How Accurate Is a Zestimate? The 2026 Numbers, Explained
An automated valuation model is a starting point, not a price opinion. The gap between those two things is where deals are won and lost.
Short answer
As of 2026, Zillow's Zestimate carries a median error rate of about 1.74% for on-market homes and about 7.20% for off-market homes. On a $500,000 house, that off-market figure is a ±$36,000 swing at the median — meaning half of all estimates are off by more than that. Off-market Zestimates land within 5% of the sale price only about 38% of the time.
Why there are two very different accuracy numbers
Most people quote the on-market figure, because it is the one that sounds impressive. It is also the one that matters least.
When a home is listed, the Zestimate gets to see the list price. A model that has been handed the seller's own asking price and then predicts something close to the eventual sale price is not doing much independent work — it is largely reflecting a number a human already set. That is why the on-market median error compresses to under 2%.
The off-market number is the honest one. That is the model working from tax records, prior sales, square footage, and neighborhood signals with no list price to anchor to. Per HomeLight's 2026 comparison of home value sites, the off-market Zestimate lands:
- within 5% of sale price — 37.7% of the time
- within 10% — 61.8% of the time
- within 20% — 82.6% of the time
Flip that last one around: roughly one in six off-market estimates is more than 20% wrong. On a $500,000 home, that is a six-figure miss.
What an AVM structurally cannot see
The error is not sloppiness. It is a category limitation. An automated valuation model is inferring value from records, and records do not contain:
- Condition. A gutted kitchen and a $90,000 renovation look identical in county data.
- Why a comp sold for what it sold for. Divorce, probate, an off-market cash deal to a neighbor, a seller who needed out in eleven days — all of it lands in the dataset as a clean arm's-length sale.
- Micro-location. The side of the street that backs onto the arterial road. The unit with the view. The lot that floods.
- Concessions. A $500,000 sale with $18,000 in seller-paid closing costs and a rate buydown is functionally a $482,000 sale. The recorded price says otherwise.
- Recency at the turn. Models lag. In a market that moves quickly in either direction, the model is describing the market of six weeks ago.
None of that is fixable with more data of the same kind. It requires someone to look.
How Zestimate compares to other AVMs
Redfin publishes a 1.92% median error for on-market properties — the same order of magnitude, and subject to the same list-price anchoring. Most other consumer estimators (Realtor.com, bank-run tools, FSBO sites) publish no accuracy data at all, which is its own kind of answer.
The practical takeaway: AVMs disagree with each other by more than their stated error rates suggest they should. When three estimators give you three numbers spread across $60,000, none of them is "the value." They are three guesses with different training data.
When a Zestimate is actually useful
It is a fine tool for the job it is built for:
- Rough screening. Sorting 40 addresses into "worth a look" and "not close."
- Trend direction. Watching a Zestimate move over 18 months tells you something real about the neighborhood even if the level is wrong.
- Setting expectations with a client who already looked. Your seller has seen the number. Pretending it does not exist is worse than explaining it.
How to handle the "but Zillow says" conversation
Almost every seller arrives at a listing appointment with a Zestimate already in their head. Arguing with it directly reads as defensive. What works better is showing the seller the mechanism:
- Acknowledge the number out loud first. Say it before they do.
- Explain the two error rates. Most sellers have never been told there is an off-market figure, and that it is four times larger.
- Show the comps the model used versus the comps you would use. This is usually the moment it clicks — the algorithm pulled a 1,900 sq ft ranch eleven blocks away in a different school zone.
- Give a range, not a rival single number. Replacing one point estimate with another point estimate is just a coin flip with better manners. See how to build a CMA that wins listings for why the range framing works.
AVM vs. CMA vs. appraisal, in one table
| AVM (Zestimate) | CMA | Appraisal | |
|---|---|---|---|
| Who produces it | Algorithm | Licensed agent | Licensed appraiser |
| Sees the property | No | Usually | Yes |
| Adjusts for condition | No | Yes | Yes, formally |
| Typical cost | Free | Free / included | $400–$700 |
| Accepted by lenders | Rarely | No | Yes |
| Turnaround | Instant | Minutes to days | Days to weeks |
We go deeper on the middle two columns in CMA vs. appraisal.
The bottom line
A Zestimate is a free, instant, unverified opinion generated by a system that has never seen the house. That is genuinely useful for triage and genuinely dangerous for pricing. The 1.74% headline number describes a scenario — an already-listed home — where the model is grading its own homework.
If you are pricing a property that is not on the market yet, which is every listing appointment you will ever walk into, the number that applies is 7.20%. Treat it accordingly.
Frequently asked questions
What is the Zestimate's median error rate in 2026?
About 1.74% for on-market homes and about 7.20% for off-market homes nationwide. The off-market figure is the relevant one for any property that has not been listed yet.
Is the Zestimate ever higher than the actual sale price?
Frequently. The median error is a measure of magnitude, not direction — Zestimates run high and low roughly symmetrically. Off-market estimates fall within 5% of the eventual sale price only about 38% of the time.
Why is my Zestimate so different from my agent's price opinion?
The model cannot see condition, renovations, concessions, or micro-location, and it may be using comparable sales an agent would reject. A CMA selects comps deliberately and adjusts for the differences; an AVM applies statistical averages.
Can I get my Zestimate corrected?
You can claim the home on Zillow and update facts such as square footage, bed/bath count, and finished basement space, which usually shifts the estimate. It does not make the underlying model see condition or quality.
Should sellers ignore the Zestimate entirely?
No — sellers and buyers both see it, so it shapes the negotiation whether or not it is accurate. The productive approach is to explain what the number is measuring rather than dismiss it.