A lot of agents now pull comps from the MLS and hand them to ChatGPT to write up. It produces something that reads beautifully. Here is what it cannot do, and why that matters to the person you hand it to.
The short version. A language model is a writing tool, not a valuation method. Give it five comps and it will produce fluent, confident prose about them — but it did not choose those comps, it cannot tell you whether they were the right five, and it does not apply an adjustment the same way twice. Ask it the same question tomorrow with the same five sales and you will get a different number.
The report is not the valuation. The method is. AI makes the report look professional; it does not make the number defensible.
Value is simple. Good comps and good adjustments equal good values.
Everything else is formatting.
None of these are criticisms of the model. They are things outside what a language model is for.
The hardest and most consequential part of a valuation is deciding which sales count and how much each one counts. That happened before you opened the chat window, in your head, under time pressure. If you unconsciously picked the three sales that support the number you wanted, the AI will write a persuasive case for exactly that number and never flag it. Truppraisal scores every recent nearby sale 0–100 for similarity and lets the score decide the weighting, so nobody hand-picks the flattering three.
The sales you pasted in are current — you pulled them this morning. That is not the problem. The problem is that a model has no way to judge whether those five were the right five: whether a closer, newer, more similar sale was sitting two streets over, whether one of them was a trustee sale or a transfer between family, or whether the set you happened to pick is the set that supports the number. It will write confidently about whatever you hand it. Truppraisal reads the whole feed and rates every candidate, so the comps in the report are the best ones available, not the best ones you had time to find.
Ask it what a pool is worth and it will give you a plausible range and a caveat. Ask it twice and the range moves. A real adjustment grid applies the same rate to every comp in every report: living area, lot, parking, pool, bedrooms, baths, proximity and how long ago the sale closed — eight lines on every comp. That consistency is the entire point — it is what makes two reports comparable to each other, and what lets you defend a number to a seller who disagrees with it.
A language model's confidence is a writing style, not a measurement. It reads exactly as certain about a thin comp set in an unusual neighbourhood as about twelve near-identical sales on the same street. Truppraisal grades each report's confidence from how much the adjusted comps actually agree with each other, and prints the basis for that grade.
| Comps pasted into an AI chat | Truppraisal | |
|---|---|---|
| Chooses the comparable sales for you | No — you pick them first | Yes — it rates every sale in the feed |
| Rates each comp by measured similarity | No | TruComp Rating, 0–100 |
| Dollar adjustments at consistent rates | A different answer each time you ask | The same rates, every comp, every report |
| Same inputs produce the same number tomorrow | No | Yes |
| Confidence measured, not asserted | Confidence is a writing style | Graded from how closely the comps agree |
| Goes to your whole database every quarter, under your name | No | Yes, automatically |
This is not an argument against using it. It is an argument about which job you are giving it:
The distinction worth holding onto: use it to communicate a valuation, not to produce one.
Truppraisal is an automated valuation model — arithmetic applied consistently to MLS data, not a language model writing prose. The TruComp Rating is a scoring formula over measurable attributes: size, age, lot, garage, pool, distance, how recently the sale closed. The adjustments are consistent rates multiplied by measured differences. Run it twice on the same data and you get the same number, which is the property that matters.
It is enough right up until a seller asks where the number came from, or lists with someone else who could answer the question. A polished document built on an unexamined comp set is a risk you are carrying without knowing it — and every other agent has the same tool, so it stops being a differentiator the moment it becomes common.
No. It is an automated valuation produced by a computer model, not prepared or reviewed by a licensed appraiser and not prepared under USPAP. It cannot be used for lending, mortgage, credit, tax, legal, insurance or estate purposes. It uses the sales comparison approach an appraiser uses, which is why it shows its work.
You type an address. The comps, the scoring, the adjustments and the report are produced for you — including from your phone, standing at an open house, sent to the client before you leave.
See a complete report › See pricing ›
Also comparing Truppraisal and Homebot · Truppraisal and HouseCanary · all of them on one page.