How to get your business cited by AI search: Google AI Overviews, AI Mode, ChatGPT and Perplexity, for a UK firm.

Google Search Central’s page on AI features and your website states that there are no additional requirements to appear in AI Overviews or AI Mode and no special optimisations, because those features are rooted in Search’s core ranking and quality systems, and that traffic from them is included in Search Console’s overall figures. Its May 2026 guide to optimising for generative AI features repeats the fundamentals, a clear technical structure and unique, valuable content, and adds the emphasis that matters: expert-led content that provides value beyond common knowledge. In June 2026 Search Console gained a dedicated generative AI performance report showing impressions within AI Overviews and AI Mode. Everything else on this page is those three documents applied to a small firm.

For a query with a place in it, “AI consultancy London”, “dentist Wandsworth”, “accountant near me”, Google’s AI answers lean on the local results, which come from Google Business Profile. A profile with the right primary category, a complete description, opening hours, services, photos, a website link and reviews with replies is the single largest factor for a local firm, and it is free. The same signals belong on the website as LocalBusiness structured data with the same name, address, phone and opening hours, so the two agree. Our own experience: the change that put us into the local answer for our main query was fixing the profile’s category and description, not anything on the site.

AI answers are assembled from passages, so the passage has to exist. Each page should take one question a buyer actually asks, state the answer in the first paragraph in plain terms, then support it with sections that a reader could skip to. Put the specifics in: the price range, the timescale, the rule, the threshold, the source. Attribute every figure to where it came from, with a link, because a claim with a source is quotable and a claim without one is not. Add a short set of questions and answers at the end that restate the specifics, marked up as FAQ structured data. And write it in the language the buyer uses, not the industry’s.

AI answers favour what they can quote. A firm that publishes its price ranges gets named in answers to “how much does X cost”; a firm that says “contact us” does not. A page with a named author, a role and a short biography carries the experience and expertise signals Google’s guidance describes; an anonymous page does not. Case studies with a measurable outcome, even without a client name, give the model something specific to cite. None of this is a trick. It is the difference between a page that says something and a page that says nothing, which is what “value beyond common knowledge” means.

Fast pages that render their content as HTML rather than behind scripts; a sitemap that lists every page and nothing else; structured data for the organisation, the local business, the services and the FAQs; and a plain-text summary of the site at a well-known path so AI crawlers that read it get the facts in one request. Check which AI crawlers visit in the server logs, and decide deliberately which to allow in robots.txt: a firm that wants to be cited allows the search-related ones. Our web team builds all of this in as standard, and our performance work on our own sites is a matter of public record.

Three measures, monthly. The Search Console generative AI performance report for impressions in AI Overviews and AI Mode, by page and by query. A fixed list of twenty buyer questions asked, signed out, in Google AI Mode, ChatGPT and Perplexity, with a note of which firms are named, which is a spreadsheet and an hour a month. And referrals from AI assistants in analytics, which need the referrers tagged. The first shows Google’s view; the second shows the buyer’s; the third shows whether anyone arrived. Expect movement over months, not weeks, and expect the local answer to move first.

Invented statistics, which are the fastest way to be excluded once a model checks them against a source. Pages written for a keyword rather than a question. A hundred thin pages instead of ten complete ones. Blocking AI crawlers and then asking why the firm is not cited. Paying for “AI search optimisation” services that promise a submission process, because none exists. And copying a competitor’s content, which produces the same answer with the competitor named.