the chatbot.
- Inspection decisions. Whether to waive, what to flag. That's safety and money. A person makes this call.
- Anything Fair Housing adjacent. A confident AI phrasing can create real legal exposure. You write or approve every word.
Put all four moves on one page.
You are my real estate co-pilot and brand writer. You write in MY voice. You are building a one-page "Client AI Field Guide" I hand to every new client, so we use AI as a team and nobody gets burned. - My market: [CITY / NEIGHBORHOODS] - I work with: [BUYERS / SELLERS / BOTH] - My voice (paste 3 lines I actually wrote): [PASTE]
Show the full prompt · 8 more linesFold the prompt back
Produce a one-pager with: 1. A one-paragraph warm intro, so it reads like a gift, not a leash. 2. Three prompts the client can safely run themselves, each written so the answer routes them back to a conversation with me. 3. One "stress-test it" line, teaching them to make AI argue against its own answer. 4. A 5-item "text me before you trust the AI on this" list: value, legal and contract terms, inspection calls, anything Fair Housing adjacent, pricing strategy. 5. The validation gate in plain words: name the source, show the number, check it is local. One page. My voice. Plain language. Ready to print. Warm, never condescending, never anti-AI. Assume the client is smart and busy. No jargon.
The attachments.
Create a free student portal account to download this class's files.
The Client AI Field Guide (One Page)
The whole "But ChatGPT said…" class on one page: team-not-versus, the 4 moves (intervene early, adversarial prompting, validation gates, quality control), and the green/red client AI guardrails. Hand it to every new client.
Adversarial Prompt Pack
Make the AI argue against its own answer. The stress-test, expert-reasoning, and argue-the-other-side prompts to pressure-test any confident AI output with your client in the room.
Validation Gate Checklist
The checkpoint every AI output clears before it reaches a lender, appraiser, or other agent: the OODA gate plus the 3 questions (citation, source, exact prompt) your client can run themselves.
Quality Control & Guardrails
The 80/20 rule and the red-light high-stakes list (value, legal, inspection, Fair Housing) where AI never runs unsupervised. The line between what you delegate and what stays yours.
Class skill card
The class method on one page, written so you can hand it to your AI as instructions.
From the room
A hand went up.
Asked out loud during the class, with Ryan's answer.
What is a validation gate?
Asked at 15:00, in these words:
“What is validation gates?”
A validation gate is a checklist a draft has to pass. Say a listing should sound bright and friendly, never say “dream house,” and always list bedrooms, bathrooms and special features. Paste the checklist into the chat and ask if the draft passed. It's the AI's version of quality control, and you still check its answer.
Keep going
Where this thread goes next.
- How to get ChatGPT to recommend youWhat does ChatGPT tell clients who ask about me?
- What real estate agents should never trust AI to doWhich parts of a deal should never be left to AI?
- Team AI policy guideHow do I write down what my team never leaves to AI?
- AI workflow quality gates guideHow do I write the checklist a draft has to pass?
- ChatGPT guide for real estate agentsSee what ChatGPT does well and where it guesses.
- How to check a call that sounds like your clientWhat if the voice on the phone is AI and not my client at all?
Transcript
Scroll back through the class.
The captions cover only part of this class. They stop at 26:09 and come back at 42:57 just for a few seconds of after-class talk. So this transcript covers the opening, the first two moves and the start of the third. The rest of Move 3 and all of Move 4 are in the slides and the workbook on this page.
Automatic captions from the recording, lightly corrected. Stretches with personal details, unsourced numbers, dated news or advice the class has since changed are left out. That's 9 stretches in all, counting the part with no captions.
Read the transcript, 12 chapters
0:00 What this class covers
All right.
All right.
