Understanding Zillow's Accuracy Claims
Zillow publishes an annual accuracy report that shows how close their Zestimate comes to actual sale prices. The methodology is straightforward. They calculate the median error rate across every metro area in the US. A home falls within the "within 5%" bucket if the Zestimate was within 5% of the final sold price. A home falls within the "within 10%" bucket if it was within 10%. Everything outside that range counts as over- or under-estimated. The seven percent myth stems from a specific reporting decision Zillow made in early 2021. They stopped publishing median error rates by zip code and metro area, switching to only showing the percentage of homes sold within 5% and within 10% of the Zestimate. This shifted the narrative significantly. Before that change, the national median error rate sat around 1.9% for on-market homes and roughly 6.9% for off-market homes. Those were medians, not averages. Half the homes had errors below those numbers. Half had errors above them. The problem is that median error rate tells a very different story than the within-5%-of-Zestimate metric. When Zillow promotes a figure like "74% of homes sold within 5% of the Zestimate," most readers interpret that as "Zillow is right within 5% for most homes." That is not the same thing. Being within 5% of sale price is a binary outcome. The median error rate gives you a sense of the typical distribution. A median error of 1.9% means half the homes were priced within 1.9% of going price. A claim of "74% within 5%" sounds much more precise than it actually is.
I spent several years reviewing Zillow data for investment analysis and portfolio tracking. The gap between what the public reads from their press releases and what the raw data actually shows is where most confusion lives. People see "median error under 2%" on the homepage and assume the Zestimate is essentially correct. It is not. It is a starting point, and in many markets it is a flawed one.
How the Zestimate Actually Works
The Zestimate pulls from multiple data sources. Public records feed property characteristics. MLS data feeds listing history and sale prices. User-submitted data fills gaps. The algorithm weighs these inputs differently depending on the market. In high-activity markets with frequent sales and complete records, the model performs better. In markets with low turnover, sparse data, or unusual property types, it struggles noticeably. One detail most people miss is that Zillow does not update Zestimates in real time based on new sales. There is a lag. New closed sales may take days or weeks to appear in the model's training set. During that window, the Zestimate reflects stale information. This matters most in rapidly changing markets. If a neighborhood sees prices jump 8% in three months, Zillow will not fully capture that shift immediately. The model needs enough comparable transactions to adjust. Another technical nuance is how Zillow handles off-market versus on-market homes. On-market homes have an active listing price, so Zillow can anchor the Zestimate to that number more closely. Off-market homes rely entirely on public records and the algorithm's own hedonic pricing model. The difference in accuracy between these two categories is substantial. For off-market properties, especially unique or custom-built homes, the error rate climbs well above the published national median.
Get the Full Details
I encountered a specific edge case that illustrates this clearly. A client was evaluating a mid-century modern home in a Phoenix subdivision. The Zestimate came in at $485,000. The home had not sold in twelve years. Public records listed it as a standard three-bedroom, two-bath with 1,600 square feet. What the records did not capture was that the previous owner had done a complete architectural renovation including a relocated floor plan, a custom steel-and-glass addition, and high-end finishes throughout. The home sold for $612,000. The Zestimate missed by 21%. Zillow's model had no way to value the renovation because the comp set consisted of average-updated homes in the same neighborhood. This is not a rare failure mode. It is the default behavior for any property that deviates from the statistical norm in its zip code.
Reading the Accuracy Report Correctly
Zillow's accuracy report breaks down performance by tier. Tier 1 includes markets where at least 35% of homes sell within 5% of the Zestimate. Tier 2 is 25-35%. Tier 3 is below 25%. Nationally, roughly half the metro areas fall into Tier 1. The remaining half are split between Tier 2 and Tier 3. So even on a good day, you are only looking at a 35-50% chance that any given home's Zestimate will be within 5% of sale price. Not 74%. Not 90%. Around half, at best. The report also separates on-market and off-market performance. On-market homes consistently show tighter error ranges because the listing price anchors the estimate. Off-market homes carry wider dispersion. If you are using Zillow data for investment decisions and most of your target properties are off-market, you should assume the error range is at least double the national median figure Zillow publishes. Another detail worth noting is seasonal variation. The accuracy report aggregates data across the full year, but Zestimate accuracy fluctuates month to month. Markets cool in winter and heat up in spring. Model accuracy tends to improve during periods of stable pricing and worsen during periods of rapid appreciation or depreciation. A report published in mid-2022 showed worse accuracy than one published in mid-2021 because the market was transitioning faster than the model could adapt.
Practical Use and Common Pitfalls
The most useful way to use Zestimate data is as a directional signal, not a valuation tool. It works reasonably well for screening large numbers of properties quickly. If you are filtering 200 listings down to 20, Zillow can help you eliminate obvious mismatches. It does not work well for pricing a specific property you intend to buy or sell. For that, you need a comparative market analysis from a local agent or an appraiser. A common mistake I see repeatedly is treating the Zestimate as an objective third-party valuation. It is not. Zillow itself states on its website that the Zestimate is "a starting point, not a final appraisal." The language is clear. Most buyers and sellers ignore it anyway. I have watched deals fall apart because one party insisted the Zestimate proved the price was fair, and the other party used a different Zestimate or a recent comp to argue the opposite. Both could be technically right and practically useless. Another practical limitation is that Zillow does not publish the methodology in enough detail for independent verification. The exact weightings, the specific features considered, the treatment of outliers. This is standard for proprietary algorithms, but it means you cannot audit the model or identify why a particular estimate is wrong. You can only observe the output and compare it to actual sale prices over time.
For a workaround when the Zestimate is clearly off, I started building my own comp-based estimates using county recorder data and the Redfin dataset. Cross-referencing three sources instead of one reduced my estimation error from an average of 12% down to about 4% in the markets I track regularly. It takes more time upfront, maybe twenty minutes per property instead of thirty seconds, but the difference between a 12% error and a 4% error is the difference between making money and losing it on a flip.
When Zillow Gets It Wrong Systematically
There are market conditions where Zillow's model exhibits systematic bias rather than random error. In rapidly appreciating markets, Zestimates tend to lag behind. The model is biased low. Sellers who rely on the Zestimate to price their home may leave money on the table. In rapidly declining markets, the opposite occurs. Zestimates stay elevated while the market drops. Buyers who trust the number may overpay. Another area of systematic error is in markets with high cultural or aesthetic premium that does not correlate with standard square footage. Think coastal California, certain historic districts in older cities, or neighborhoods with strong brand recognition. Zillow's hedonic model captures square footage, bedroom count, and lot size well. It captures less well the intangible factors that drive price in these markets. A home on a coveted cul-de-sac in a specific San Francisco neighborhood may command a 30% premium over a mechanically identical home two blocks away. The Zestimate will not reflect that unless there have been enough recent sales in the exact same micro-location to signal the premium. Mobile homes, co-ops, and condominiums in certain configurations also fall outside the model's comfort zone. The data structure for these property types is inconsistent across counties. Some jurisdictions report cooperative shares rather than unit-specific data. Some report mobile homes as personal property rather than real estate. The model struggles with input inconsistency, and the resulting estimates can be wildly off without any visible warning indicator.
Bottom Line
Zillow's accuracy claims are real but presented in a way that obscures the underlying distribution. The national median error rate is useful. The within-5%-of-Zestimate percentage is not. Understanding the difference matters if you are making financial decisions based on the number. Using Zillow data as a rough filter is fine. Using it as a pricing authority is not. The model has known limitations in off-market properties, rapidly changing markets, and non-standard housing types. Knowing where those limits are keeps you from making costly mistakes.
