How to improve your business's visibility in AI search

The short version: a checklist for one location

If you run one location without a marketing team, five moves do most of the work, and almost all of them are free. The rest of this guide gives the evidence and the timings for each, and GEO for a local business shows what changes from one trade to the next.

• Complete your Google Business Profile: categories, hours, services, photos and a real description. Gemini and Google's AI Mode read it directly.
• Earn recent, genuine reviews and keep them coming, on Google and on the platform your customers use, such as Yelp in the US.
• List your business on the main directories and registers for your trade and town, with exactly the same name, address and phone everywhere.
• Get onto the local best-of lists and trade pages that cover your area; a mention there can count for more than your own homepage.
• Write one clear page per service that answers what customers ask before they call, and check that AI crawlers can read your site.

What decides whether AI recommends your business?

An assistant does not know your business first-hand. When a customer asks for a plumber or a physio nearby, it searches, reads a handful of sources and repeats the names those sources agree on. Two large studies show what those sources are. Yext analysed 6.8 million citations from ChatGPT, Gemini and Perplexity in 2025: 44% pointed to websites and 42% to business listings, with reviews and social sites far behind. BrightLocal's August 2026 study of 1,355 locations found the Google Business Profile was the most cited domain overall, and Yelp appeared as a source in 80% of ChatGPT's local answers.

So improving your AI visibility is mostly about being present, accurate and well reviewed on those sources, with your own site as supporting evidence. If you first want to understand why you are missing today, read why your business might be invisible to AI.

Which sources each assistant reads, and what to fix first

The groundwork is shared, but each assistant leans on a different source for local businesses, so the first fix depends on which one your customers use. The table sums up what is documented as of October 2026.

• Yelp's shareholder letter of 6 August 2026 says Yelp ratings and reviews "recently began powering ChatGPT's local experience". Yelp's coverage varies by country, so elsewhere the directories that matter for your trade carry more weight: in our study, the NHS directory for health trades in the UK, and PagesJaunes and Doctolib in France.
• Google's help for the Gemini app says it automatically uses public information from Google Maps, and Google's guidance for AI Overviews and AI Mode asks businesses to keep Business Profile information up to date. Our guide on how the Business Profile affects each assistant goes into detail.
• Assistants disagree more than people expect. In our own study of 756 answers, ChatGPT and Gemini named at least one business in common only 51% of the time for the same trade and city, so being named by one tells you little about the other (which websites AI trusts).

Earn reviews steadily, and know their limits

Reviews are the clearest correlation in the data. Insites studied 10,000 businesses in May 2026: those recommended by both ChatGPT and Perplexity averaged 133 Google reviews, while those invisible to AI averaged 11. In SOCi's index, locations ChatGPT recommended averaged 4.3 stars, and SOCi reads reviews as a gate rather than a ranking signal. Neither study proves cause, but both point the same way.

Reviews are necessary, not sufficient. In one of our own audits in July 2026, a London clinic rated 4.9 with 189 reviews was named by ChatGPT in none of 45 answers, while Claude named it 13 times and Gemini 6. Across the three assistants, the answers about its trade leaned on a medical directory, a doctor-listing site and a review platform far more than on the clinic's own site. Customers care about recency too: BrightLocal's 2026 consumer survey found 74% look for reviews from the last three months.

• Ask every satisfied customer, right after the job, with a direct link to your Google review form.
• Spread the effort to the platform each assistant reads, such as Yelp for ChatGPT in the US.
• Reply to reviews briefly and specifically; the details customers mention, like the job and the area, are what assistants quote back.
• Never buy or gate reviews. It breaks platform rules and puts the listing itself at risk.

Get mentioned on the pages AI reads

Third-party mentions are the lever most owners skip. Ahrefs studied 75,000 brands and found that branded web mentions correlated with AI Overview visibility at 0.664, far more than backlinks at 0.218. In Whitespark's 2026 survey of 47 local search experts, presence on "best of" lists curated by experts ranked as the top factor for AI visibility. Research on generative engines points the same way: AI search favours earned media over a brand's own pages, as our AEO vs GEO explainer details.

For a local business, that means the local press, your trade association's member directory, "best [trade] in [town]" roundups, supplier and partner pages, and community sites. One accurate mention on a page an assistant already trusts can do more than another page on your own site. All of this evidence is correlation or expert opinion, so treat it as a strong bet rather than a guarantee.

Publish pages that answer specific questions

Your website is where an assistant checks the details once it has a reason to consider you. Google's guide to its AI features warns against commodity content, writing that something like "7 Tips for First-Time Homebuyers" adds "little unique insight". A page that states your service, your area, your price range and how long a job takes gives an assistant something specific to repeat. Answer engines look for this kind of page: in our Gemini study, 72% of local answers ran at least one advice search such as "what to look for when choosing a plumber in Dubai" (what is answer engine optimization).

Formats matter less than substance. SE Ranking's study of 216,524 pages found sections of 120 to 180 words averaged 4.6 ChatGPT citations against 2.7 for sections under 50 words, while pages with FAQ sections or question-style headings did not get cited more on average. Freshness does count: Ahrefs found AI assistants cite content that is 25.7% fresher than what ranks in classic search, and Seer found 75% of the pages AI cited had been updated within the past year. Changing a date without changing the page does not help.