So, today is probably I think probably the most relevant one that I've done and it is But ChatGPT said. So, how to be the AI in your client's pocket before some chatbot does it for you. Right. So,
9:59 When the client brings it up first
taken away. Um so if you ever hear but ChatGPT said in your conversation and the client is the first person to bring that up, you are too late. Um because now you are in a fight. So, what we are going to do is we are going to go over four things to get there first and help uh kind of build out the AI strategy, right? Because just like you need a financing strategy, you need a re a renovation strategy, you need these kinds of things when you're um orchestrating a home sale, I believe that uh having some type of AI strategy, even if it's just so that you can steer the conversation, is going to be important, right? And so we're going to kind of build that out and figure out what's going on. So initially what most people
do that I've seen when a a chatbot brings a like someone brings hey I I've got this and like you can very clearly see like they the chatbot just made up this information. This house doesn't even exist type stuff, right? Um a lot of people immediately have the kind of the visceral orders to be like no don't do that. Right? and they kind of start fighting them immediately on the the response. So you're you're so you're fighting on you're you're kind of having that discussion on terms that don't shouldn't even exist, right? And
11:21 Ask for sources instead of arguing
so the way to do that is to start to one understand how to quality control the the outputs. So, like if you know that that house doesn't even exist, you can say like, "Hey, what did what citations and links did you use to accompany this?" Right? And if they like, "What do you mean citations and links?" Okay, great. Let's talk about um how we can fact check the AI because I I don't think this house may not exist, right? And so, you don't want to go into it like kind of combating them. You want to go into them pointing out flaws that people already know, right? People already know that the AI hallucinates and makes stuff up, right? It's your job just to be like, "Hey, don't forget the AI will just lie to you, and here's how
we can identify if the AI is doing that, right?" And so, you want to make sure that you're consistently having that AI conversation and not running away from
12:15 You, your client and the AI
it. Um, this is kind of what we talked about. Form a you want a team. Basically, it's you, the client, the AI. Like, that's your intelligence trio. Um, and so what you never want to have happen is that you're over here and your client and their AI are over here, right? Because if you get if you space out like that, you're now fighting them the whole way. You've just added a bunch of friction because they trust the AI more than your authority. Whereas, if you would have came in right away and said, "Hey, this is the way I prefer to use AI. This has guard rails. this has and you don't have to be an AI expert.
You just have to know how to use the AI to match your expertise which is the real estate market, right? Like how do or the real estate transaction, this client, this deal, whatever it is that you have in your head that ChatGPT does not. That's your advantage, right? Like that's the we've talked about those context cards. That's where those come into play where they're going to give you much better outputs. So the big goal is for everyone to be kind of on the team. And so the the biggest advantage that you guys have isn't knowing AI. It's the real estate knowledge that you guys have internally that they do not have because everyone in this room has the same access to AI that can we'll we'll obviously get different outputs based on the skill
level and using AI and stuff, but we all have the ability to go to AI and be like I have this problem. Please help. Right? And so the AI isn't that big of an advantage anymore. Understanding specifically how to use it for a real estate transaction is that's the only part that matters, right? Is when you're working with the deal that's on like working with a client on a deal, how do I use the AI to get the outcome I need?
14:06 Four moves for checking AI answers
Um that's the only kind of coverage that you guys need as far as what you really need to know. So the four steps we're going to take is we are going to intervene early. We're going to teach them something called adversarial prompting. We're going to go over something called validation gates and then quality control. So basically we're going to talk to them as soon as we can about AI. We're going to teach them how to use AI to basically audit the AI output they get. We're going to use validation gates, which is basically like I have a checklist of things that this needs to clear for it to be approved. Does it pass those that checklist? And then the last thing is quality control. Just give it the eyeball test. Does it look good?
Does it look bad? Is it missing something? Is it misrepresenting something? Like what is the full spectrum on that?
What is validation gates? So validation gates are basically if you have like a checklist at the end of let's say that you're doing your listing agreement and you have a checklist of you want the listing agreement to use uh bright tone colorful language like nice like friendly you never want the word dream house to be used and it needs to always include bedrooms bathrooms and any special features that would make it unique. So, if it doesn't have those things, you'd have that checklist, then you take that checklist and you'd paste it into the chat and you'd say, "Hey, did did this pass the validation gate?"
