Real estate valuation: can we really talk about a reliable price?
Between artificial intelligence, statistical models and professional experience, a fundamental question remains: can a real estate appraisal really determine the price at which a property will be sold?
Artificial intelligence can produce a detailed property valuation in seconds, accompanied by a compelling rationale. But does a figure presented with confidence necessarily become a reliable price?
The issue goes far beyond the performance of digital tools. It touches on the very nature of real estate valuation, market knowledge, the experience of professionals and, above all, the uncertainty inherent in any transaction.
A real estate agent with thirty years of experience can provide a well-reasoned valuation. An algorithmic model can analyze thousands of real estate data points. Neither can guarantee that the property will actually sell at the asking price.
The essential principle: a property valuation is a reasoned assessment of the probable value of a property at a given date and under specific market conditions. It is neither a guarantee of sale, nor a promise of price, nor a mathematical certainty.
Estimating the value of a property is not the same as knowing its future selling price
In everyday language, the expressions "estimate," "market value," and "selling price" are often used as if they referred to a single reality. However, they refer to different concepts.
The valuation
This is an assessment based on market references, the characteristics of the property, its environment, and the economic context. It may take the form of an indicative amount or a range.
The presentation price
This is the price at which the owner decides to offer their property on the market. It may incorporate a negotiation strategy, personal expectations, or marketing objectives.
The price finally accepted
It results from the meeting between a seller and a buyer, and their subsequent agreement on the financial terms of the transaction. It may differ from the initially asked price.
Real estate appraisal
A formal appraisal is based on a defined mission, methodology, and documentary framework. It can be used in various situations, including those related to property, legal matters, or taxation. It remains a reasoned valuation opinion, not a definitive prediction of the future sale price.
"An estimate can be rigorous without being a certainty. Its credibility rests on its foundations, not on the confidence with which the figure is announced."
This distinction is fundamental. A professional can perfectly well produce an estimate consistent with the available data and find several months later that the sale has concluded at a different price.
Simply noting this discrepancy is not enough to demonstrate that the initial analysis was irrelevant. Marketing conditions, demand, negotiation, and the specific characteristics of the transaction must also be examined.
Generative AI and real estate valuation: what the Yanport comparison reveals
On October 9, 2026, the Journal de l'Agence revisits a study published on August 3 by Yanport, a company specializing in real estate data and valuation solutions.
The comparison pits two language models, Gemini and Claude, against an automated property valuation model (AVM) developed by Yanport. The study covers a sample of 1,071 properties.
According to Yanport, the systems received the same information on the housing units in order to produce estimates and, in particular, price ranges.
Data provided by Yanport for its comparative protocol. These percentages do not represent a systematic error applicable to all real estate valuations.
The result is of statistical interest: under the conditions of this test, the specialized model shows a median absolute error lower than that of the generalist models evaluated.
Yanport also notes differences in the dispersion of results and the stability of responses. The Agency's Journal reports, in particular, an experiment where several identical requests to a model produced different estimates for the same dwelling.
But what exactly is the reference being used?
An estimation error is always calculated relative to a reference value. To fully interpret the published percentages, it is therefore necessary to understand the composition of the sample, the nature of the reference prices, and the processing methods applied. Yanport specifically indicates that it first estimates an asking price and then deducts negotiation margins. The public presentation alone is insufficient to establish that each comparison is based directly on a sale price authenticated by a notarial deed.
The fact that the study was produced by the publisher of the specialized model being evaluated also calls for careful methodological analysis. This does not invalidate the results, but rather suggests interpreting them within the limitations of the presented protocol.
A 15% error: 15% compared to what?
Let us imagine, for purely educational purposes, a house offered on the market and ultimately sold for 800,000 euros.
One tool had estimated it at 920,000 euros. Taking the actual sale price as a reference and calculating the difference on that basis, the absolute error would be 15%.
Fictitious example: actual selling price: €800,000; prior estimate: €920,000; difference: €120,000, or 15% of the selling price.
The calculation is perfectly possible. It allows you to compare an estimate with an actual observed price.
But one question remains: could the professional or the tool that formulated the initial estimate have known in advance the behavior of future buyers, the sales times, the negotiations or a possible change in the economic situation?
The answer is no.
This is why it is necessary to distinguish between the statistical accuracy measured after the fact and the certainty of the amount announced before the transaction.
Notarial databases provide a reference point, not a universal formula
In France, transaction data from notarial deeds are essential references. The Land Value Inquiry (DVF) database, published by the administration, allows users to consult transfers of ownership that have taken place across a large part of the country.
This data is valuable for analyzing a market. However, it does not replace the need to assess the differences between properties: condition, quality, configuration, precise location, amenities, and transaction conditions.
The tax authorities themselves point out that the valuation of a property must take into account its own characteristics, beyond the observation of comparable sales.
Do thirty years of real estate experience provide access to a fair price?
An experienced real estate agent accumulates knowledge that cannot be reduced to a transaction log. They have seen properties sell quickly, others remain on the market for several months, buyers back out, and owners accept unexpected offers.
It sometimes has very localized peculiarities: a street more sought after than another, an exceptional exposure, a nuisance that is difficult to perceive from a distance, the quality of an environment or the expectations of a specific clientele.
This professional knowledge is a major resource for real estate appraisal.
But experience also has its limits. A series of past transactions does not guarantee that a future operation will reproduce the same results. The market evolves, expectations change, and each buyer retains their own motivations.
