AVMs, appraisals, and asking prices: which number to trust

Ask what your home is worth and you'll get several confident answers that don't agree. The listing site says one number. A different site says another, $40,000 away. Your neighbor's house is listed at a price that would make yours worth more than both. And the appraisal from your refinance last year says something else entirely.

None of these numbers is lying. They're just different kinds of evidence, with very different reliability — and treating them as interchangeable is how homeowners end up over-borrowing, under-pricing, or planning around a fiction. Here's the hierarchy.

The four kinds of numbers

Closed sales are what buyers actually paid for comparable homes, recently, near you. This is the bedrock — the only category where a real buyer and real seller agreed on a price with money behind it. Every other number is, one way or another, an attempt to predict this one.

Appraisals are a licensed professional's documented opinion, built from those closed sales and adjusted for your home's specifics — after physically seeing it. Banks lend hundreds of thousands of dollars against this number, which tells you how the financial system ranks it.

Listing prices are hopes with a sign in the yard. A list price is a seller's opening position — often inflated by optimism, occasionally lowballed for strategy. Homes routinely close meaningfully above or below asking. Your neighbor's ambitious listing is a data point about your neighbor, not your home, until it closes.

Automated estimates (AVMs) — the single numbers on listing sites — are statistical models run over public records: square footage, beds and baths, tax data, and nearby sale prices. Useful, instant, free — and blind in specific, predictable ways.

Where AVMs go wrong — predictably

The major AVMs are reasonably accurate on average across millions of homes. The catch is that you don't own the average — you own one specific house, and the model has never seen it. AVMs are systematically weakest when:

• Condition diverges from the neighborhood. The model can't see your new roof or your tired kitchen. It largely assumes you match the homes around you — flattering for neglected homes, unfair to maintained ones.
• The home is unusual. Odd lots, additions, unusual layouts, mixed-use, acreage: anything without close statistical siblings gets guessed at.
• The market is thin or fast. Few recent sales nearby — rural areas, unique segments — leave the model extrapolating. Rapidly moving markets leave it lagging.
• Public records are stale. That permitted addition from 2019 might not be in the data the model eats.

This is also why the same house gets numbers $40,000 apart on two sites: different models, different data, both plausible, neither inspected anything. The spread between AVMs is itself useful information — it's roughly the width of honest uncertainty about your home.

What an appraiser does that a model can't

An appraisal isn't magic — it's mostly the same comparable-sales logic an AVM uses. The difference is judgment applied to specifics: the appraiser walks the home, selects the genuinely comparable sales rather than the merely nearby ones, and adjusts line by line for what your home actually has — condition, updates, defects, the things that actually determine a home's value.

Appraisals have their own limits: they're opinions, they can vary between appraisers, and in fast markets they anchor to slightly stale sales. But for a single point-in-time number on a specific home, a recent appraisal outranks every screen estimate you'll find.

Ranges are honest; points are marketing

Here's the deeper problem with every single-number estimate, automated or not: it dresses uncertainty up as precision.

Your home's value is genuinely a range — what fifty plausible buyers would pay this season spreads wide. A model that says "$487,300" hasn't eliminated that spread; it's hidden it behind false precision. A range with a reason — "likely $460,000 to $500,000, supported by three closed sales within half a mile, with the spread driven by kitchen condition" — tells you more than any point estimate, because it tells you what the number rests on and what could move it.

A practical habit: whenever you're handed one number, ask what the range around it is and what evidence supports it. If there's no answer to either, you've been handed marketing.

Matching the number to the decision

Different decisions deserve different evidence:

• Casual tracking. AVMs are fine. Watch the trend, ignore month-to-month wobble, and remember the model can't see your house.
• Renovation planning. Use closed sales of homes that already have the feature you're considering — that gap, not an AVM delta, is what a project can realistically return.
• Borrowing against the home. The lender will order an appraisal anyway; don't size plans off a screen estimate the appraisal may not support.
• Selling. Closed comparable sales plus a clear-eyed read of your home's condition — ideally with documentation of what's been done — beat both the AVM and your neighbor's asking price.
• Estate, divorce, taxes. Get the appraisal. These decisions punish imprecision.

The pattern across all five: the more money rides on the number, the further up the evidence hierarchy you should go — from model, to professional opinion, to what real buyers actually paid. The number on the listing site isn't your home's value. It's the start of the question.