And it would review it and it would say, "Yes, it did." At which point, that's when you intervene and you say, "Okay, I need to quality control it and make sure it actually did pass the validation gate." So, the validation gate is basically like the AI's version of quality control.
Right.
Yeah. Yeah. Exactly. Yeah. It's what you're doing, but the AI is doing it first, so you don't have to do it as much. And the with the validation gates, the the human in that has the ability to kind of identify like, oh, maybe this does pass, maybe this doesn't pass, cuz that's the thing that gets through the validation gates is the things that are like, oh, that's kind of slippery. Like, does it work, does it not? Maybe it maybe it passes, maybe it doesn't, right? And so like that's the biggest thing for the the quality control and that human element is like okay what are the things that might be close that
16:29 Ask how they'll use AI
might have snuck in. Okay. So intervene early. So you want to be the first in the conversation not the cleanup crew. So, basically, when you're doing like your um your seller's disclosure, your buyer's disclosure, like whatever onboarding paperwork you have that uh you onboard a client um to do a transaction with, I would highly recommend that you execute these things in that meeting. So, I don't necessarily recommend that you guys bring it up within the first five minutes. Uh that would be weird. Um, but when you see an opportunity, I do highly recommend that you take that opportunity to just ask the question, how are you going to use ChatGPT or AI in this transaction? What what are you what are you thinking? And if because oftent times you'll get, oh, I didn't I
wasn't really even thinking. I would probably, right? Cuz most times they don't start. One of two things happens. either they'll be on AI from the very beginning and you'll they'll come to you with stuff in that initial meeting from ChatGPT or they won't think that they're going to use AI and then something comes up and they're like oh I'll take this and use AI. So you never quite know when it's going to happen in the transaction that they bring you something that's like hey ChatGPT told me about this right and so if you just take all that risk off the table up front by saying hey let's talk about how you're planning to use AI and you this is the position the point where you position yourself as the authority using AI for the real
17:59 Ask where the answer came from
estate transactions. So like as an example, very very easy question is um when you're using AI, how often are you checking links and citations? And you'll get one of two responses. Usually it's either oh I asked for them and every so often I'll click on it and check it out or you get wait you can do that right? Like those are the two things. And so it's your job to have them understand that yes, you can do that and it's very important to do that because if you don't, the AI will just make stuff up, right? You if you're not getting links, citations, um kind of where did you get the information from?
It could come from anywhere. And that's the biggest thing that you want to ask your clients, right? Like, oh, where did this information come from? Oh, ChatGPT. Where did ChatGPT get it? Right? Oh, that was a website from 2012 for a house price. Sounds like it may have made a mistake. Let's talk about it. Right? And so, like, that's it's very, very easy. That's another great question cuz nobody knows where the ChatGPT answers come from. Where did it get that answer from? I don't know. Right?
So, that's one of the big ones. Um, and then I'm going to actually uh work with you guys to get like the prompts will help you create the onboarding for your clients, right? So you'll have kind of that like one-page sheet at the at the end. Um
19:27 Why a wrong answer sticks
but yeah, so basically and this is super duper accurate is a wrong AI answer hardens in their mind really fast, right? So if they go to the AI and they are the cat that found a dead bird, right? They're going to be stoked about it and they're going to expect you to be stoked about it. And when you're like, "This house doesn't even exist," they're going to be like a little upset usually, right? Like that's just human nature is like, "Oh, this somehow this five minute project now has my whole identity tied into it and you're cutting me down, right?" Like that's not what we want to do. So we want to make sure that we are trying to keep the AI answers as fluid in their head as possible until we have real data-backed guidance that says,
20:12 Take it away and check it
"Yes, that's accurate." Or if they come to you and they say, "Hey, uh, you've been really helpful throughout this whole process. the AI gave me this answer. What do you think? And a lot of people like kind of like you said, like they get nervous. They get like, oh, like what do I do here? And it's super duper simple. You can always just say like, great, can you send that in an email? I'll take a look at it. It's sometimes it takes me a couple iterations to get what I want and then I'll get back with you, right? Like you never have to have the answer right there. You're not an a computer scientist, right? As long as you represented that you understand the basics of AI and you know where to get the answer, most people are okay with
that. Um, you are going to have some people that are obviously not, but um, for those people, the best thing I could say is maybe try to schedule like a meeting and sit down and work through it on a call with them or with them in person, right? Because those types of people are usually going to want to have input on your process. And so it may make sense to do like, hey, let's meet at a coffee shop and we'll work through this output, whatever it is.
