Some links on this page are affiliate links: if you buy through them we may earn a commission, at no extra cost to you.
Zillow’s Zestimate is an automated estimate of a home’s current market value. Its modern “Neural Zestimate” uses a neural-network model that combines property facts, public records, listings, MLS and brokerage data, prior sales, geography, time and seasonal market patterns. That design helps the estimate adjust as local and national conditions change, but it remains a statistical estimate—not an appraisal, guaranteed sale price or lending decision.
Zillow says Zestimates are refreshed multiple times per week. In its 2025 filing, the company reported a median error rate of 1.8% for listed homes and 7.2% for off-market homes. Those are aggregate company metrics, not a promise that an individual home’s estimate will fall within either percentage.
What Zestimate is—and what it is not
Zillow launched Zestimate in 2006 and describes it as a proprietary automated valuation model. It is intended as a starting point for researching likely market value, not a formal valuation of a specific property. Zillow explicitly says a Zestimate is not an appraisal and should be supplemented by an in-person assessment, a comparative market analysis (CMA) or a professional appraisal when the decision is consequential. (Zillow Help Center)
The 2025 Zillow Group Form 10-K reports median error rates of 1.8% for listed homes and 7.2% for off-market homes. “Median error” describes the middle result in a set of percentage errors; it does not mean every estimate is within that range, nor does it identify the likely error for one address. Listed homes generally provide more current market signals than homes that are not for sale.
#1 Best Overall
Do not confuse an individual Zestimate with the Zillow Home Value Index (ZHVI). A Zestimate estimates one property. ZHVI aggregates Zestimate-derived information to describe typical values across an area and over time.
From random forests to the Neural Zestimate
Zillow’s valuation system has changed substantially since launch. Early production models used collections of random-forest models trained on historical transactions, property facts and listing information. Zillow’s historical account reports about 14% median absolute percentage error in early national backtests; that statistic came from an earlier model, period and evaluation setup and should not be compared directly with current figures. (Zillow’s 20-year history of AI valuation)
| Stage | Approach | What changed |
|---|---|---|
| 2006–2007 | Multiple random-forest models | Built initial estimates from transactions, property facts and listings. |
| Neural Zestimate | Unified neural-network architecture | Learns richer relationships among home attributes, location and time, with fewer separate models to maintain. |
| Current direction | Improved data pipelines and richer, potentially multimodal representations | Ongoing work includes better neighborhood context, explainability, listing text and images. |
Zillow says the neural architecture can generate estimates faster and at lower operating cost than the prior fragmented system. The model also uses quantile regression to represent a range of plausible values rather than implying that one number is certain. (Building the Neural Zestimate)
Do these 3 things before closing this tab:
1Clear out junk files and repair common Windows errors2Fix the driver behind crashes, sound loss and screen glitches3Repair Windows errors before they cause bigger problemsWhat data feeds the estimate
The exact formula is proprietary, but Zillow publicly describes several input groups:
Rank #2
- Property characteristics: square footage, bedrooms, bathrooms, property type, location, lot and structural details where available.
- Public and historical records: county and tax-assessor records, prior sales and other public property information.
- On-market information: listing price, listing description, comparable homes, days on market and related listing data.
- Market and time signals: historical transactions, local conditions, geographic relationships, changing values and seasonal demand.
- Professional and homeowner data: MLS and brokerage feeds plus corrections or updates submitted through Zillow’s property-data tools.
More data is not automatically better. An incorrect square-footage record or stale renovation information can be more damaging than a smaller but accurate dataset. Zillow’s public description of the inputs is available in How is the Zestimate calculated?
How machine learning responds to changing markets
Time-aware valuation
The model learns how similar homes behave at different points in time. That lets it incorporate observed price changes instead of treating a sale from an earlier market as if it occurred under today’s conditions. Mortgage rates, inventory and buyer demand affect an estimate only insofar as those effects appear in the available transaction and listing data; Zillow has not published a simple one-variable adjustment for each factor.
Geographic relationships beyond the nearest comparable
Zillow says the Neural Zestimate can learn across county borders, longer historical periods and geographic boundaries. If a neighborhood has few recent sales, the system may use a broader area—potentially at county scale—to infer how similar homes are moving. This is more flexible than looking up only the nearest recent sale, but it can also import differences between neighborhoods that a local expert would recognize.
Seasonality and turning points
Seasonal demand can make the same type of home sell differently in spring, winter or another part of the cycle. Zillow’s methodology work says the neural model captured stronger seasonal patterns and responded more clearly when prices changed direction. That is a response to observed signals, not a claim that Zestimate forecasts a crash or knows future prices.
What happened during the volatile 2022 market
In a 2023 methodology revision, Zillow reported that the neural model was nearly 20% more accurate than the prior model at predicting 2022 sale prices, a period that included a sharp market shift. Neural-backed ZHVI also showed a larger decline from the July 2022 peak to January 2023 than the previous version, indicating faster recognition of the downturn. Zillow reported that one-month-ahead systematic error for neural ZHVI was close to zero over its January 2020–September 2022 test period. (Zillow Home Value Index Methodology, 2023 Revision)
These are model-level and index-level results. They do not establish that every individual Zestimate tracked every local market accurately, particularly where data was thin or a property was unusual.
