Ask a general AI tool what a four-bedroom home on a specific street is worth and it will give you a confident answer. The answer will sometimes be roughly right, often by accident. Most of the time it will be off by a margin large enough to lose you the listing if you walk into a kitchen-table conversation with it. The model has read enough about your suburb to sound informed. It has not seen the three sales that closed last week, the local council change that’s about to lift values north of the railway line, or the fact that the comparable two doors down sold under-market because the owners were divorcing.
This matters because the listing presentation is the moment a seller decides whether they trust you. AI used badly will make you look lazy. AI used well will make you look like the agent who did three days of work in three minutes. The difference is the workflow.
What AI is actually good at on pricing
Three things, reliably. Synthesising a list of comps you provide into a clear, defensible narrative. Translating a technical pricing analysis into language a homeowner understands. Stress-testing your own thinking by playing devil’s advocate. None of these involve the model deciding the price. They all involve the model helping you communicate or pressure-test a price you’ve already arrived at using your own market data.
What AI is not reliably good at: knowing current local prices, applying recent market shifts, accounting for off-market sales, or weighing the kind of nuance (“the kitchen is original 1995 and that scares 35-year-old buyers more than the price suggests”) that an experienced agent picks up at the inspection. Don’t ask it to do those things. Do them yourself, then let AI help with the parts that follow.
The comp synthesis prompt
Start by gathering your comparable sales the way you always have: from your MLS or local equivalent, plus any off-market intel you’ve collected. The CMA tools built into Lofty, Follow Up Boss, and the major MLS portals will produce a list of comps with sale prices, dates, and details. That list is your input.
Paste it into ChatGPT or Claude with this:
“You are helping me prepare a pricing narrative for a listing presentation. I will provide the subject property and a set of recent comparable sales in the same suburb. Do not invent or estimate prices that aren’t in the data I provide. Your job is to: (1) group the comps into tiers based on how truly comparable they are to the subject (close match, partial match, weak match) and explain the reasoning, (2) identify the median and range of sale prices in the close-match tier, (3) note any patterns in time-on-market or sale-to-asking ratio, (4) flag any comps where the sale circumstances suggest the price may not reflect open-market value, (5) write a 120-word narrative paragraph I can paste into my listing presentation that explains the pricing recommendation in plain English. Subject property: [ADDRESS, BEDS/BATHS, SQ FT, KEY FEATURES]. Comps: [PASTE 5-10 COMPS WITH PRICE, DATE, SQ FT, KEY DIFFERENCES].”
The output won’t tell you what to price the property at. It will tell you what your data is saying, in language a seller can follow. You decide the price. The narrative is what makes the price defensible across the kitchen table.
The seller-language translation prompt
The pricing analysis a seller needs to see is almost never the one your CMA produces. The CMA is dense, full of acronyms, and structured for an internal audit. The seller wants three things: where their home sits in the market, why, and what would change if they moved the asking price up or down. AI is excellent at this translation.
“Below is a technical pricing analysis I’ve prepared for [ADDRESS]. I need to present this to the homeowner, who has no real estate background. Rewrite it as a one-page document with three sections: (1) ‘Where your home sits in today’s market’ (75 words, plain English, no jargon), (2) ‘What we’d expect at three different price points’ (a short table showing aggressive, fair, and conservative pricing with the implication for time-on-market and likely buyer pool at each), (3) ‘My recommendation and why’ (60 words, confident but not pushy). Avoid percentages where a sentence would work better. Don’t oversell, the seller will be more receptive if I’m honest about trade-offs. Technical analysis: [PASTE].”
A specific failure mode to watch for. The model will sometimes round numbers in ways that change the meaning, or smooth over a trade-off you wanted explicit. Always cross-check the numbers in the rewritten version against your source CMA. The narrative is for the seller, the numbers have to match what your data actually says.
The pre-meeting pressure test
This is the prompt better agents run before a tough listing meeting. You’ve got a price you believe in. The seller will push back. The pressure-test prompt prepares the conversation.
“I’m walking into a listing presentation for [ADDRESS]. My recommendation is [PRICE], based on the comps below. The sellers are likely to push back because [SPECIFIC REASON]. Play a sceptical homeowner who wants a higher number. Ask me the three hardest questions, and after each give me a short, evidence-based response that doesn’t sound defensive. Flag any weakness in my pricing logic. Rationale: [PASTE]. Comps: [PASTE].”
The model will sometimes invent objections that don’t apply, but it will reliably surface the one or two questions you hadn’t thought to prepare for. Five minutes of this before a hard meeting is worth doing.
A counterintuitive insight on AI and pricing accuracy
The agents who price most accurately are not the ones using AI most aggressively. They are the ones using AI to communicate a price they arrived at the old way. AI doesn’t improve the price recommendation, it improves the seller’s confidence in it. That second effect is what wins listings, even when the recommended price is lower than what a competing agent suggests, because the seller can see the work. The competing agent in the same kitchen, with a slightly higher number and no articulated reasoning, loses to the agent whose AI-translated narrative makes the lower number feel like a more credible plan.
Where the model can hurt you
Three places to watch. First, never paste a seller’s name, address, or personally identifiable detail into a free AI tool that you wouldn’t paste into a public forum. Second, never use AI’s gut estimate of a price as your starting point. It is anchored on training data that may be a year or more out of date. Start with your own data, then bring AI in. Third, watch for confidently wrong neighbourhood claims. The model will sometimes assert a school zone or a transport upgrade that isn’t actually true. Verify before you put anything in writing.
The agents winning on pricing right now are using AI to do in three minutes the writing and translation work that used to take three hours, and using the saved time on the parts of the job that genuinely require a human in the room.