Why every home value site gives your house a different number

They disagree because each one is a separate prediction, built by a separate company, from data that only partly overlaps.

An automated valuation model, or AVM, takes what is known about a home and about nearby sales, then produces the price it thinks is most likely. Every company chooses its own inputs, its own way of weighting them, and its own rules for which sales count as similar. Change any of those choices and the answer moves.

So when two sites land far apart, it does not mean one of them made an error in the usual sense. It means two sets of assumptions produced two outputs, and neither set of assumptions was checked against your actual house.

Mostly the same public raw material, cleaned and filled in differently.

Typical inputs include county property records, tax assessments, past sales, old listing details and recent sales nearby. One site may have picked up the listing photos and description from when the home last sold. Another may only have the county file, which might still show the original bedroom count from decades ago. A third may have a newer record of a permit.

Small differences in the facts produce big differences in the answer. A missing bathroom or a wrong square footage figure is enough to shift a whole estimate.

The gap grows wherever the model has less to go on.

A plain house on a street full of similar homes that sell often gives every model plenty of close matches, so the estimates tend to bunch together. A home on acreage, a custom build, a property with major renovations, or anything in a place where few homes change hands leaves each model reaching further for comparisons. Each reaches in a different direction, and the numbers fan out.

None of them is right in any official sense, because none of them is an opinion that anyone has signed.

It is tempting to trust the highest number, or the site you have heard of most. Neither habit makes an estimate more accurate. Averaging does not help much either: if every model is missing the same fact, such as a new roof or a failing foundation, the average misses it too.

A wide spread is a warning light, not a problem to be solved with more lookups.

When the estimates cluster, the models are at least agreeing about the records. When they scatter, the models are uncertain, and the reason is usually something about your house or your area that the records do not capture. That is the situation where a person who inspects the property adds the most.

Lenders feel that uncertainty too. Federal agencies adopted quality control standards in 2024 for AVMs that mortgage originators and secondary market issuers use to value homes, including requirements meant to ensure a high level of confidence in the estimates. A free estimate on a consumer website is not something a lender treats as an appraisal.