"Experience helps to better argue a probable value. It does not, however, guarantee the price a buyer will accept tomorrow."
A quality professional appraisal must therefore be able to be explained: references used, elements of comparison, strengths of the property, possible reservations and market assumptions.
It also benefits from being re-examined when marketing conditions change or when market feedback becomes sufficiently significant.
In the luxury sector, value is not simply measured by the price per square meter
The limitations of automatic estimation become particularly noticeable when dealing with rare properties: mansions, castles, vineyards, equestrian properties, waterfront villas or historic residences.
In these segments, the number of truly comparable transactions may be reduced. Architectural quality, the history of the place, outbuildings, land, views, privacy, or the condition of amenities can strongly influence the perceived value.
Two properties with the same surface area in the same region can have very different levels of desirability.
A restored home, respecting its original architecture, will not necessarily be valued as highly as a property requiring extensive renovations. A villa with direct access to the sea is not automatically comparable to a house located a few hundred meters away.
As goods become more unique, qualitative analysis becomes increasingly important. It should complement statistical data, without entirely replacing it.
Scarcity can create value, but it doesn't guarantee a sale
A rare property can generate particularly high interest from a buyer whose criteria perfectly match the property. It can also require a significant marketing period if the potential clientele is small.
The exceptional nature of a property does not, therefore, in itself, allow us to determine the amount at which it will be sold or the time required to find a buyer.
In the luxury market, the valuation therefore benefits from being presented as a detailed reasoning, accompanied by a marketing strategy adapted to the positioning of the property.
AI and professionals: two complementary contributions
Systematically opposing artificial intelligence to human expertise is too restrictive a way of approaching the evolutions of the real estate profession.
A specialized valuation model can leverage vast datasets. Generative AI can help structure an analysis, synthesize observations, or explain hypotheses. The professional can then compare these results with the realities on the ground and the specific characteristics of the property.
The data
Study transactions, analyze market trends and match properties with comparable characteristics.
Artificial intelligence
Assist in reading information, highlight hypotheses and contribute to a structured presentation of the analysis.
The professional
Examine the property, assess local features, explain the limitations of the valuation opinion and support the owner's decision.
However, a distinction must be made between general language models, which produce text in particular, and automatic models specifically developed for real estate valuation.
Their technical objectives, available data, and validation methods can be very different. A serious comparison must therefore precisely identify what has been tested.
The right question is not simply: "Who gives the closest figure?" It is also: "How was the figure constructed, what references is it based on, and what uncertainties remain?"
How to intelligently interpret a property valuation?
When a homeowner receives an estimate, it's tempting to focus primarily on the quoted amount. However, several factors deserve just as much attention as the final figure.
Request the references used
An estimate must be able to rely on identifiable elements: relevant transactions, market conditions and comparison criteria.
Understanding the property's specific characteristics
The architectural qualities, the technical condition, the work carried out and the environment must be explicitly considered.
Distinguishing between probable value and pricing strategy
The price chosen to present the property on the market may correspond to a commercial choice different from the central valuation.
Accepting a margin of uncertainty
A reasoned range, accompanied by understandable assumptions, can be more useful than an isolated amount presented with excessive precision.
Reassess the situation throughout the marketing process
The interest of buyers, the nature of visits, the offers received and the evolution of the market provide new information that needs to be analyzed.
Reliability must not become a misleading promise of certainty
It is legitimate to seek more effective valuation tools and to measure their results on sets of transactions. This work contributes to improving valuation methods.
But the concept of reliability must be precisely defined.
A median statistical deviation can measure the relative performance of a model on a given sample. It is not sufficient to guarantee the relevance of a particular estimate, especially when the asset has rare characteristics or is poorly represented in the data.
Similarly, extensive professional experience is a considerable asset without transforming a value opinion into a certain prediction.
The true value of an estimate lies in its method
A quality real estate appraisal must be explainable, contextualized, and honest about its limitations.
It can draw on statistical data, professional experience, digital tools, or a combination of these resources.
What matters is not presenting an amount as indisputable, but helping the owner make an informed decision.
In luxury real estate, where the uniqueness of properties often reduces the comparability of transactions, this requirement for rigor and transparency appears all the more important.
The references for this analysis
This report offers an independent editorial analysis. It distinguishes between the statistical results published by the study's authors, the available administrative references, and the resulting professional reflections.
Study published on August 3, 2026. Comparison of 1,071 goods, error indicators and protocol elements. See the original study.
Article published on October 9, 2026, presenting Yanport's results and the limitations of the responses produced by generalist models. Read the article.
Presentation of transaction references and available valuation resources. See the official document.
Information on searching for comparable transactions and taking into account the specific characteristics of a property. Consult the administrative reference.
Aerial view of the Château de Vaux-le-Vicomte, Carsten Steger, photograph taken on June 19, 2025, original resolution 5100 × 2800 pixels, Creative Commons CC BY-SA 4.0 license. The presentation framing may differ from the original. Photograph details and license.
Methodological note: The reported performance figures are from a study published by Yanport, the publisher of the specialized model being compared. They characterize the conditions of the test presented and do not represent universal error rates for all artificial intelligence systems or for all categories of real estate. The illustrative numerical examples do not correspond to identified transactions.
Discover character and prestigious real estate
Propriétés De Charme is a real estate portal dedicated to character properties, prestigious residences and luxury properties in France and internationally.