Real estate remains a relationship business. A client expects a well-explained price, useful viewings, and clear follow-up through to signing. Artificial intelligence mainly helps with repetitive tasks: preparing an estimate, drafting a listing, or tracking prospects. The Immonot survey cited below shows, however, that the lack of human contact remains individuals' main obstacle to AI.
This article takes stock of AI uses in the real estate sector in 2026, their limits, and a method for testing a first use case.
Key Takeaways
- 33% of French people cite the lack of human contact as the main obstacle to AI in real estate, ahead of privacy concerns (19%).
- 65% of French people have never used AI for a real estate project: use among individuals remains limited.
- 26% of French micro and small businesses say they use AI, a share that has doubled in a year.
- Three tasks lend themselves to automation: estimation, listing writing, and prospect qualification.
- AI can prepare and sort information. The agent keeps the exchanges and decisions that commit the client.
Three levels of adoption not to confuse
Available studies do not all cover the same market. So we need to separate global commercial real estate, French micro and small businesses, and individuals.
In commercial real estate, the JLL study from December 2025, conducted among 1,000 decision-makers and 500 investors across ten countries including France, measures that AI pilot deployment went, globally, from 5% in 2023 to over 88% among investors and 92% among occupiers in 2025. These figures cover global commercial real estate, not French residential agencies.
According to the France Num 2025 Barometer, 26% of French micro and small businesses say they use AI, a share that has doubled in a year. Generative AI use went from 10% to 22%. Bpifrance counts an average of 14 AI use cases per company, 93% of them with a direct impact on productivity.
Individuals remain cautious. The Immonot survey, a leading notarial real estate site, finds that 65% of French people have never used AI for a real estate project. An agency therefore needs to distinguish internal tasks it can automate from exchanges the client wants to keep having with an agent.
Estimation: a support, not a verdict
An estimation tool can cross-reference market data: price per square meter, recent sales, property characteristics, and neighborhood trends. It produces a range to check, not a ready-to-send valuation.
French notaries estimated 921,000 existing-home transactions over the twelve months ending September 2025, up nearly 11% year on year. This volume gives the market context, but it does not change the limits of an automated estimate.
But an automated estimate remains a starting point. It replaces neither the comparative market analysis prepared by an agent nor the formal valuation. It does not see the property's actual condition, the charm of a room, or street noise. AI provides a base, the agent brings judgment. That is the division of labor that works: the machine processes the data, the human interprets it.
In the same Immonot survey, 11% of respondents say they have used an automated estimation solution. The agent therefore keeps a central role: explaining the method, checking the data, and defending the valuation.
Listings: an easy first test to review
Writing listings follows a stable template: property description, strengths, neighborhood, price, and photos. From these elements, AI can prepare a first draft adapted to the target audience. The agent must then check every fact and remove excessive claims.
The value lies mainly in preparing the first draft. The agent keeps control of the final result, corrects errors, and checks the statements that commit the seller or the agency.
This work lends itself to a first test because its form is stable and the result can be reviewed before publication. The same principle of human oversight applies to the document-related uses described in our article on AI in law.
The agent's place in the customer relationship
In the Immonot survey, the top obstacle cited to AI in real estate is the lack of human contact (33%), ahead of distrust of algorithms (22%), fear of errors (20%), and data privacy (19%). 55% of French people believe AI will remain a secondary tool in their real estate dealings.
These answers set a clear limit. AI can prepare or sort information, but the agent must remain available and know the file.
An AI voice agent can handle phone reception, qualify a call, note the request, and forward urgent matters to the agent while they are out showing a property. A lead sorter ranks contacts by how qualified they are and follows up at the right time. The agent, in turn, spends more time on the phone with serious clients and less time chasing missed calls.
Privacy must be framed
An agency handles sensitive data: client identity, financial situation, offer amounts, and documents from a sale agreement. France Num notes that data entered into some free consumer tools can be used to train the model. Professional offerings can provide additional guarantees. The agency should therefore check a service's terms before sending it a file.
In July 2026, the Conseil supérieur du notariat (the body overseeing France's notaries) chose Mistral AI and Scaleway for its artificial intelligence. The Conseil presents this choice as a way to innovate without compromising on data security.
An agency can also look into local AI. If the model runs on infrastructure it controls and no external call is configured, files can stay within its own environment. Our article on local AI and privacy details this architecture and its limits.
Where to start
An agency can start with a bounded test, in this order:
- Choose a stable, repetitive task. Writing listings, responding to estimate requests, sorting incoming leads. A good candidate recurs often and can be checked with a quick review.
- Keep a human check in place. Every AI output goes through review before it is used. Without this step, no one trusts the pipeline and the tool sits unused.
- Settle the data question before scaling up. What can go to a cloud service, and what must stay local. Client files and sensitive documents stay in-house.
- Train a point person at the agency who tests use cases and documents what works. Adoption happens through use, not installation.
An advisory audit helps map tasks, identify which ones lend themselves to automation, and rule out poorly defined projects. It is also the time to check which GDPR and AI Act rules apply to the intended use.
FAQ
Will AI replace real estate agents?
The Immonot survey suggests not counting on it: a third of French people cite the lack of human contact as the main obstacle to AI. The tool can automate certain steps. The agent remains responsible for the estimate, the negotiation, and the support provided.
Can client files be handed over to an AI tool?
Not without oversight. France Num notes that data entered into some free consumer tools can be used to train the model, while professional offerings add guarantees. For client data or sale-agreement documents, check the contract, privacy settings, hosting, and retention period. A local solution can also avoid sending data to an external service if it is configured with no outbound calls.
What is the first use to automate at an agency?
Writing listings is a possible first test, since its template is stable and the result can be reviewed before publication. Start with one specific property type, not the whole agency.
How long does it take to set up automation?
The timeline depends on scope, data, and tools already in place. A targeted test requires, at minimum, choosing the task, preparing test cases, defining human review, and training users. Broader rollout comes only after this test.
Conclusion
The studies cited show strong experimentation in global commercial real estate, still partial adoption among French micro and small businesses, and limited use among individuals. For an agency, choosing a first use depends on three things: how stable the task is, whether the result can be reviewed, and how sensitive the data is.
Start with a repetitive task, keep a human check in place, and verify where the data goes. To examine the uses suited to your agency, contact NexeAI. NexeAI designs no-code AI agents that can be deployed on-premises.
