When scrolling online for a place to live, you might come across a bright-colored house for sale listed online with fall foliage and folksy furniture.
But then you notice that a few things look off: deformed electrical outlets or furniture that looks too big for the room. A floor lamp is coming out of a table in one picture and does not appear in the next, or the same piece of artwork is found in every room.
Real estate images like this are, most likely, made with AI. Listers are increasingly using digital platforms to spruce up living spaces for sale. Several online forums have called out AI real estate photos for being misleading and untrustworthy.
A user on an Irish housing subreddit posted side-by-side photos of a house with furniture, plants, and decorations and the real-life empty house. But in the AI version, the couch is blocking the stairs and the living room wall is bigger.
“AI now being used by estate agents to sell houses. Is this even legal?” the user wrote on 22 September.
Real estate agents have long used staging to help their clients visualize the potential of a space. Adding a sofa or bright flowers can make an empty apartment seem more appealing and livable.
But staging a place takes a lot of time and money. AI can speed up the process and cost much less — or almost nothing at all.
Some argue that virtually staging with AI is similar to normal staging — it just uses a new technology, says Steven Lee from the Professional Institute of Real Estate Agents.
“Agents already use home staging to make a place look nicer when you visit it,” Lee said. “So this is basically easier and cheaper to do, of course, on a computer.”
Yet changing the appearance of a home or apartment with AI could possibly trick some into believing a room is much bigger or a kitchen has more modern appliances. This is a particular concern for those relocating to a new city who may not be able to see a listing in person.
Sometimes it is hard to tell when an image is real or altered. The elderly are especially susceptible to AI deception, and some do not know what a deepfake is.
Claudio Teixeira is the head of digital policy at the European Consumer Organization. He said AI deceit in online marketplaces, such as the housing industry, are not talked about enough.
“Not everyone has the capability to see at first glance that this is indeed not real,” Teixeira said. “So this content needs to have a clear warning that it is AI generated.”
New EU regulation might help
New EU AI transparency rules may prevent the practice known as “housefishing.”
Article 50 of the EU AI Act, which became applicable in early August, requires AI-generated or modified content to be labeled.
The EU classifies an image of an apartment that is “furnished using AI” as partially AI-modified and requires the photos to have an AI transparency icon.
“Article 50 is more or less an expression of this protection of fundamental rights,” said CMS senior law associate David Rappenglück, who monitors applications of the act. “The message is basically quite easy: if you, as a customer or a bidder, are confronted with a picture which might be AI generated or manipulated, you should be informed about it.”
The providers of AI content, a company that develops AI systems such as Open AI or Anthropic, must label images in a machine-readable format. A deployer, someone who uses an AI system to create or enhance an image, must also include a visible label on the image.
But if a deployer does not label an image, it can be difficult for a regular person to check if AI has been involved at the provider level.
In order to check, they must plug the machine-readable code into another website which verifies it, said Akvilė Šikšniūtė, who works on the digital policy team in the European Consumer Organization. The extra step is often not very clear.
“I think we’re still quite far away to the point where the consumers feel that they have places to turn to if they feel like they’re misled or the content that they’re seeing is not accurate,” said Šikšniūtė.
At what point should AI be labeled?
Using AI to add minor details such as changing the lighting or color of a wall is very different from expanding a room or creating an entirely new image. But the AI Act does not distinguish between partial manipulation or full generation, according to Rappenglück.
People may not know if an image is just slightly modified or entirely changed. So how do we determine what should be labeled? The details remain uncertain.
“It isn’t clear because, for example, very subtle changes like changing a background wall could be considered as a minor change that AI was used to change that. Therefore, it perhaps would not even be marked,” said Šikšniūtė.
“I feel like this is where perhaps the misconception comes from for consumers slightly, because we expect now from all the communications we see that all the content will be labeled, and that’s not the case,” she said.
Practical enforcement remains unclear
The punishment for not labeling content is steep. Those who do not follow the act could be fined up to €15 million or 3% of their worldwide turnover.
“We as lawyers would really recommend taking this quite seriously,” Rappenglück said.
“There’s quite a hard enforcement mechanism and it’s also enforced by the national authorities.”
Because the act only came into force last month, there are still a lot of unknowns about how it will work, and enforcement remains difficult in practice.
According to Rappenglück, images can be screenshotted and reported to the national government.
But because real estate moves so quickly, listings may only be up for a few days. That makes it harder for individual cases to be monitored.
“How do you actually ensure accountability of this sector to make sure that these practices do not gain hold? Right now, the apparatus is clear that this is not happening,” Teixeira said.
The industry itself is trying to enforce AI transparency. The Professional Institute of Real Estate Agents in Belgium is currently updating their code of ethics to align with the rules of Article 50.
A real estate agent remains responsible for what they publish. They can use AI but only if they disclose it, said Lee.
“They have to follow the code of ethics to be able to exercise their activities. And if they don’t, I’m afraid they can get into quite a bit of trouble,” Lee said.
It is case-by-case, but that trouble could include suspension or revocation of their license.
Next steps
AI deception is happening beyond housing, in many different sectors. Teixeira hopes that AI enforcement becomes stronger and that consumers are not held to blame for falling for fraud.
“It must be a whole system at work, in order not just to ensure the consumers have more information to make our decisions, but also if everything fails, to ensure that rogue actors who are going on a rampage to make content more engaging by changing it willfully into something which they know is not true,” Teixeira said.