Before any of this was possible, one of us worked on food photography for a very large fast food brand. The ice cream in those shots was not ice cream. It could not be, because it would have collapsed in the first ten minutes under the lights. The burger was propped, pinned and built to survive an hour of being looked at.
Nobody called that a fake. It was understood as what commercial imagery has always been, which is constructed.
That history is worth holding onto, because it tells you what the objection to AI images actually is. It is almost never about the technique.
So what is the real objection?
A broken promise.
When someone says an image feels wrong, dig into it and you will usually find they are not upset that software was involved. They are upset because the picture told them something that is not true. The patio in the photo is not your patio. The room is bigger than the room. The product has a feature it does not have. The team in the shot is not your team.
People have made peace with styling. They have not made peace with being misled, and they are much better at spotting it than most owners expect.
What is the test, then?
One question, applied to every public-facing image:
Does this picture promise the customer something they will not get when they turn up?
If the answer is no, the tool you used stops mattering. A background texture behind your own product, a bit of colour correction, a distracting sign cloned out of the corner: nothing there is a promise. It is the same work a retoucher has done for forty years.
If the answer is yes, the tool you used still does not matter. A real photograph of a location that is not yours is just as misleading as a generated one, and it will land just as badly.
Is there a rule I can just follow?
Yes, and it holds up under pressure because it is true and because anyone can check it:
AI in the back office. Real in the shop window.
Everything the customer looks at and forms an expectation from is real: your place, your product, your people, your work. Everything behind that is fair game.
The rule survives a sceptical audience precisely because you can be open about it. “The photos are ours, the scheduling and the drafting are automated” is a sentence you can say out loud without flinching. That is a good test of any policy.
What counts as the back office?
More than most owners realise, and it is where nearly all of the time saving lives anyway.
- Drafting captions, then editing them so they sound like you
- Resizing and reformatting one asset into every placement it needs
- Cleaning up a real photo: exposure, colour, a cluttered background
- Laying out templates so your posts look consistent
- Writing alt text and file names properly, which also happens to be what makes your images findable
- Sorting, tagging and naming your image library so you can find the shot you already have
One client engagement makes the point about where the hours actually go: a founder was spending six-plus hours a week planning, drafting and composing social content. A system that handles the production work took that to about 45 minutes of review. None of that saving came from generating a picture. It came from the work around the picture.
What if my photos are not good enough?
They are probably better than you think, and the bar is lower than the vendor examples suggest.
A clear phone photo, taken in good light, of the actual thing, beats a polished generated image almost every time. It reads as true, and audiences reward that. Slightly imperfect often does better than slightly perfect, because the imperfection is the signal that a human was there.
The practical move is boring and it works: build a library of your own. Photograph your products properly once, in consistent conditions. Get someone on the team into the habit of shooting a clip whenever something interesting is happening. Put it all in one folder with a naming convention you will still understand in a year. That library is an asset, it appreciates, and it removes the temptation to invent something because you had nothing to post.
If capturing it is the bottleneck, that is a real constraint and worth solving directly. A student, a local videographer for a day, or an hour a week with a phone will produce more usable material than any generator.
Will AI images hurt my search visibility?
Not by existing. Google’s published guidance is that its focus is on the quality of content rather than how the content was produced, and that using automation is only a problem when the purpose is to manipulate rankings (Google Search Central, guidance on generative AI content).
Two things are worth knowing anyway. For e-commerce listings, Google asks that AI-generated images carry IPTC DigitalSourceType: TrainedAlgorithmicMedia metadata, and that AI-generated product titles and descriptions be labelled as such (Google Search Central). And the risk that should actually concern you is not a search penalty. It is a customer who arrives, sees the real thing, and quietly concludes you overstated it.
What about my logo?
Different question, and the answer is firmer: get it designed.
A logo has to work at one centimetre and on a vehicle, in one colour, embroidered, and as a favicon. That is a set of technical constraints, not a picture. Generators produce something that looks like a logo in the one size you generated it at. You will also want clean vector files you own outright, which matters the first time you order signage.
Where to start this week
Open your last twenty public posts and run the test on each one: does this promise something a customer will not get? Most businesses find one or two that fail, and the fix is usually swapping in a photo they already had.
Then write your own version of the rule down, in a sentence, and tell your team. A stated policy is what stops the awkward judgement call being made in a hurry by whoever is posting that day.
If you want the production side handled so the only thing left is capturing the real stuff, book a free readiness call. Twenty minutes, no pitch. We will look at what you already have and where the hours are going.