AI SEO for a local business

Nearly everything written about AI and search is written for brands with content teams and a budget for monitoring software. Very little of it survives translation to a business with one location, one van, or one chair.

This page is the local version. It assumes you are not going to publish twelve articles a month, and it assumes that when we say something is worth doing you want to know roughly what it costs you in hours.

The uncomfortable number first

Analysis in 2026 found ChatGPT surfaced roughly 1.2% of local business locations in its local recommendations. Google's local results show 35.9%. Gemini named about 11%, Perplexity about 7.4%.

That gap is the whole story of this page. Google local was a broad net: be reasonably set up and you appeared somewhere, for something, to somebody. AI recommendation is a narrow one. It names two or three and the rest of the trade in your town does not exist for that question.

Two readings of that, and both are true. It is harder. And the businesses that get inside it early are inside something far more exclusive than page one ever was.

The one thing that matters most, from our own testing

When we built the test on this page, we watched what the models cited to justify naming a business. The pattern was consistent enough to be worth building a whole approach around:

GPT-5 and Gemini both recommended local businesses by citing the businesses' own service-and-city pages. Not directories. Not review sites. Their own pages.

That is unusually good news, because it is the part you own outright. No platform, no gatekeeper, no waiting on somebody else to write about you. A page that exists, states the service and the place, and can be read without running any code.

What that page looks like

Concretely, for a mobile notary in Palm Beach Gardens, the page is titled and headed with the service and the city, and then it does four things:

1. Answers the question in the first two sentences. What you do, where, and how someone starts. Before any history, any values, any hero copy.
2. Names the actual places. Neighbourhoods, landmarks, the adjacent towns you genuinely cover. Not a keyword list; the way you would describe your patch to a new hire.
3. Says the specific things. Price ranges if you can bear it, turnaround times, hours, what is and is not included. Specifics are extractable. “Competitive rates” is not.
4. Answers the four questions you get on every call. Written as the questions, in the customer's words, each answered immediately underneath.

One page per service you genuinely want more of. Not one per keyword combination. If you would not be pleased to receive the call, do not build the page, because a thin page built for a model to read is a page a human will eventually land on and judge you by.

If you have no storefront

Service-area businesses have a particular problem here. You cover six towns, you are physically in one, and the honest instinct is to avoid naming any of them too loudly in case it looks like you are pretending to be local everywhere.

The way through it is to be straightforwardly truthful about the shape of the business. Say where you are based. Say which towns you actually serve and roughly how far you travel. Then build a real page for the two or three you most want work from, with genuine local specifics in each, rather than twelve near-identical pages with the town name swapped.

Twelve swapped pages is the oldest trick in local SEO, it reads as spam to a human, and a model comparing them arrives at the same conclusion for the same reason. Three honest pages outperform twelve hollow ones and cost less to write.

Reviews, and the thing nobody tells you about them

Your star rating is not the useful part. The useful part is the words.

A model deciding whether to name you is looking for corroboration in the terms the customer asked about. Forty five-star reviews saying “great service, highly recommend” corroborate almost nothing, because they do not say what the service was. Twelve reviews that mention water heater replacement, the neighbourhood, and how long it took corroborate the exact thing somebody is about to ask for.

Which changes what you should ask for. “Would you mind leaving us a review?” produces the first kind. “Would you mind mentioning what we actually did and roughly where you are?” produces the second. Same effort, completely different asset.

The order, with rough hours

1. Check for a crawler block. Half an hour, usually free. Both kinds.
2. Check you are readable without JavaScript. Fifteen minutes to diagnose. The fix depends on how the site was built, and on a builder-based site it can mean a rebuild.
3. Get into Bing. Twenty minutes, free, and almost nobody has done it.
4. Write one real service-plus-city page. Half a day done properly. The highest-value item on the list.
5. Fix your structured data. An hour or two, and check before rewriting: schema is often reported missing when it is present but using a specific subtype.
6. Change what you ask for in reviews. Free, ongoing, slowest to compound and hardest to fake.
7. Publish llms.txt. Twenty minutes. Do it last and pay nobody for it.

That is a realistic week of work spread across a month, and it is most of what a small business can usefully do. The item that most often turns out to be a bigger job than it looks is number two, because if the site assembles itself in the browser there is no setting to change. That is the conversation we end up having most often, and it is the one we would rather you reached with evidence in hand than took on trust.