How to Check Whether a Home Value Estimate Is Accurate
You audit an estimate by opening its comparable sales, not by comparing it to a second estimate. A six step check on any individual number, plus the four questions any accuracy claim must answer.
You check whether a home value estimate is accurate by auditing its comparable sales, not by comparing it to a second estimate. Two AVMs agreeing means they chose similar comps from similar data, which is a statement about their inputs rather than evidence that either number is right. There are really two separate questions here, and it helps to keep them apart. The first is whether the model has a credible track record, which is a question about the provider. The second is whether this particular estimate on this particular house is sound, which is a question you answer yourself in about ten minutes. Here is how to do both.
Part one: does the model have a real track record?
Most accuracy claims in this industry are unfalsifiable as written. Four questions separate a real one from a marketing number.
When was the estimate made relative to the outcome?
This is the question that matters most and gets asked least. Grading that estimate against the closing price is measuring the seller's judgment as much as the model's. An accuracy claim is only meaningful if the estimate was fixed before the outcome was knowable. Ours are frozen at the moment a property lists and then graded against the actual closing price when the sale records, which is what the accuracy page reports.
What was the sample?
Ask whether the figure covers all properties in a market or a filtered subset. Filters that quietly improve a number include: on-market homes only, single family only, sales within a normal price band, properties with a recent prior sale, and homes with complete data. Each filter is defensible in isolation, and stacked together they can turn a mediocre model into an impressive headline.
What was it graded against?
The recorded closing price is the only real answer.
Can you see the sample size and the distribution?
A median with no sample size behind it is a slogan. A median with hit rates around it, the share of estimates within 5%, 10%, and 20%, tells you the shape of the error and therefore your realistic downside. More on how to read those figures in median error rate explained.
Part two: auditing one specific estimate
Step 1: open the comps
If the tool will not show you which sales produced the number, stop. You cannot verify a black box, and every remaining step depends on seeing the comparison set. Ours are visible and adjustable by design, and you can pull a full set for any address with the CMA report.
Step 2: check each comp against a short list
| Check | What good looks like | What should worry you |
|---|---|---|
| Living area | Within roughly 20% of the subject | A comp 40% larger carrying heavy weight |
| Sale date | Within about 6 months in a moving market | Sales over a year old presented as current |
| Distance | Same subdivision or same submarket | Nearest by distance but across a boundary |
| Property type | Same type and ownership form | A townhouse comping a detached house without adjustment |
| Age and style | Same era and construction | A 2019 build comping a 1962 ranch |
| Condition | Comparable finish level | Renovated flips comping an unrenovated subject |
| Transaction type | Arm's length, market exposure | Family transfers, auctions, or off-market deals treated as normal |
Step 3: check the condition assumption
Ask what condition the model assumed. If it assumed average and the house needs a roof, the estimate is high by roughly the cost of the work plus the discount buyers apply for the inconvenience and risk of doing it, which is usually more than the raw cost. If it assumed average and the home was recently renovated, the estimate is low. This is the largest single correction most people need to make, and it is directional, so you can apply it with confidence even when you cannot size it precisely. If you need to size it, the free rehab cost estimator gives a defensible starting budget.
Step 4: check the submarket, not the radius
Pull up a map and ask what sits between the subject and each comp. Highways, rail lines, school attendance boundaries, city limits, flood zones, and subdivision lines all create price steps. A sale a quarter mile away on the other side of one of these is frequently a worse comp than a sale a mile away inside the same community. This is one of the most common causes of a correct valuation that looks wrong, and of an incorrect valuation that looks reasonable.
Step 5: sanity check the price per square foot band
Take the adjusted comps, compute price per square foot for each, and see where the subject's implied figure sits. It should sit inside the band, not at the edge and not outside it. If the estimate implies a figure above every comp in the set, something in the model's adjustments is doing work you should be able to name. If you cannot name it, do not trust it. Price per square foot is a poor primary valuation method and an excellent error detector.
Step 6: check the dates on everything
When was the estimate generated? When did the most recent comp close? When was the assessor record last updated? In a market moving several percent a quarter, a three month old estimate built on nine month old comps is describing a market that no longer exists, no matter how well the model performed when it ran.
Red flags in any home value estimate
- •No visible comps, or comps you cannot exclude or adjust. - A single number with no range and no confidence measure. - Accuracy claims with no sample size, no date, and no statement of what was graded against what. - Comp sets that cross an obvious physical or administrative boundary. - An as-is value that equals the after-repair value on a property that visibly needs work. - Rural or non-disclosure market properties presented with the same apparent confidence as a suburban tract home.
The honest bottom line
You cannot make an automated estimate correct, but you can nearly always find out why it is wrong, and wrong for a knowable reason is a usable number. The comp set is where the reasoning lives. Providers who hide it are asking you to trust a brand instead of a method. We publish our graded results at /accuracy and keep the comps in the open for the same reason. If you want to run a check right now without an account, start with free tools.
Frequently Asked Questions
Is comparing two home value estimates a good accuracy check?
No. Consumer AVMs draw on heavily overlapping data, so agreement between them mostly means they picked similar comparable sales. It is a statement about their inputs, not evidence that either is right. Auditing the comps behind one estimate tells you far more than lining up three numbers.
How recent do comparable sales need to be?
In a market moving several percent a quarter, aim for sales within about six months. Six to twelve months is usable with a market adjustment. Beyond twelve months you are describing a different market, and any estimate leaning on those sales should be treated as a rough range rather than a number.
What is the single biggest red flag in a home value estimate?
No visible comps. If you cannot see which sales produced the number, you cannot check any of the things that actually cause error: whether the comps crossed a submarket boundary, whether they are the right size, whether they are renovated, or whether they are stale. A hidden comp set asks you to trust a brand rather than a method.
Ready to Start Investing Smarter?
Join 2,000+ investors using Resideline.
Start free with 3 reports a month.