How AI is changing what your home is "worth" — and why you should care
Three browser tabs, three different home-value estimates, one house. The spread between them is sometimes wider than a typical down payment, and none of them is exactly right. Here's what they're actually doing, where they break, and how to read them without overweighting any single one.
A short history of the AVM
Automated Valuation Models — AVMs — started as glorified property-tax-assessor algorithms: square footage, lot size, bed and bath count, location, recent comps. Reliable enough for the assessor, blunt enough that no one took them seriously as a sale price.
The 2010s consumer products layered on neighborhood-level data and more aggressive comp matching. The 2020s versions started reading interior listing photos — granite versus laminate, refinished versus carpeted — and built models around comp density, commute corridors, and walkability scores. By 2026 the front-running AVMs are multi-modal: they read the photos, the listing copy, the public records, and the recent comps all at once.
Accuracy has gotten better. It still varies considerably by market. The major AVM providers publish their own median error rates, and the spread is significant — dense, active markets are reasonably reliable; thin or idiosyncratic ones are wildly variable. The model can't see what it hasn't seen.
Where they break
A few categories of homes the model gets wrong, systematically.
Unique homes. A custom build, an architectural one-off, a renovated historic — anything where the photo speaks to a value the comp set doesn't capture. The model averages toward the neighborhood. The house refuses to be averaged.
Post-purchase renovations. If a kitchen was gut-redone last spring and the listing photos still show the old layout from three owners ago, the model believes the photos. The renovation is invisible until the house lists again.
Hyperlocal pockets. One side of a block backs up to a busy commercial corridor; the other side backs up to a park. The model often uses block-level data and averages the two. The fence line tells a different story.
Thin submarkets. Five sales in 12 months isn't enough data to learn a market. The model leans harder on the comps it has, and those comps don't carry much weight.
A sane way to read them
Pull three AVMs from three different providers. Treat the spread as the signal — a tight cluster suggests the model is confident; a wide spread says the comp set is noisy or the house is unusual. Take the midpoint as a hypothesis, not an answer.
That gets a homeowner most of the way to the right ballpark. It does not get them to a listing price.
The Graham inversion
Benjamin Graham, writing in The Intelligent Investor, gave investors a discipline that turned out to apply just as well to homes. His move: do the diligence before buying, then stop calling for an appraisal four times a day. The daily quote is noise. The diligence is signal.
From the bookshelf — The Intelligent Investor, Benjamin Graham
"In the short run, the market is a voting machine, but in the long run, it is a weighing machine."
Graham's frame for stock prices — short-run noise, long-run weight — is exactly what AVMs are now doing to home values, three browser tabs at a time.
AVMs flip the picture. Homeowners can now — and do — call for an appraisal four times a day, across three different services, on their own house. The daily fluctuation is mostly noise. The discipline still applies; the direction just reversed. Glance occasionally. Don't trade on it.
Why a real CMA still wins
A comparative market analysis from an agent who has walked the block sees what a model can't. The new deck. The recently replaced furnace. The awkward setback. The neighbor with the chickens. A CMA weighs the active competition the way a buyer's agent will weigh it, anchored in the recent comparable sales the data actually supports.
The AVM has a place — first ballpark, casual curiosity, the napkin sketch before a refinance conversation. For an actual pricing decision, it's a starting point, not a finishing one.
For buyers on the other side of this conversation, the inverse problem hides in HOA fees — a monthly cost that lenders fold into the qualifying math but most house-hunters never run. Wrote that one up: the hidden cost of buying in a Cincinnati HOA neighborhood.
Anyone curious what a real CMA looks like on a Cincinnati-area home: send the address.