Where the line is on AI-edited property photos

The Guild of Property Professionals issued a warning to its members on 25 September 2026 covering two things at once: AI-edited listing photographs, and buyers turning up to viewings wearing camera glasses. The photographs half is the one worth reading twice, because most of the coverage of it stopped one sentence too early.

What the Guild actually said

Its compliance officer, Paul Offley, drew the line roughly where you would expect. Removing distracting furniture from a room, or furnishing an empty one so a buyer can read the space: acceptable, provided the image is clearly labelled as AI-enhanced. Removing a structural element such as a wall, or altering the apparent dimensions of a room: not acceptable at all. His summary of it was that "AI is available as a tool to help and support all parties, not as a tool to deceive anybody."

That is a sensible line and it is not a novel one. What makes it worth a longer look is that the coverage largely condensed it to "label it", and labelling is the part that is widely misunderstood.

The label is not a permission slip

Read the Guild's position carefully and the label attaches to one side of the line only. Furnishing an empty room is acceptable and needs a label. Removing a wall is not acceptable, full stop. There is no version of a wall removal that a label rescues.

The Advertising Standards Authority has said much the same thing about AI content generally, and rather more bluntly. Its position is that there is no blanket requirement in the UK to disclose the use of AI in an advert, and that the question to ask instead is whether the audience is likely to be misled if you do not. Then comes the line that matters here: "disclosure alone is very unlikely to mitigate the harm caused by a fundamentally misleading message."

So the sequence is not "edit, then label". It is two separate questions, in order.

First: does the edit change what a buyer would understand the property to be? If it does, no label saves it, and it should not be published.

Second: if the edit does not change that, is the label clear enough that a buyer will actually see it?

Most of the argument in the trade is about the first question. Most of the actual failures are on the second.

There is no AI photo law, and that is rather the point

Agents keep waiting for a rule specifically about AI imagery. There probably is not going to be one, because the existing test already reaches it, and it is broader than a rule about AI would have been.

Under the Digital Markets, Competition and Consumers Act 2024, a misleading action includes giving "an overall presentation that is likely to deceive the consumer, even if the information is factually correct". Read that clause again with a virtually staged photograph in mind. The furniture is not claimed to be real. Every fact in the written description may be accurate. The Act is not asking about the facts. It is asking about the overall presentation.

Separately, a misleading omission is the omission of material information where that omission is likely to cause a consumer to take a transactional decision they would not otherwise have taken. Whether a photograph has been altered is, for a growing number of buyers, exactly that kind of information.

And the Chartered Trading Standards Institute's guidance for property descriptions has said the same thing for years, long before anybody was generating anything: "Do not doctor photos or use extreme lenses that affect the perspective of the image." A tool that stretches a room with a text prompt is doing what a tool that stretched it with glass was already doing.

The enforcement position is the part that moved. Since 6 April 2025 the Competition and Markets Authority has been able to decide that consumer protection law has been breached without taking the business to court, and to impose fines and order redress directly. Propertymark has pointed out that the guidance is ambiguous on several points that land on our sector specifically, and it is, but ambiguity is not slack. The CMA has also declined requests for sector-specific guidance, which means agents are working to the general test and will go on doing so.

The failure mode has already happened

A report in The Times, picked up by Property Industry Eye in June 2026, found listings where rooms had been made to look larger and a chimney breast had been edited out. That is the wall-removal side of the line and needs no further analysis.

The detail worth keeping is what some of those listings had done about disclosure. The AI disclosure statement was present. It was in white text on a white background.

Whoever did that knew a disclosure was expected and treated it as a box to be ticked rather than as something a buyer should read. It is a useful example because it fails on exactly the ground the law already anticipated. On the small-print approach generally, the Trading Standards guidance is flat: "General disclaimers in small print, telling buyers not to rely on details, won't be effective in preventing offences." Invisible text is small print taken to its logical conclusion.

Propertymark's chief executive Nathan Emerson put the requirement plainly at the time: "Any images that have been created or altered using AI must be clearly labelled as AI-generated so consumers understand what they are viewing." Clearly, and understand. Both of those words are doing work, and neither is satisfied by text nobody can see.

So where does the label actually go?

This is the question we would ask an agency about its own website, and it turns out to be a systems question rather than a wording one.

A single line at the foot of the property description does not do it. A buyer scrolling a gallery of fourteen photographs on a phone sees the images, not the description, and a note at the bottom saying some images have been AI-enhanced does not tell them which room they are looking at. If three of the fourteen were enhanced, the buyer has learned that the set is unreliable and nothing more useful than that.

The disclosure belongs with the image, which means per image rather than per listing. In practice that means the enhanced flag has to be a field on the photograph record, not a sentence somebody types into the description, because a sentence in the description cannot follow the image into a lightbox, a thumbnail or a portal feed. A field can. It is the same argument we keep making about material information, for the same reason: a fact held as prose can only be displayed in one place.

Three things follow, and all three are site-build questions rather than compliance-policy questions.

The badge has to be visible on the image itself, in the gallery and in the enlarged view, at a size and contrast a phone user reads without trying.

Alt text does not count. It is not a disclosure. It exists for screen readers and search engines, and a sighted buyer never sees it. An AI label that lives only in an alt attribute is the white-on-white problem in a more respectable font.

And the flag has to survive the export. Your own site is the easy part. It is worth asking whoever supports your feed what happens to a per-image flag when the listing goes out to a portal, because if it is dropped in transit then the version of the listing most buyers actually see carries no disclosure at all.

A short test before publishing

Three questions, in this order, and only the first one can stop you.

Would a buyer standing in that room feel they had been shown something different? If yes, do not publish it, whatever the label says.

If no, can a buyer on a phone see the label on that specific image without looking for it?

And does the label still exist by the time the listing reaches the portal?

Where we land on it

There is nothing wrong with virtual staging and we would not tell an agent to avoid it. An empty room photographs badly, buyers struggle to read scale from bare floorboards, and a clearly labelled furnished version genuinely helps them. Some of that is simply a photography problem that software has got better at solving. That is Offley's point and he is right about it.

What we would push back on is the idea that this is a disclosure problem with a disclosure solution. Half of it is a judgement call about the edit itself, which no amount of labelling touches. The other half is a data problem: whether your photograph records can carry a flag, whether your gallery displays it, and whether your feed keeps it. Those are things a website either does or does not do, and most do not, because until fairly recently nobody needed them to.

The Guild's warning also covered viewings recorded on camera glasses, which is a separate question about consent and data protection, and it deserves a piece of its own rather than a paragraph at the end of this one.

A note on sources. The Guild of Property Professionals warning and Paul Offley's quotations are from guidance it issued to members on 25 September 2026, reported by Estate Agent Today, Property Industry Eye and The Negotiator. The altered images and the invisible disclosure text were reported by Property Industry Eye in June 2026, following a report in The Times; we have not repeated the agency names. Nathan Emerson's quotation is from that reporting. The misleading actions and misleading omissions wording, and the guidance on doctored photographs and small-print disclaimers, are from the Chartered Trading Standards Institute's Business Companion guidance on the Digital Markets, Competition and Consumers Act 2024 and on property descriptions. The Advertising Standards Authority position on AI disclosure is from the Committee of Advertising Practice's published guidance.