Why not use listing prices as a reference?

Because a listing measures what a seller hopes for, not what a buyer paid. In a turning market the gap between the two can exceed 10% for months. DVF sales are recorded by notaries: they describe completed transactions, at a known price and a known date.

How often is DVF data updated?

The tax administration publishes twice a year, which leaves several months between a sale and its publication. The dossier always dates the sales it cites, so you know which period the comparison covers.

Does DVF cover the whole of France?

No. Alsace, Moselle and Mayotte fall under the livre foncier and do not appear in the DVF file. In those areas the dossier says so and leans on the other sources rather than producing a median that does not exist.

Why the median rather than the mean?

Because a single atypical sale — an exceptional property, a sale between relatives, a mis-recorded floor area — moves a mean and does not move a median. On samples of a few dozen sales, that is the difference between a benchmark and an artefact.

What happens when a figure is missing?

The dossier writes it down. A missing piece of information is neither a defect of the property nor a reason to lower a score: it becomes a question to put to the seller or a document to request. A listing never says everything, and that is not a fault of the property.

Can the AI get things wrong?

It can mis-rank an attention point or phrase a nuance badly, like any writer. That is precisely why it produces no figures: the price gap, the median and the range remain verifiable calculations, and every fact it puts forward cites its source so you can check it.

Methodology

How we analyse a listing

The figures come from public registers, the calculation is deterministic, and the AI only steps in at the end — to read the result, never to produce it.

A property listing is a sales document. It states a price, a floor area and an energy label, but it says nothing about what the neighbours actually paid, what was voted at the last general meeting, or what the certificate filed with the administration really shows. Our job is to set what the ad claims against what the public registers establish.

One principle governs everything else: we do not invent data. Every figure published in a dossier comes from a dated public source, or is presented explicitly as an assumption. When a figure is missing, the dossier says so — it does not replace it with an estimate that would read like a fact.

A less obvious corollary: no competing listing enters our price references. A market of listings measures what sellers hope for, not what buyers pay. Only transactions recorded by notaries count.

Our sources

DVF — actual sales

Published by the French tax administration on data.gouv.fr from notarial deeds: price paid, floor area, property type, address, date. Updated twice a year. The file does not cover Alsace, Moselle or Mayotte, where land registration follows the livre foncier — the dossier flags that rather than passing over it.

ADEME — the energy-certificate register

The certificate actually filed for a home, with its consumption and its date. It is what lets us compare the advertised label with the official one — a discrepancy is both an attention point and a negotiating lever.

Géorisques

The public hazard service, plot by plot: flooding, clay shrink-swell, radon, seismicity, soil pollution, natural-disaster orders. Queried as soon as the listing gives coordinates.

Your documents

The files you upload never leave your dossier and serve only to enrich it. They are the only sources that describe the co-ownership from the inside: works voted, arrears, litigation, real charges.

What we do not use

Prices advertised on other listings, portals' automatic valuations, proprietary "indices" whose calculation is not published. They measure sellers' hopes, not transactions — and you cannot verify them.

The five steps of a dossier

1
We read the ad, without believing it
Price, Carrez floor area, floor level, year built, energy label, charges, property tax, plot coordinates. Everything is extracted and kept as-is: these are the seller's statements, and they are treated as such until verified.
2
We query the public registers
DVF sales around the address, the energy certificate filed with ADEME for this home, the hazards Géorisques records for the plot. Those three queries always distinguish "nothing recorded" from "register unreachable": an unavailable register is a check to run, never an absence of risk.
3
We compute the gap, without the AI
Comparable sales are filtered on the same property type, a close floor area and a tight radius around the address. The dossier takes the median — not the mean, which one atypical sale would distort — then derives the gap in percent and in euros, along with a coherent price range. It is arithmetic on public data.
4
The AI reads, ranks and explains
It arrives once the figures are settled. Its job: to say what the gap means for this particular property, spot what the ad omits, rank the attention points by severity and phrase the questions to ask at the viewing. It produces no market figure and cites the source of every fact it puts forward.
5
Your documents sharpen the dossier
Diagnostics, general-meeting minutes, charge statements, property tax: this is where the decisive attention points live. Every document you upload is read and the dossier is updated — works voted through, a proceeding under way or a real charge figure outweigh anything the ad says.

Frequently asked questions about the method

Check for yourself

Paste a listing: the dossier cites every source, dates every sale and lets you trace each figure back to the original data.

Analyse a listing See an example dossier