A restaurant, a salon, a garage. The AI advice aimed at small businesses is mostly about volume, and volume is the wrong goal when your customers can walk in and check.
- Never generate the product. A generated plate of food is an advert for a dish you do not serve.
- Generate everything around it — social posts, menu layout, seasonal graphics, the boring copy nobody has time to write.
- Your advantage is being real, and a local business that posts obviously synthetic images throws away the only thing the chains cannot copy.
- Never generate a review or a reply that pretends to be a customer. Platforms delist for it, and in the US the FTC rule banning fake reviews — AI-generated ones included — has been in force since October 2024.
- Photos of your premises must be your premises. Same rule as the food.
- The best use is the invisible one: admin, drafts, translations, descriptions — not pictures.
A customer who sees the photo will be standing in the room within the week.
National brands can advertise an idealised burger because you will never meet the person who made it. A local business is judged against the thing itself, immediately, by someone who will tell their friends.
The food is not decoration
Why generated food is worse than a bad photo
Generated food looks superb and is always slightly wrong: the wrong number of prawns, a garnish you do not use, a plate you do not own, a burger with a physically impossible cross-section. Customers do not analyse it. They notice the gap when the real one arrives.
And the cost is asymmetric. A mediocre real photo disappoints nobody, because it sets an accurate expectation. A beautiful fake photo creates a disappointment you paid to manufacture — plus a complaint, and possibly a claim that your advertising was misleading.
Take the photo. A modern phone by a window, dish on a plain surface, shot from the angle a diner sees it. Ten minutes, and it will outperform anything generated because it is true.
What generation is genuinely good at here
Menu descriptions have one trap, and it is the same one as everywhere else: a model asked to make a dish sound appealing will add ingredients. Anything on a menu is a claim, and some of those claims are allergen claims. Give it the real ingredient list and forbid additions.
Photograph the thing you sell. Generate the frame around it. Every rule on this page is that sentence applied to a different asset.
Reviews are the bright line
Writing a review is not marketing
Generating reviews of your own business, or replies written as though from customers, is fabrication. It is not a grey area and it is not an efficiency. Platforms suspend and delist for it — and in the United States it is now specifically illegal.
The FTC’s rule on consumer reviews and testimonials took effect on 21 October 2024. It bans fake reviews and testimonials outright, expressly including AI-generated ones, along with paying for positive or negative reviews, insiders posting without disclosing the relationship, and company-run sites posing as independent. Knowing violations carry civil penalties in the tens of thousands of dollars per violation.US Federal Trade Commission, Trade Regulation Rule on the Use of Consumer Reviews and Testimonials — approved 14 Aug 2024, effective 21 Oct 2024; AI-generated fake reviews are covered. The maximum civil penalty per knowing violation is adjusted for inflation annually, so no figure is quoted here — it was above $50,000 when this was checked. US rule; other countries handle fake reviews under their own consumer-protection law. An earlier version of this page said only that "the regulatory direction has been tightening", which was vaguer than the facts allow. Corrected 6 Sep 2026.
Using AI to draft your reply as the owner is fine — that is your voice, assisted. Read it before it goes out, because a generic apology to a specific complaint is worse than no reply, and readable as machine-written by exactly the customers you are trying to keep.
Reply to the specific thing they said. That is the whole skill, and it is the part a model does not know.
The thing you lose by sounding like everyone else
Every local business now has the same tool, which means the default output is the same. Feeds fill with identical warm-lit stock imagery and identical friendly captions, and the effect is that a real place starts looking like a franchise.
Your genuine advantages are unglamorous and unfakeable: the actual room, the actual people, the regular who has come every Tuesday for nine years, the dish that is a bit odd and sells out anyway. None of that is generatable, which is exactly why it is worth posting.
The practical version: use AI for the parts customers never see, and put a real camera on the parts they do.
Spend the AI time on admin, not on images. Supplier emails, listings kept current across every directory, translated menus, a month of caption drafts you then edit.
That work is invisible, genuinely tedious, and has no downside if the model gets it slightly wrong — which is the profile of a good AI task, and the exact opposite of a photograph of your food.
Before you post
No platform policies are quoted and no figures are given. Review and advertising rules differ by platform and by country and change often; a specific claim here would be stale before it was useful, and the ones that matter to you are on the platform you actually use.
The reasoning is craft, not measurement — it follows from one asymmetry that does not change: a local customer verifies your advertising in person, usually within days. The related pages carry the sourced versions: AI product photography for the photograph-the-object method, and how to spot AI writing for why generic replies read as generic. Checked 6 September 2026.
The through-line: the chains can afford an idealised photograph because nobody checks. You cannot, because everybody does — and that is a strength if you let it be one.