• One page per core service, opening with a plain answer: what you do, where, for whom, and roughly what it costs.
• The questions customers ask before calling you, answered with specifics: timings, permits, what to prepare.
• Real details only you have: areas served, certifications, the brands you work with, before-and-after photos.
• A visible last-updated date, changed only when the content changes.

Let AI crawlers read your site

An assistant cannot cite a page it cannot reach. OpenAI says that to appear in ChatGPT search you should allow OAI-SearchBot in your robots.txt, a separate bot from GPTBot, which collects training data; a change takes about 24 hours to apply. For Google's AI features, a page must be indexed and eligible to show with a snippet; how to appear in Google's AI Overviews covers the rest. Claude's web search runs through Brave, and Copilot through Bing, so being indexed by those engines matters too.

If your site was built by an agency or uses a security plugin, check that it does not block these bots by default. It is a five-minute check that can rule out a whole assistant.

Do you need an llms.txt file?

No. llms.txt is a proposal published by Jeremy Howard in September 2024 for a plain-text file that helps AI agents use a website. Google's guide to AI features says you do not need new AI text files to appear in Google Search, and that adding one will neither help nor harm your visibility, because Google Search ignores them.

OpenAI's and Perplexity's crawler documentation is built on robots.txt and does not say their answers read a site's llms.txt, and we found no such statement from Anthropic either (all read on 15 October 2026). For a local business, the hour is better spent on robots.txt, indexed service pages and consistent listings. If your web person adds an llms.txt file anyway, it costs little, but do not count it as an AI visibility fix.

What to ask your web person to check

If someone else looks after your website, these checks take them under an hour. None of them gets you recommended on its own; each one removes a reason to be left out.

• The robots.txt file does not block OAI-SearchBot, which ChatGPT search uses, or Googlebot and Bingbot, which Google's AI features and Copilot rely on.
• No security plugin, firewall or CDN setting blocks AI crawlers by default. Some do, so ask them to look.
• Your service pages are indexed in Google, as shown in Search Console's page indexing report, and load without errors.
• Your business name, address and phone number are identical on every page and in any structured data the site already has. Mismatched details are what lead assistants to wrong answers.
• An llms.txt file is optional. There is no evidence that it changes AI answers, so it should not be the first job.

Keep your details identical everywhere

Assistants often get basic facts wrong. Insites checked 10,000 US businesses in July 2026: ChatGPT's details fully matched the Google Business Profile for 55.6% of them, and gave a wrong phone number for 30.1%. Mismatched names, addresses or numbers across your site and listings make those errors more likely and send customers to the wrong place. In our own test of 40 local businesses, most wrong details found by search were copied from another page; what to do when AI has wrong information about your business shows where to fix each one.

There is no published evidence that consistency by itself changes who gets recommended, but it reliably changes whether the answer is right. Pick one exact name, address, phone and website, and correct every listing that differs.

How long does it take to improve AI visibility?

Some fixes apply in hours and others take months, and no study has measured how long a new listing or a batch of reviews takes to change an AI answer. Here is what is documented. If you have just opened, moved or changed your name, the waits for new listings and what we saw in a test of new restaurants are in how long until ChatGPT knows a new business.

What does not work

Some popular tactics show no measurable effect, and a few backfire. The same list is a good filter if you pay someone: our guide to AI search optimization services covers what agencies charge and the red flags.

• An llms.txt file or special AI markup. Google says its search ignores them, and in Insites' study of 10,000 US businesses llms.txt showed only a weak link with being recommended and FAQ schema no meaningful one, far behind Google reviews.
• A separate version of your content written for AI. Google treats that as scaled content and asks for pages written for people.
• Keyword stuffing. It reads badly and has done nothing for AI citations in published tests.
• Buying reviews or mentions. Fake reviews risk your listings, and paid placements on low-quality list sites are easy to spot.
• Judging from a single answer. The same question gives different names from one run to the next.

Can you pay to appear in AI answers?

You can buy an ad next to some answers, not a place inside them. ChatGPT began testing ads in the US in February 2026 and extended them to 31 European countries, including France, in August 2026, to Free and Go users only. They appear below the answer, labelled as sponsored, and OpenAI states that advertising does not influence the answers ChatGPT gives. Google shows ads above and below AI Overviews, and inside them for English-language searches in 12 countries, and is testing ad formats in AI Mode; Microsoft runs ads in Copilot.

None of these buys a mention in the organic answer, which is still earned through the levers above. ChatGPT ads for local businesses sets out which trades may advertise, what it costs and when a test is worth it.

How to know if it is working

Ask the same set of customer questions every month, several times each, on each assistant your customers use, and record whether you are named, who is named instead and which sources sit under the answer. Improvement looks like being named more often, on more questions and more assistants, with new sources starting to cite you.

AI visibility metrics and KPIs explains what to record and how to track AI sources gives the log to keep. A free Peekl check runs your questions on ChatGPT or Gemini and shows who is named instead and from which sources, so you know which lever to pull first.