Fake sofas, hidden stairs: Europe's AI Act now requires labels on staged property photos

In the European Union, since early August 2026, that picture is supposed to arrive with a label attached. Whether it does is another matter.

What "housefishing" actually means

Estate agents have been editing property photos for as long as the software has existed: wide-angle lenses, brightened windows, a floor plan drawn on a laptop. The newer step is AI virtual staging, where a model furnishes an empty flat — or re-furnishes a cluttered one — in seconds, for the price of a coffee. Done honestly, it is a genuinely useful tool: it helps a buyer see what a bare room could become, and it is not banned anywhere in Europe.

"Housefishing" is the dishonest version. The model does not merely add a sofa; it widens a room, paints over a damp patch, removes a radiator, or — as in the Irish example — hides a staircase behind furniture. The result still looks like a photograph, so the brain files it as evidence rather than advertising. That gap between what a buyer believes they are looking at and what is physically there is the entire problem, and it is a gap that costs people real money at the viewing, after they have already paid for travel, surveys or a deposit.

Article 50: a label, not a ban

The legal mechanism is Article 50 of the EU AI Act. The transparency duties became applicable in early August 2026, after a long stretch of voluntary draft guidelines and pre-enforcement grace periods. The Digital Omnibus Regulation (EU) 2026/1744 adjusted parts of the implementation schedule, but the outcome is clear: since August, this is statutory, not optional.

The logic of Article 50 is narrow and surprisingly practical. It does not stop anyone from generating or altering an image. It obliges them to say so, and it splits that duty into two halves:

• Providers of generative systems must ensure outputs carry machine-readable marking identifying them as artificially generated or manipulated.
• Deployers — the agency, portal or landlord publishing the listing — must disclose clearly that the image has been altered. In practice, that means a visible notice on the page, not a sentence buried in the terms and conditions.

AI-altered and virtually furnished property photos fall into the category of modified content, not into some convenient grey zone. That is why professional bodies have started rewriting their own rules: the Professional Institute of Real Estate Agents in Belgium updated its code of conduct to require members to disclose when AI staging tools were used. An empty room is rarely the issue. A room whose geometry has changed is.

What non-compliance costs

This is the part of the conversation that changes an agency's tone. To make the scale concrete, we put the AI Act's transparency ceiling next to the one most European businesses already know from data protection law:

For a large portal or a listed developer, the percentage matters more than the flat figure: 3% of global turnover dwarfs €15 million. For a small agency with a handful of staff, the ceiling is far beyond a year's revenue — which is why the deterrent does its work long before anyone issues a fine. And money is not the only risk: professional bodies can suspend or revoke an agent's licence, and for most agents, losing the right to practise is the heavier consequence.

Who is legally on the hook

Legally, the agent publishing the listing stays responsible. Software vendors will point at the tool's settings; portals will say they only host what they are handed. But the disclosure duty sits with the person deploying the system — the one who uploads the photo, writes the price and signs the mandate. For a buyer, that has a practical consequence: the party you can complain to is the party whose name is on the listing, not the anonymous software behind it.

Why the label is easy to miss

Here is the uncomfortable part. The provider-side marking is machine-readable metadata, and metadata is precisely what disappears when an image is screenshotted, dropped into a chat, re-uploaded to a portal or forwarded to a partner. The visible disclosure travels; the hidden one usually does not.

Checking whether a photo is AI-generated therefore requires a deliberate extra step — opening the original listing rather than the copy, or running the image through a provenance checker such as the Content Credentials verifier, which reads C2PA data. Very few buyers do that at ten in the evening while scrolling on a phone. That is the enforcement gap in one sentence: the law is binding, but verification is manual, and the incentive to be careless sits with the side that wants the flat to look bigger. The full text of the obligation is set out in Article 50, and the Commission frames the underlying idea as giving people the ability to recognise AI-generated content when they encounter it. Euronews reported on how this plays out specifically in European real estate.

What a buyer can do in a few minutes

• Ask for a phone video walkthrough of the flat in its current state. It is a reasonable request and it settles most arguments.
• Ask for a floor plan with measurements and compare the layout against the photos.
• Run a reverse image search — the same staged living room often turns up in several listings.
• Look at the boring details: skirting boards that bend, plug sockets floating above a sofa, shadows falling in two directions, windows at impossible heights.
• Look for a visible AI notice on the page. If the picture is obviously staged and there is none, ask the agency in writing. A written answer is also a record.

None of this is an argument for rejecting virtual staging outright. A labelled, furnished image of an empty flat is honest and helpful; it sells a possibility. The problem starts when that possibility is drawn over a wall that is not there, and the buyer only finds out at the viewing, with the decision already half made.

For more on how the AI Act's transparency rules are landing across European sectors, see our magazine coverage.