How often a Zestimate changes
Zillow says estimates for all homes are generally refreshed multiple times per week. The normal schedule can be interrupted during algorithm changes or when new analytical features are introduced. A change may follow a new listing or sale, an updated public record, additional comparable transactions, a seasonal adjustment or a broader model update. Zillow says the estimate is automated and cannot be manually changed for one specific property.
Crashes, No Sound, or Screen Glitches?
Random freezes, missing sound and display glitches usually trace back to one bad driver. Find and replace yours safely.Free scan · under a minuteWindows Errors? Fix Them Before They Spread
Repair common Windows errors and clear accumulated junk for a smoother, more stable PC - no reinstall needed.Free scan · no reinstallCorrecting a home fact does not guarantee a new value. The model evaluates whether the correction has a measurable market effect: an omitted bedroom or materially wrong square footage may matter, while a detail such as paint color may not. Public records can also lag behind an addition, renovation or other physical change.
How Zillow measures accuracy
Backtesting and holdout sales
Zillow trains a model on historical information and tests predictions against sales that occurred after the relevant training period. Backtesting recreates that process using older data. Median absolute percentage error (MdAPE) measures the typical percentage distance between an estimate and the observed sale price, without allowing unusually large errors to cancel out unusually small ones. Listed and off-market homes are reported separately because their information environments differ. (Building the Neural Zestimate)
| Statistic | Scope and qualification |
|---|---|
| 1.8% median error | Zillow-reported 2025 figure for listed homes; an aggregate median, not an individual guarantee. (2025 Form 10-K) |
| 7.2% median error | Zillow-reported 2025 figure for off-market homes; an aggregate median, not an individual guarantee. (2025 Form 10-K) |
| Nearly 20% improvement | Zillow’s comparison of the neural model with the prior model for predicting 2022 sale prices; a different test context from the 2025 filing figures. (Zillow Research) |
Why a Zestimate can be unavailable or wrong
- There may be too few recent sales in the property’s area.
- County, tax or listing records may be incomplete or incorrect.
- The home may have unusual architecture, condition, amenities, views, water access or development potential.
- Major renovations or additions may not be recorded.
- Rural, luxury, newly built or thinly traded properties often have fewer useful comparisons.
- A rapidly changing neighborhood can move faster than the data refresh cycle.
- Structured data cannot fully encode condition, buyer urgency, seller motivation, negotiation strategy or other human factors.
Zillow may suppress a Zestimate when available information does not meet its internal accuracy standards. An absent estimate can therefore be a confidence-control decision rather than a website malfunction. (Zillow Help Center)
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.What to do when the number looks wrong
If it differs sharply from the asking price
- Check the home’s facts on Zillow for incorrect bedrooms, bathrooms, square footage, lot size or property type.
- Review recent closed sales, not only active listings, and compare homes with similar condition and features.
- Ask a local agent for a CMA that accounts for renovations, micro-location and buyer behavior.
- Use a licensed appraisal when a lender, estate, tax authority or legal process requires a formal valuation.
A list price can reflect urgency, strategy or an unrecorded feature, so a difference from Zestimate is not by itself evidence that either number is wrong.
Recommended Free Tools
If the Zestimate changes suddenly
Look for a new sale or listing, a public-record correction, new comparable transactions, seasonal movement or a broader algorithm update. Zillow does not describe manual one-off edits to individual estimates.
If you want to correct the property record
- Claim the home on Zillow.
- Review the displayed facts.
- Correct inaccurate or incomplete fields and add available details such as architectural style, roof type, heat source and building amenities.
- Report material additions or renovations to the appropriate public-record authority where applicable.
- Allow time for Zillow’s systems to process the information.
Corrections are intended to improve accuracy, not to increase the estimate automatically.
How to use Zestimate responsibly
For ordinary research, Zestimate is most useful as a quick, data-informed starting point: pair it with closed-sale evidence, neighborhood context and a review of the underlying facts. For a financially material decision, escalate to the method that fits the decision—a CMA for pricing strategy or a licensed appraisal when an independent formal valuation is required.
There is also an analytical caveat. An external 2023 academic paper discusses how algorithmic price estimates could create feedback loops if buyers and sellers anchor on them. That paper is not a Zillow corporate study and does not prove that Zestimate causes prices to move; it is a reason to treat any widely visible automated estimate as one influence among many. (Academic discussion of housing-market feedback loops)
The Bottom Line
Machine learning has made Zestimate more scalable and better able to incorporate geography, seasonality and market turning points. It improves a model of market behavior; it does not remove property-level uncertainty. Treat the number as an informed starting estimate, then verify the facts and use a local CMA or licensed appraisal when the stakes require it.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