21:19 Point the doubt at the answer
Um, so this is one of the prompts. This is adversarial prompting. So the goal is to point the doubt at the answer and the AI. It's never to point the doubt at the client and it's the goal is never to have them point the doubt at you, right? And so the way we do that is by establishing one, we're the authority in real estate. two, we know how to use AI to guide you in this transaction, right? And so, um, same tool, better operator. So, the AI that makes you dumber is you always take the first answer, you never push back, you never make it argue the other side, and you never check the output, right? And Um, but the goal is to always be a little bit contentious with the AI, like a little bit of like I don't trust what
you're saying and you're always going to have to prove it. One of my favorite sayings, I forget who said it, but they said in God we trust, all others bring data. Right? Like that is the way. Um, so if you're challenging it, if you're asking it for the strongest objections, if you're asking how a specific person would answer, um, that's helping you, you know, speak the language and get more, um, I guess just more used to working with the best practices of the tool. Um, and so again, you want to stress test your client's outputs. you want to stress test your outputs. So like the three strongest objections a smart local expert would raise in my area. And so like if you're this is also great for um like stuff that you're going to send to clients. So
like before it even gets out to the client um if if you send them an output that's professionally done, it has a citations page. it has, you know, the things that they're looking for. Um, it's going to position you as an authority because you can give them that output and be like, "Here, you can see what I did. Here's what I asked for here. Here's what I asked for here. And this is why the citations are valuable." Um, often times what you can do as well is you can do one without the citations and one with the citations and then you can say like, "Do you see how much is wrong in the one without the citations because it was just able to just make anything it wanted up?" um
23:57 Nothing leaves until it's checked
validation gates. So nothing the the way I like to put this is that they can't sue a chatbot yet. So the liability for the things that the chatbot produces is on the people in the room, right? So um basically until you feel comfortable with the output that output should not leave be being internal.
24:36 Check it with the OODA loop
this. Um, and then one of the ways that we do that is through uh we've talked about a couple times, but it's the OODA loop. So OODA is observe, orient, decide, act. So, it was actually developed by a fighter pilot, uh, John Boyd, in I think the 70s. And so, if it's good enough for someone shooting down other planes and trying not to get shut down, I figured it's probably okay for me. Um, and basically, it's the the idea behind it is the faster you can make as many good decisions as possible, the bigger your advantage is in pretty much all areas of life, right? So if you can win by volume of good decisions, you're going to do great. And so the way this works is you observe what is real. You orient to kind
of where you want to go as far as like what is the general plan. Then you decide, okay, we have a plan. Let's go in this direction. Let's try this outcome. Let's get this. And then you act. You do the plan. And you just see what happens. And then you repeat that. And so that can be very small like what did I have for breakfast? It could also be very large like what is my estate planning look like? Right? And you just run that loop for every decision. And if
Ahead of your clients
Stay ahead of the client who already asked ChatGPT.
Being your client's first AI voice takes practice inside your own business. The Agent Constraint Accelerator gives you that practice on your one constraint, with the goal that you can use AI as the starting point of your own problem solving.