Why the Zestimate is wrong on your block.
Automated valuations are useful at the metro level and frequently off by tens of thousands at the property level — especially in older Maryland neighborhoods where condition and renovation history drive value more than the model can see.
In this guide
1. The published median error rate
Zillow publishes its own accuracy data for the Zestimate. As of 2026, the nationwide median error rate for the Zestimate is approximately 1.74% for on-market homes (homes actively listed for sale) and approximately 7.20% for off-market homes [1].
Two things to know about that 7.2% off-market number [1]:
- "Median" means half the homes are off by more than that. On a $400,000 home in a less-comp-dense area, errors of $40,000–$80,000+ are routine.
- Accuracy varies wildly by location. Where there's high data density (active market, recent sales, similar properties), accuracy is better. Where there's not, accuracy degrades fast.
The on-market 1.74% figure looks better only because Zillow has the benefit of your listing's own price as an input. Pre-listing, you're firmly in the off-market category — where the model is weaker.
| Home value | Median Zestimate error (off-market, 7.20%) |
|---|---|
| $250,000 | ±$18,000 |
| $400,000 | ±$28,800 |
| $600,000 | ±$43,200 |
| $1,000,000 | ±$72,000 |
And again — that's the median. Outliers are larger.
2. Why AVMs miss on historic stock
Automated valuation models (AVMs) work by mass-matching properties against recent sold comps with similar square footage, beds, baths, lot size, and a few other model inputs. They struggle systematically with:
- Renovation level. The model has no idea whether your Federal Hill rowhome has been fully restored or has 1970s wall paneling. Two homes with identical bed/bath/sqft can be $100,000 apart in actual value.
- Condition. Wear, deferred maintenance, system age — invisible to the model.
- Layout quality. A rowhome with an awkward layout and one with an open-concept renovation read the same in the model. They don't sell at the same price.
- Block-level variation. In Baltimore's neighborhoods, the difference between two streets four blocks apart can be material — the model doesn't see the line.
- Off-street parking. A rare amenity in Federal Hill or Canton that adds significant value but isn't in most AVM data sets.
- Recent improvements without permits. Unpermitted renovations don't show up in tax records the model pulls from.
3. What a real comp looks like
A defensible comp set typically has these criteria:
- Recency: sold within the last 6 months (ideally 3). The market moves fast enough that 12-month-old sales are stale.
- Proximity: same neighborhood, ideally within a half-mile. In dense urban neighborhoods, often within a few blocks.
- Similarity: same property type (rowhome, detached, townhouse), within ~20% on square footage, same bed count, same general condition tier.
- Adjusted: condition and amenity adjustments applied. Example: comp had a finished basement, subject doesn't → subtract for the difference. Comp had no off-street parking, subject does → add a premium.
You typically want at least three solid comps, and ideally one or two "pending" or "under contract" comparables to read where the current market is heading.
4. The pricing walk: 4 steps
- Pull recent sold comps from the MLS — not Zillow, not Redfin (both lag and miss). Filter by neighborhood, property type, bed/bath, sqft band, sold within 6 months.
- Walk through the comps virtually — look at the photos. A comp that sold for $50K above what you'd expect probably has a feature you don't. A comp that sold $30K below probably has condition issues.
- Make condition adjustments. Roughly: $5–15K for kitchen-grade differences, $3–8K per bath, $10–30K for a finished basement, $15–30K for off-street parking in Baltimore neighborhoods (your number will vary by block).
- Land on a range, then pick a price. Aim for a list price slightly under your highest justified comp — leaves negotiation room and motivates offers in the first weekend.
Want a real comp set for your specific home?
Send me the address — I'll send back an MLS-sourced CMA with adjustments shown.
Request a CMA →5. When pricing the AVM-number is dangerous
- If the AVM is too high: you list above the real market, sit on the MLS, lose first-weekend momentum (the most important sales window), then have to cut — and cuts read as "something's wrong with the house."
- If the AVM is too low: you list under, get fast offers, accept thinking you did well, and leave $20–40K on the table that a proper comp set would have captured.
Either way, the AVM number is a reference, not a pricing decision. The pricing decision belongs to you, your agent, and the real comp set — verifiable, defensible, and adjusted for your specific home.
Sources
- "How Accurate Is Zillow's Zestimate? Median Error Rate 2026" — Irina Norrell (with Zillow's own published figures) — https://irinanorrell.com/how-accurate-is-zillows-zestimate/ (accessed 2026-06-15)
- "What is a Zestimate?" — Zillow — https://www.zillow.com/zestimate/ (accessed 2026-06-15)
- "Zillow Zestimate Accuracy: How Reliable Is It in 2025?" — Clever Real Estate — https://listwithclever.com/real-estate-blog/how-accurate-is-a-zillow-zestimate-5-things-to-know/ (accessed 2026-06-15)
Automated valuation accuracy varies by property and market. The figures above are nationwide median rates published by Zillow and aggregated third parties; individual properties may be substantially more or less accurate. This guide is general information for Maryland real estate consumers and is not appraisal, financial, or investment advice. For a binding valuation, consult a licensed Maryland appraiser. Evan Kundrat is a Maryland-licensed real estate salesperson (Lic. #5003434) at Keller Williams Flagship of Maryland (Designated Broker: Barry Hess, Lic. #517943). Equal Housing Opportunity.