Getting recommended by ChatGPT as a local small business: what actually works

At the scoping call, there's a moment I rather like. I take out my phone, I open ChatGPT in front of the owner-manager, and I type the question their customers genuinely type. Not the company's name, they'd find that for sure. The question from a customer who doesn't know them yet, with their trade and their town in it: "I'm after a good plumber to redo a bathroom, who would you recommend?" Three names come up. Last time, none of the three was his, and one of the three was his direct competitor, the one he'd been rolling his eyes about since the start of the meeting.

The silence afterwards, I know it by heart. It isn't an angry silence, it's the silence of someone who's just realised a game started without them. So let me tell you straight away what I wish someone had told him sooner: how these systems choose who to cite, what a small business can do this very week, and what takes time.

Are your customers really asking AI who to call?

Yes, but not yet on a massive scale, and I'd rather start there than sell you the panic. In France, France Num estimates that AI engines account for under 20% of searches and around 1% of the traffic sent to websites. One percent. If someone tells you Google is dead and you should bet everything on ChatGPT by tomorrow morning, they're telling you a story.

What's shifting is the speed. In the United States, BrightLocal's annual survey measures that 45% of consumers used AI to get a business or tradesperson recommended to them in the past year, against 6% the year before. From 6 to 45 in twelve months. The French market isn't the American market, but nearly 44% of working-age French people already use ChatGPT or an equivalent for something else. The habit is already there; it's only waiting for the occasion.

The day an AI stops giving your name, nobody tells you.

The difference with Google is right there, and it's brutal. On Google, being fifteenth means being far down; people can still find you, if they search. AI doesn't make a list. It gives two or three names and stops. There's no page two.

How does AI decide who to cite?

It doesn't rank, it cross-checks. That's the whole difference, and it's what explains why the reflex of "I'll optimise my site" isn't enough. Before putting a name forward, these systems go looking for what several independent sources say about you, and check whether they tell the same story.

Yext analysed 6.8 million citations across more than 1.6 million answers generated by Gemini, ChatGPT and Perplexity. The breakdown speaks for itself: 44% of citations point to the business's own site, 42% to third-party listings and directories, 8% to reviews and social media. In other words, your site accounts for less than half of what the machine looks at. The rest is what others say about you, and you don't write a single line of that rest.

Concretely, if your site states one set of hours, your Google listing another, and an old directory a phone number you haven't had for two years, you don't come across as messy. You come across as uncertain, and a machine whose job is not to say anything wrong avoids what's uncertain. Consistency between your sources isn't a tidiness detail, it's the signal.

Next come reviews, which carry weight because they're the one part of the file you didn't write yourself. SOCi's 2026 Local Visibility Index observes that businesses recommended by ChatGPT average around 4.3 stars. And it gives a figure that puts things in perspective.

I want to flag this, because someone will soon tell you the opposite: this study covers American chains, not the tradesperson on your street. You can't derive a percentage of French small businesses cited or not from it, and I haven't seen anyone measure that seriously. What it does say, though, applies to you: the filter is tight for everyone, including those with a budget. And ranking well on Google protects you from nothing, since in retail, fewer than half of the businesses most visible on Google are among those most recommended by AI. These are two different games.

What are the five moves within reach of a small business?

None of the five needs a budget, or technical skill. They need a morning, then a bit of consistency. Here they are, in the order I'd do them in your place.

1. Your Google listing, properly looked after. It's the most cross-checked source there is for a local business, and it's free. The right trade category, your genuine service area, accurate hours, the right phone number, recent photos, and a description that says precisely what you do. A half-filled listing is an empty file.
2. Reviews, regularly. Two or three a month are worth more than thirty all at once in September and then nothing. Ask for them at the end of the job, while it's still fresh, not three months later. And reply to all of them, including the bad ones: a measured reply under a two-star review says more about you than ten glowing ones.
3. A legible site. Legible doesn't mean pretty. It means fast, correct on a phone, and above all precise. "Reliable tradesperson at your service for 20 years" says nothing, because your competitor writes exactly the same line. "I bring 1960s and 1970s house electrical panels up to code" gives something to grab onto. The test fits in one question: could my competitor copy this sentence onto their own site word for word? If yes, it's useless.
4. Your prices, shown. This is the rarest move and the most rewarding. A price is a verifiable fact, so it's citable; "free quote" is nothing at all. You don't need pricing to the penny: an honest range and a rough figure are enough. I show mine, and that's why I can recommend it to you without blushing.
5. An FAQ, with the real questions. Not the questions you'd like to be asked. The ones you're asked on the phone, ten times a week, in the customer's own words: "do you work Saturdays?", "how much does it cost to change a water heater?", "how far do you travel?". One answer per question, in two or three sentences, written the way you speak. That's exactly the format AI knows how to pick up.

