Why Tourism Should Not Use AI-Generated People or Places

We wrote about China's labeling regime for AI-made content and set one question aside at the time: the label settles a duty, and says nothing about whether the picture was a good idea. That post handled the duty; the labeling rule is here and it still applies. This one is the argument we deferred.

It comes down to something a compliance answer cannot reach. Your product is a place. An image of that place does not decorate the product — it describes it, and the customer eventually stands inside the thing being described.

The short version: in tourism a picture of the location functions as a specification, so a generated one is a claim about what the traveler will find, and the traveler goes and checks. Generated people are worse than generated places, because an image of visitors enjoying your site reads as evidence that visitors enjoyed your site. Chinese travelers in particular arrive having compared your images against photographs taken by people who actually went. Generated imagery is fine anywhere it is not asserting what somebody will see.

A picture of a place is a specification

Sell a jacket and the studio shot is a rendering of a thing the buyer will hold; slight differences in light are not a broken promise. Sell a day at a lagoon and the picture of the lagoon is the offer. The colour of the water, how many people are in the frame, whether the sun is low — every one of those is a term the customer thinks they agreed to.

A generated image gets all of them wrong in the same direction. These tools default to golden light, empty foregrounds and improbably saturated water, because that is what the training material rewarded. Your ten o'clock departure in August lands in flat light and a car park with four other coaches in it. Nobody labelled anything dishonestly, the gap is there anyway, and that gap is what the complaint names.

Generated people are a different and worse case

A generated place misdescribes the product. A generated person makes a claim about other customers.

An image of a family looking delighted at your viewpoint says something specific: people like this came here and this is how it went. That sits much closer to a testimonial than to illustration, and a fabricated one at that. We hold the line on not inventing reviews or screenshots for clients; an invented photograph of invented happy visitors belongs in the same category, whatever the technical route to producing it.

There is a practical layer under the ethical one. Generated people are still where these models are most detectable — hands, teeth, the way a group's eyelines do not agree, clothing that resolves into nothing at the seams. An audience that spends hours a day scrolling images has calibrated for it, and the cost of being caught is not a complaint about the picture. It is a reader who now treats everything else you published as possibly fake too.

Three ways it goes wrong in practice

Composite geography. Models merge. You ask for your valley and get your valley with a mountain that belongs three hundred kilometres away, or a building in a style that exists in the next country. This is invisible to you and obvious to somebody who has been.

Season and weather that you do not sell. Green where your season is brown, snow where there is none in the month you run, a sky that does not happen there. Travelers book a month, not a place, and the mismatch is specific to that.

Details that cannot be true. Signage in a language nobody uses there, a vehicle type that could not be licensed there, a plant that does not grow at that latitude. Each one is small. Each one is also the kind of thing a person planning a trip notices, because planning a trip means looking at hundreds of images of the same place.

Why this bites harder with Chinese travelers

Your image does not get judged on its own. It gets judged next to a Xiaohongshu note from somebody who went last month and posted eleven unretouched photographs.

That comparison is built into how the trip gets planned — the research happens inside the platforms, against first-hand material, and it goes deep. A generated picture sitting beside real ones from the same viewpoint does not read as aspirational. It reads as the odd one out, and the reader's next question is what else on the page was made rather than photographed.

The same dynamic is what makes real guest photographs worth so much, and the permission side of using them is worked through in using traveler content.

Where generated imagery is genuinely fine

The rule that survives: use it anywhere it makes no claim about what somebody will see.

Backgrounds and textures behind text. Diagrams, maps, icons, charts. An abstract header for a post about payments. Anything where nobody could mistake the image for a record of a place. None of that touches the problem above, and all of it still carries the labeling duty, which is a separate matter from whether the choice was wise.

What falls outside: product pages, the cover of a note about a specific location, anything showing people at your site, and anything a customer could reasonably read as "this is what the day looks like".

The cheaper answer was always there

The real constraint for most operators is that photographing your own product properly costs a day and some attention, and generated images feel like a way around that.

They are not, and the alternative is closer to hand than it looks: your guests are photographing your product continuously, and what they choose to photograph is more useful than what you would have staged — an argument we made in designing for the photo. A real photograph of a grey day with the right people in it outperforms a generated perfect one, because the second kind sets up a comparison the first kind never has to survive.

Where this stops being a rule

This is an argument about representation rather than a prohibition. A labelled generated image is lawful, and these tools keep getting better at the things listed above, so some of the tells will stop being tells.

What will not change is the structure of the problem: the customer travels to the thing depicted and compares. No improvement in image quality removes that, because the failure was never about the image looking fake. It was about the image showing something other than what is there.

Where CN1X fits

When we build your Chinese pages we use your photographs, and where you do not have enough of them, that is a gap we name rather than fill with something generated. Sometimes the honest recommendation is that you need a day of shooting before the page is worth building.

Images of people who were never there sit with invented reviews and invented screenshots, and we supply none of the three. If you are unsure which of your current images would survive being compared against what travelers post from the same spot, send us the page and we will tell you which ones we would replace.