Lights, Camera, Bathroom

The bread looked like reptile skin. That was the polite version of the local consensus in July, when a small restaurant in San Francisco put up a menu board illustrated with AI-generated pictures of its own cooking. The coleslaw looked like something recovered from a drain. Locals photographed it, the internet did what the internet does, and then somebody walked up to the building after dark and sprayed the word "seriously" across the awning in red paint. The owners took the sign down and said, reasonably enough, that they would rather people came in and ate the real thing. Every operator in the world shoots the suite at golden hour and almost nobody shoots the bathroom, which is a shame, because the bathroom is where the guest finds out whether any of it was true.

That story is worth sitting with, because it happened in the city most invested in generative AI, to a business that had done nothing worse than take a shortcut its software vendors had actively encouraged. The lesson was not about the technology. The lesson was that in hospitality, a fabricated image reads as a statement about how much you care, and the audience makes that inference instantly and without mercy.

Hospitality is exposed here in a way that almost no other category is. A skincare brand can retouch a photograph and the customer will never be able to run the comparison. A hotel sells a promise that gets redeemed physically, in daylight, by a person who has travelled to check. Every guest is an auditor. Every arrival is a reconciliation between the picture and the room, and the entire economics of the business depend on that reconciliation going well.

The reconciliation is going badly at scale. A forensic image analysis conducted this year examined more than twenty-five thousand hotel photographs across seven European destinations and found that roughly one in five carried at least one indicator associated with AI generation or heavy AI editing. Deception has also outrun the properties themselves. Travellers have driven hundreds of kilometres in Malaysia to board a cable car that never existed, gone looking outside Bangkok for a lotus lake that had been composited against a skyline it does not sit near, searched Peru for a sacred canyon invented in a caption, and planned trips to Tasmanian hot springs that a tour company later admitted had been generated by an outsourced content supplier. Even the review layer has started smoothing things over, with AI summarisation quietly sanding the sharpest guest complaints off properties that had earned them.

Regulation is arriving, though it will not save anyone. Under the EU AI Act, from the second of August this year, materially AI-generated or AI-edited images on booking platforms have to be labelled, and the requirement reaches small independents as well as chains. Labels establish a floor. They do nothing at all for the operator who wants to be chosen.

Which brings us to the actual opportunity, and it is a large one. For thirty years, polish signalled investment. A beautifully lit room photograph told a guest that somebody had spent money, hired a professional, cared about the presentation, and by extension probably cared about the towels. That signal has been devalued to nothing, because perfect imagery now costs less than a sandwich and takes eleven seconds to produce. When a signal becomes free, it stops carrying information, and the market goes looking for the next thing that is expensive to fake.

The next thing that is expensive to fake is specificity. Not beauty in general, which any model can produce endlessly and boringly, but beauty that exists in one place, at one hour, under one particular quality of light, with the wear and the asymmetry and the odd detail that nobody would have invented. The stain on the marble where a hundred thousand glasses have been set down. The way the sun comes through the west window of room 214 for about twenty minutes in October. The chef's hands. The scratch on the bar. These are the things a generative model cannot know, because they were never written down anywhere, and they are the things a returning guest recognises.

Unabashed is the right word for what this demands. The instinct in hospitality marketing has been to smooth, to declutter, to remove the human evidence from the frame, to shoot the suite with nobody in it. That instinct now works against the property, because the empty, immaculate, slightly airless room is exactly what the slop produces by default. Showing the real thing, including the parts that are old and worn and particular, has become the most credible thing an operator can do.

Practically, we think that means a handful of shifts.

Shoot with named humans and say who they were. A photographer's credit is a provenance claim, and provenance is the currency now. Publish the unretouched set alongside the hero images and let guests move between them. Show the room at the hour it is actually least flattering and let the honesty do the selling.

Invite the audit rather than surviving it. The properties that will win are the ones that put guest photography beside their own, at equal size, without curating the difference away, because a hotel confident enough to do that is making a claim no competitor can copy cheaply.

Put the staff in the frame. A generative model produces plausible people. It cannot produce the specific bartender who has worked at the property for nineteen years and knows what you drink. That person is the single most defensible asset in the building.

Sell the thing that cannot be photographed at all. Temperature, noise, smell, the weight of the door, the silence at four in the afternoon. Language does work here that images no longer can, and most hospitality copy has been written as though nobody would ever read it.

None of this is an argument against AI in hospitality. It belongs in revenue management, in scheduling, in translation, in service recovery, in a dozen operational places where it earns its keep quietly. The line sits exactly where a business begins fabricating the thing it is selling, and hospitality of all industries should be the last to cross it. This is the sector that has always traded on the real thing. Counterfeiting it is a strange way to compete.

10 prompts to run at your marketing team

1. Delete every image on our website that could have been generated by a model in eleven seconds. What is left, and is it enough to book a room?

2. Photograph the least flattering hour of our worst-selling room and write the caption that makes somebody want it anyway.

3. If a guest arriving today compared our gallery against the property with a stopwatch, where would they first feel lied to?

4. Our competitor just published every unretouched frame from their last shoot. Write our response by Friday.

5. What does our property smell like, and why has nobody ever put that in a piece of paid media?

6. Build the case for firing our retoucher and hiring a documentarian instead. Then build the case against it. Which one do you actually believe?

7. Name the five things in this building that took thirty years to happen and could not be recreated with any budget. Why are none of them in the brand book?

8. Write the one-star review we deserve. Then write the operational change that makes it obsolete.

9. If polish is free and provenance is expensive, what should we now be spending our production budget on?

10. Our longest-serving employee is more valuable than our entire content library. Prove it or disprove it with numbers.