You'll notice that none of these moves is a trick. They're the same things that serve you with Google, and that serve the human who ends up reading your page. That's the best news in this whole story: you're not working for three audiences, you're working for a single principle, clarity. I've seen an estate agency's organic traffic take off with no other secret than that.

And what about structured data?

It helps, but it isn't the magic key you'll be sold. Structured data is invisible markup in your site's code that tells a machine "this is a price", "this is an opening time", "this is a question, and here's its answer". Useful, yes. Magic, no, and it's Google itself that says so in its official documentation: there's no special markup to add to appear in its generated answers, and no extra requirement beyond standard SEO.

One example to inoculate you. FAQ markup used to display neat drop-down menus directly in Google's results for a long time. On 7 May 2026, Google stopped showing them, with no announcement, just a quiet note added to a technical page. Those who'd built their strategy on that trick lost it overnight. Those who'd written genuine answers to their customers' genuine questions lost nothing at all: the content itself is still read, by Google as much as by AI. The moral applies to this whole story. Markup follows the substance, it never replaces it.

The real lever is less glamorous: facts. France Num sums it up well, the content that gets cited is the kind that brings numbers, cites its sources, stays clear and up to date, and this work can increase visibility by up to 40% on complex questions. Facts, dated, verifiable. Not adjectives.

How long does it honestly take?

Months, not days, and nobody can guarantee you the result. I'd rather say it plainly: I don't know exactly how ChatGPT chooses its three names, nobody really knows from the outside, and anyone who guarantees you first place in an AI is lying to you. These systems change their minds, don't answer the same way twice, and update on their own schedule. Your reviews take months to build up, because you can only ask for them as you go along.

What can be guaranteed is the opposite: what excludes you for certain. An empty Google listing, three reviews dating from 2023, a site that says nothing precise, no price anywhere. That's measurable, and it's fixable. Nobody can promise you'll be cited, but you can be made to deserve it, and that's already a great deal more than most of your competitors will do this year.

So make the five moves. One morning, and they'll serve you either way. Only after that does the question become: has anything shifted, and on which questions? The guide above, you can follow alone. Knowing where you stand, you can't: you have to go and put the questions to the engines, one by one, and see who comes up. That's the job of my AI visibility audit: you find out exactly which questions you appear on and which ones go to the business next door, screenshots included, for €490 excl. VAT, with the firm price of the project written into it if you go ahead. And if you suspect your site is the weak link, we can almost always rebuild it without losing your Google ranking, but check first: most of the time, it isn't the problem.

Sources: France Num, "Optimisation pour les moteurs génératifs (GEO)", published 9 February 2026, updated 24 April 2026 (under 20% of searches and 1% of traffic, 44% of working-age French people using generative AI according to the Microsoft Global AI Adoption 2025 study, GEO levers and gains of up to 40%, no paid placement) · BrightLocal, Local Consumer Review Survey 2026 (45% of consumers, against 6% the year before; 1,002 American consumers surveyed) · Yext, analysis of 6.8 million citations across 1.6 million AI-generated answers (breakdown: 44% sites, 42% third-party listings, 8% reviews and social media) · SOCi, Local Visibility Index 2026 (1.2% of businesses recommended by ChatGPT against 35.9% in Google's local block, average of 4.3 stars, 45% overlap in retail; nearly 350,000 outlets from 2,751 multi-location chains, US market) · Google Search Central, "AI features and your website" (no specific markup required) · Google, FAQPage documentation (end of FAQ rich results display on 7 May 2026).

Frequently asked questions