Why Do Zillow and Redfin Show Different Home Values?
Two home value estimates can sit ten percent apart and both be defensible. Here is exactly why AVMs disagree on the same house, and what the size of the spread is telling you about the property.
Zillow and Redfin show different values for the same house because they are different models, built on partly different data, making different assumptions, and quietly answering slightly different questions. Neither one is reading a value off a shelf, because the value of a specific house is not a fact that exists until someone buys it; each model produces one point estimate from a wide distribution of plausible prices, and two reasonable models land in different places on that distribution. A gap of a few percent up to roughly ten percent between two major estimates is normal and is not evidence that either is broken. A gap of twenty five percent or more is a signal, and it almost always points at something specific and knowable about the property. This article explains the mechanisms that create the disagreement and how to read the spread as information.
Why do two AVMs looking at the same market disagree?
They are not looking at the same data
Coverage is uneven. Listing data comes from hundreds of separate multiple listing services with different rules about what gets shared, with whom, and how fast. A model can have a rich direct feed in one metro and be working mostly from county records in the next county over. Add off-market and pocket sales that never hit a public feed, plus owner-submitted corrections that one platform accepts and another ignores or weights differently, and the two systems are describing slightly different houses before any math happens.
They choose different comparable sales
This is usually the largest single cause. Every AVM has rules for which recent sales are allowed into the comparison set: how far out to search, how far back to look, how much size difference to tolerate, whether a townhouse may be compared to a detached house, whether the search may cross a school district line or a municipal boundary. Change any one of those rules and the answer moves. Two models with identical math and different comp rules will not agree.
They adjust for differences differently
Once a comp set exists, the model has to reconcile houses that are not identical. One system may work from a price per square foot band, another from a regression that assigns a coefficient to each attribute, another from a gradient boosted ensemble that has learned interactions no one wrote down. The same five sales can support meaningfully different conclusions depending on how the adjustment is done. Two platforms looking at the same address on the same afternoon can be holding different information vintages. This is a very common explanation for a temporary gap that closes on its own within a week or two.
They are optimizing for different things
An AVM has a loss function, and the choice of loss function is a product decision. A model tuned to minimize its median error will behave differently at the extremes than one tuned to avoid large misses, or one calibrated to be deliberately conservative because a lender is going to rely on it. These are all defensible choices that produce different numbers on the same house.
What makes the gap wide?
| Driver of divergence | What it does to the number | How you can spot it |
|---|---|---|
| Thin comp density | Both models extrapolate, and they extrapolate differently | Few recent sales within a mile |
| Non-disclosure state | Sale prices are inferred rather than known | Texas and roughly a dozen other states |
| Recent renovation | One model caught it in the listing text, the other did not | Photos show new finishes, records show none |
| Property needs work | Both assume closer to average condition, by different amounts | Dated photos, as-is language in the listing |
| Submarket boundary nearby | One model crossed it for comps, the other did not | A highway, a district line, a different HOA within half a mile |
| Unusual size or lot | The subject sits outside the comp cloud | Largest or smallest home on the street |
| Different update dates | One model has priced in a recent event, the other has not | A price cut or a nearby closing in the last two weeks |
Is the average of two estimates better than either one?
Sometimes, and less often than people assume. Blending independent estimators does reduce error in general, and that is real statistics. The problem is the word independent. Consumer AVMs draw from heavily overlapping sources: the same listing feeds, the same county records, the same recent sales. They share their blind spots, so averaging them does not cancel the shared bias, it just relabels it. Worse, when two estimates are far apart for a structural reason, say one of them pulled comps from across a boundary into a cheaper submarket, the average splits the difference between a defensible number and an indefensible one. You end up half wrong on purpose. It is more useful to find out why they disagree than to average them.
How should I use the spread?
Treat the spread as a confidence reading on the property, not a scoreboard between the companies. - Tight spread, a few percent. The house is ordinary for its area and there are plenty of comps. Any of the estimates is a reasonable starting frame. - Moderate spread. Something is ambiguous. Usually condition, a boundary, or a size that sits at the edge of the comp cloud. Go look at the comps before using any number. - Wide spread. Do not use an automated number at all without human review. This is exactly the population where a walkthrough, an agent's opinion, or an appraisal earns its cost. If you want the comparison set in front of you rather than a single figure, our CMA report shows the comps that produced the value and lets you adjust or exclude them, which turns "these two numbers disagree" into "here is the specific sale causing the disagreement."
Which estimate is actually right?
The only honest way to answer that for any AVM, including ours, is to grade estimates against real closing prices on the same properties, with each estimate fixed before the outcome was known. That is the reason we freeze every estimate when a property lists and then grade it against what the home actually closed for, and publish the running result at /accuracy. It is not a claim that we beat anyone. It is a claim that the scoring is done in public, on a sample we did not get to pick after the fact, which is the part of this industry that has historically been missing.
What to remember
Two different numbers for one house is the normal output of two different models working on incomplete information. The number that matters is not which brand is higher, it is how far apart they are and why. Wide disagreement is a flag on the property. If you want to understand the machinery underneath, start with what an AVM is, then read why estimates get it wrong. If you would rather just run the numbers, the calculators at free tools are open without an account.
Frequently Asked Questions
Which home value estimate should I trust when they disagree?
Neither, until you look at why. Open the comparable sales behind each number. Usually one of them included a sale from a different submarket, an unrenovated comp for a renovated house, or a stale sale. Once you find that, you know which figure to discount without needing to pick a favorite brand.
Is it better to average the two estimates?
Rarely. Averaging works when the estimators are independent, and consumer AVMs are not; they draw on heavily overlapping listing feeds and county records, so they share their blind spots. When two estimates are far apart for a structural reason, averaging just splits the difference between a defensible number and an indefensible one.
Does a wide gap between estimates mean something is wrong with my house?
Not wrong, but usually unusual. Wide spreads cluster on homes with thin comp density, recent renovation, deferred maintenance, an odd size for the street, or a submarket boundary nearby. The disagreement is a property signal, and it is a good reason to get a human opinion before relying on any single figure.
Ready to Start Investing Smarter?
Join 2,000+ investors using Resideline.
Start free with 3 reports a month.