The listing photo is the first document of a tenancy. It is what someone decides on, sometimes before they have stood in the room, and it is the thing they will hold up in six months when they say the place was not what they were shown.
In 2026 it stopped being only marketing. There is now a statute about altered listing photos, and a major city has proposed one aimed squarely at rentals. Neither of them binds a property manager in Utah or Colorado today. Both of them are worth reading anyway, because of what they ask for.
This is not legal advice.
What the law actually says, and who it actually covers
California’s AB 723 was signed on 10 October 2025 and became Chapter 497. It adds a new section to the Business and Professions Code, and it is in force from 1 January 2026.
Here is the operative line:
A real estate broker or salesperson… who includes a digitally altered image in an advertisement or other promotional material for the sale of real property shall include… a statement disclosing that the image has been altered.
Read that last phrase again. The sale of real property. The bill does not mention rentals, leases or tenancies anywhere. If you manage rental property, AB 723 does not cover your listings, and it does not cover your virtual staging — and a good deal of what is written about it online implies otherwise.
That matters mostly because it tells you where you actually stand. It does not make the rest of the bill uninteresting, because of the second thing it asks for:
… and a link to a publicly accessible internet website, URL, or QR code that includes, and clearly identifies, the original, unaltered image.
The statement has to be “reasonably conspicuous and located on or adjacent to the image”. And the altered image is defined tightly:
an image… that has been altered through the use of photo editing software or artificial intelligence to add, remove, or change elements in the image, including, but not limited to, fixtures, furniture, appliances, flooring, walls, paint color, hardscape, landscape, facade, floor plans
with an explicit carve-out for “lighting, sharpening, white balance, color correction, angle, straightening, cropping, exposure, or other common photo editing adjustments”. Making a room look its best is not what this is about. Putting in a washer-dryer that is not there is.
The rule that is aimed at rentals has not been passed. On 16 July 2026 New York City’s mayor published a Rental Ripoff Report — twenty-three proposals drawn from hearings with renters — and one of them would require landlords, brokers and listing platforms to disclose when rental listing photos or videos were created or altered with AI. It is a proposal. It may change, and it may not happen. But it is the first one written for tenancies rather than sales, and the word the coverage has attached to the problem, housefishing, is going to outlive the proposal either way.
The obligation underneath all of it is a filing problem
Strip out the disclosure language and look at what AB 723 actually requires a California agent to be capable of: producing the original, unaltered image on demand.
That is not a photography rule. It is a records rule, and it is the one almost nobody is set up for.
Think about where the original lives in a normal rental workflow. The photographer shoots the unit and sends you finished JPEGs. The virtual-staging vendor sends you finished JPEGs. The portal re-encodes whatever you upload. At no point in that chain does the file the camera made arrive on your drive, and at no point does anyone ask for it. Six months later the original is on a freelancer’s laptop, if it exists at all.
The day that becomes your problem is not the day a regulator calls. It is the day a departing tenant says the listing photos were faked, or the day a prospective one asks whether the second bedroom is really that size — and the only files you can produce are the same ones they are already doubting. That is the same corner a manager gets backed into when a tenant says the move-out photos are AI, and it has the same answer: the original file, or nothing.
What an altered photo records about itself
Some of this is answerable from the files you already have.
When a photograph is altered with generative AI in an editor that follows the industry convention — Adobe’s generative fill, Google’s Magic Editor, Samsung’s generative edit — the edit is written into the file’s own metadata as a standard IPTC term. Our checker reads that and says so:
The file’s own metadata says generative AI was used to alter or add to this picture. It does not record what was changed, and the rest may be a real photograph… Ask for the file as it was before editing.
It names the tool when the file names it. A picture that was generated outright, rather than edited, reads differently again: the file names its generator. In both cases the finding is the file’s own declaration, not our opinion about the picture. We do not look at a photograph and decide whether a sofa was added. Nobody can do that reliably, ourselves very much included.
Which brings the honest half of this.
A staging vendor whose renderer writes no metadata leaves no trace at all. A file that says nothing is not a file that was honestly made; it is a file we cannot read anything out of.
And a photograph that has been re-encoded loses all of it. Listing portals generally re-encode what you upload, the same way a messaging app does — but we have not measured which portals strip what, and we are not going to assert it about a site we have not tested. Find out about your own in two minutes: download one of your own listing photos back off the portal, and drop it on the checker next to the file your photographer sent you. If the portal copy comes back Can’t Verify and the original does not, you have your answer, and it is an answer about the pipe rather than about the photograph.
Either way, the file worth checking is the one your photographer or your staging vendor sent you.
About that 11%
The New York coverage carries a figure: an analysis of roughly 40,000 listing photographs found nearly 11% “showed signs of AI alteration”.
We cannot corroborate that, and we are not going to repeat it as though we can. We do not know what was counted as a sign, and “signs of AI alteration” is precisely the kind of judgment that our own trade keeps getting wrong in public. We have tried twice to build a classifier that looks at a picture and scores it. The second attempt looked at eighteen ordinary photographs and called seventeen of them fake. Both of those are switched off.
We do ship a third one, and it is worth being exact about where. It lives only in our free browser extension, where it can mark something “likely AI” as you scroll past it. What it produces there is a judgment about a picture, not evidence from a file, it never shows a percentage, and it is deliberately kept out of the checker and out of the dated reports. A report that mixed the two would let a guess ride into a hearing wearing the same clothes as a fact. So the reports we file carry no confidence score of any kind — not on a listing photo, not on anything.
So we will not tell you what percentage of listings are AI. We will tell you what your own files say about themselves, and where they say nothing, we will say that instead.
Four things worth doing, none of which need a law
- Ask for the original alongside the final. One line in your photography and staging agreements does it: “Vendor will deliver, with each finished image, the unmodified original file as produced by the camera, and will retain those originals for twelve months.” It costs the vendor nothing. It is the only step here that has to happen before you need it.
- Check the finals you were sent, once. Drop a set on the checker. If the finals still carry camera data, your pipeline preserves it and you are in good shape. If they already read Can’t Verify before they have been anywhere near a portal, your vendor’s export is stripping them, and that is a five-minute conversation now instead of an argument later.
- Keep the originals where you keep the lease, not in the photographer’s Dropbox. The retention period that matters is the one that outlasts the tenancy and its deposit window.
- If you virtually stage, label it on the listing. No rule requires this of a rental in most of the country today. It costs a sentence, it is true, and it removes the single most common complaint in the housefishing coverage — which is not that the photo was edited, but that nobody said so.
The limits, stated plainly
- AB 723 is one state, and it is about sales. Nothing here tells you what your own state requires, and we are not the people to ask. The New York item is a proposal, not a rule.
- We read declarations, not pictures. If an editor writes the edit into the file, we will show you. If it does not, we have nothing, and a file that carries nothing proves nothing in either direction.
- Metadata can be edited, including the parts that declare an AI edit.
- We have tested files built to the published standard, and real originals from a handful of devices. We have not tested the output of every virtual-staging product on the market, and we will not guess at what they write.
- No confidence score, ever. See above.
ImposterShield reads the evidence inside image files. It runs in your browser and uploads nothing. If a file has been stripped, it says so. That is the product working, not failing.
Check the photos your vendor sent you
Drop them on the page. You will see which ones still carry camera data, which declare an AI edit, and which carry nothing at all. Free, and nothing is uploaded.
Open the checker