How much is my house worth in the United States?
You can get a free estimate for your address here, and you can get one from five other places in about the same time. What almost nobody gives you is the second half: what the number is made of, where it breaks, and how far out it tends to be on exactly the kind of house you own. That is what this page is for.
What our United States number actually is
What we return
A value range, low, middle and high, for the address you give us, plus a count of the comparable properties behind it. It comes from an automated valuation model that we rent. It is free, it needs no account, and the range appears on screen.
What it is not
It is not a reading of recorded sales. The United States has no free national register of what houses actually sold for, so unlike France, where we read the recorded deeds, the American figure is a model output. It also carries no floor area, so we publish no price per square foot.
We say this in the first paragraph rather than in a footnote because the distinction changes what the number is worth to you. A recorded sale is a fact about a house. A model output is an opinion about a house, produced by a machine that has never seen it, and the quality of that opinion depends entirely on how ordinary your house is.
First, we can serve about forty United States valuations a month, because that is what the free allowance of the model we rent permits once we keep ten back for testing. It resets on the first. When it runs out the page says so and spends nothing on your address.
Second, the range is free and immediate. The exact figure, the split and the comparable properties are sent by email if you ask for them. We say so here rather than discovering it at the end of a form.
The proof problem, measured one site at a time
Between 23/07 and 30/07/2026 we opened the five largest American home value tools and read them the way an owner would, then wrote down what each one publishes about its own accuracy. The pattern was consistent enough to be the reason this page exists.
Redfin states 1.87 per cent median error for homes listed for sale and 7.28 per cent for homes that are not. Read 27/07/2026. That is a real disclosure and it deserves credit.
Zillow, Chase, Bank of America and realtor.com carry no accuracy figure, no sample size and no method page. realtor.com runs three rented third-party models rather than one of its own and puts its homeowner dashboard behind a login.
On 29/07/2026 the Bank of America tool returned 2,107,772 dollars for a Seattle address while its own table of six nearby sales on the same screen averaged 1,476,710. A gap of 43 per cent, with no explanation offered anywhere on the page.
A number you cannot interrogate is not information. It is a lead form with a figure printed on it.
Where an automated valuation breaks, and why it breaks worst at the top
Three things move a price more than anything a model reads from an address, and a model reads none of them.
Two houses on the same street, one with water in front of it, do not trade at the same price. No address-based model prices a view.
The gap this opens is not marginal. In Saint-Tropez we measured two four-bedroom villas at about 1,170 and about 9,376 euros a night. Identical bedroom count, factor of eight.
Above a certain price the transactions stop being frequent and stop being ordinary. Below eight genuinely comparable nearby sales we publish nothing at all.
There is a measurement behind this, and we would rather give you its exact shape than a warning in the abstract. Backtested against recorded French sales in July 2026, median models landed 30 to 48 per cent below the real price on premium properties and 52 to 72 per cent below on exceptional ones. That work was done in France, so we do not dress it up as an American result. What it tells you is the direction and the rough size of the error, and the cause is structural rather than national: the houses worth the most are the ones least like their neighbours, which is precisely the condition under which a comparables model fails. This is why, above our Standard tier, we stop publishing a central figure and publish a range with its reasoning instead.
The half of the question they do not answer
Not one of the five American tools we read returns rental income. They value the house and stop. For an owner deciding whether to keep a vacation home, sell it, or let it more seriously, the value alone answers about half of the question.
We count nights actually sold, never an occupancy percentage, and the rental income guide sets out the whole method. Fifty per cent occupancy sounds moderate until you multiply it by 365 and find you have implied 182 nights sold. Very few seasonal markets sell anything like that. In Saint-Tropez a villa sells around 45 nights at the high rate and perhaps 10 more outside the peak, which is about 92,000 euros gross rather than the 193,000 the occupancy method produces. The distance between those two figures is usually the whole of the owner's decision.
Our own American markets are benchmarked the same way: Aspen, Vail, The Hamptons, Miami, Palm Beach, Maui and Big Island.
The United States is one of the few markets where we hold both halves for the same address: a value from the model and an income figure from our own benchmarks. That makes a gross yield computable from two numbers we actually have rather than from one and a guess. Being exact about the label: what we can compute is gross rental yield, income over value. It is not a capitalization rate, which is net operating income over value, because we do not hold your operating costs. We will not print the American word over the British arithmetic and hope nobody checks.
What would change our answer
A free national record of United States sale prices joined to floor area at address level. It does not exist today. Deed records are held county by county, on different terms, and the national picture is sold rather than published. Until that changes, the American number stays a model output and we will keep describing it as one.
The same missing piece, in a different form, is why we publish no house valuation in the United Kingdom at all: there the register of prices paid is free and complete, and it is the floor area that is absent.
Questions owners actually ask us
Where does your United States number come from?
From an automated valuation model, RentCast, which we rent. That is an important difference from France, where we read the register of recorded sales directly. In the United States there is no free national register of what houses actually sold for, so the honest description of our American figure is a model output, not a record of deeds.
Why does the estimate stop working some months?
Because we buy nothing. The free allowance of the model we rent is fifty valuations a month. We stop public requests at forty so that ten always remain for our own testing, and the counter resets on the first of the month. If you hit that wall the page says so and spends nothing on your address. We would rather tell you the allowance ran out than quietly serve you a worse number.
Do you publish a price per square foot for a US address?
No. The model returns a value range and a count of comparable properties, and no floor area, so any price per square foot we printed would be arithmetic on a number we do not hold. For the same reason we publish no house valuation at all in the United Kingdom, where the register carries the price paid but not the floor area.
Is the estimate reliable for an expensive house?
Less so, and this is the part nobody puts on the front page. Backtested against recorded French sales in July 2026, median models came out 30 to 48 per cent below the real price on premium properties and 52 to 72 per cent below on exceptional ones. That measurement is French, so we do not present it as an American result, but the cause is structural and not national: the houses worth the most are the ones least like their neighbours. Above our Standard tier we publish no central figure at all.
What can you tell me that Zillow and Redfin cannot?
What the house earns. None of the large United States valuation tools returns rental income, checked one by one between 23/07 and 30/07/2026. We do, for a vacation home or a short-term rental, and we count nights actually sold rather than an occupancy percentage. For a US address we return both a value and an income figure, which is the one place a gross yield can be computed from two numbers we hold rather than from one and a guess.
If the house is also let, the rental estimator works in the United States and everywhere else. It is free, it takes three questions, and it counts nights sold. Start with the rental income calculator, compare it with house valuation in France where we read the recorded sales directly, see every market ranked by rental income, or read how we build the numbers and how we are paid.
Start with the number
Three questions. Free. And a report we would be happy to defend in front of you.
Estimate my home →Last updated 08/08/2026.