How to Get Your Business Cited by ChatGPT and Google AI Overviews (2026)

What does it mean to get cited by AI search?

Getting cited by AI search means your page appears as a linked source inside a generative answer, such as a Google AI Overview, a ChatGPT Search response, a Perplexity answer or a Microsoft Copilot result. The engine reads your content, extracts a passage and attributes it to your URL. That citation, not a blue link, is increasingly how UK customers first meet your brand.

This practice is now called generative engine optimisation, or GEO. It overlaps with SEO but rewards different things. Classic SEO earns a ranking position; GEO earns a sentence inside the answer itself. A page can rank tenth on Google yet still be the source an AI Overview quotes, because the model picks the clearest passage rather than the highest position.

For a Cardiff plumber, a Bristol accountancy firm or a Manchester SaaS startup, the stakes are practical. When someone asks an assistant "who fixes commercial boilers in Cardiff?", the businesses named in that answer capture the intent. The rest are invisible, regardless of how their site ranks in the ten blue links beneath.

Which AI engines actually cite UK sources?

Four matter for UK businesses in 2026. Google AI Overviews sit above organic results for a growing share of queries and link out to a handful of cited pages. ChatGPT Search retrieves live web pages and footnotes them. Perplexity is built entirely around cited answers, listing sources prominently. Microsoft Copilot draws on the Bing index and surfaces references inline.

Each engine crawls and ranks differently, but they share one habit: they prefer sources that state a fact plainly, back it with structure and come from a domain the model already associates with the topic. Google describes how AI features choose web sources in its Search Central documentation on AI features, and the same principles travel surprisingly well to the other engines.

How AI engines select and cite sources

AI engines select sources in two stages. First, a retrieval step pulls candidate pages from a search index using the query, the same crawling and ranking machinery that powers traditional search. Second, a generation step reads those candidates, extracts the most quotable passages and decides which URLs to attribute. You must survive both stages: be retrievable, then be quotable.

Why does retrieval still depend on classic ranking?

Both Google AI Overviews and Bing-powered Copilot build their answers from their existing search indexes. If your page is not crawled, indexed and reasonably competitive for the query, it never enters the candidate set the model reads. There is no separate "AI index" you can submit to; the foundation is the same technical SEO you already need.

This means crawlability, fast loading and clean HTML still matter. A page that returns slowly, hides content behind JavaScript or blocks bots cannot be cited because it is never retrieved. Google reiterates in its helpful content guidance that people-first, well-structured pages are what its systems reward, and AI features inherit that bar.

What makes a passage worth quoting?

Once your page is in the candidate set, the model looks for self-contained passages that answer the question without surrounding context. A sentence that begins "GDPR fines in the UK are issued by the ICO and can reach..." is quotable. A sentence that begins "As we mentioned above, this depends on several factors" is not, because lifting it out of the page loses its meaning.

Engines favour passages that pair a direct claim with a concrete detail: a figure, a date, a named authority or a defined step. The more a paragraph reads like a complete mini-answer, the easier it is for the model to extract and attribute it. That is why structure, not word count, is the lever that moves citations.

On-page patterns that get pulled into answers

The on-page patterns that get pulled into AI answers are consistent across engines: an answer-first opening sentence, short self-contained paragraphs, descriptive question-style headings, definition and comparison formats, and explicit numbers with their units and dates. These give the model clean, liftable units of meaning rather than prose it has to summarise and risk getting wrong.

How should you structure a paragraph for citation?

Lead with the answer, then support it. Put the direct claim in the first sentence of the section and the first sentence of each paragraph, then add the qualifier, figure or example afterwards. This mirrors how journalists write and how passage-indexing systems chunk a page, so the model finds the answer where it expects to.

Keep paragraphs to two to four sentences and make each one stand alone. Avoid pronouns that point backwards ("this", "that", "it") at the start of a paragraph, because they break when the passage is extracted. Where a topic has a clear set of options, costs or steps, use a short list or a table; structured formats are extracted far more reliably than long narrative blocks.

Which formats do AI engines extract most often?

Definitions, direct question-and-answer pairs, ordered steps, and comparison tables are the workhorses. A heading phrased as the exact question a user would type, followed immediately by a 40 to 60 word answer, is the single highest-yield pattern for AI Overviews and ChatGPT Search. It maps one query to one passage with no ambiguity.

Numbers earn citations when they are specific and sourced. As a rough guide, an illustrative figure such as our UK SEO cost guide shows SEO retainers in the UK can run from a few hundred to a few thousand pounds a month, which is more quotable than "SEO can be expensive". Attach a unit, a currency, a date or a named source where you can, and the model treats it as a fact worth attributing rather than an opinion it will paraphrase away.

Schema, entities and being quotable in isolation

Schema and entity clarity help AI engines understand who you are and what each page asserts, which makes your content safer to cite. Structured data does not force a citation, but it removes ambiguity: it tells the machine that this page is an article by a named author about a defined topic, published by a recognised organisation, with specific questions and answers inside it.

Which schema types matter for AI citation?

Use Organization schema to define your business as an entity, with a consistent name, logo, address and sameAs links to your verified profiles. Add Article or BlogPosting schema with a real author who has an About page, and FAQPage schema where you genuinely answer questions. These types map directly onto how engines model authors, publishers and question-answer pairs. The full vocabulary lives at schema.org.

Schema also strengthens your entity in Google's wider understanding of the web. A clearly defined Organization, linked to Companies House, LinkedIn and your social profiles via sameAs, helps the engine connect mentions of your brand across the web back to one entity. That connection is what lets a model trust that "Cambria Digital" in one source is the same firm as your homepage.

What does "quotable in isolation" really mean?

Quotable in isolation means a passage keeps its full meaning when it is lifted off your page and dropped into an answer with no surrounding text. The model cannot bring your headings, your earlier paragraphs or your images along; it takes the sentences and nothing else. If those sentences only make sense in context, they will not be chosen.

Test it the way an editor would. Copy any paragraph, paste it into a blank document and read it cold. If a stranger understands the claim, the subject and the scope without scrolling back, it is citable. If it leans on "as above" or an undefined "they", rewrite it to name the subject and restate the point. Strong copywriting and GEO are the same discipline, which is why our note on website copy that converts applies almost directly here.

Crawler access, llms.txt and brand mentions

Crawler access decides whether AI engines can read your site at all. Each engine uses named bots, and your robots.txt and server rules must allow the ones you want to be cited by. Block them and you become invisible to that engine regardless of how good your content is, which is the most common and most avoidable GEO mistake UK sites make.

Which AI crawlers should a UK business allow?

Google AI Overviews use Googlebot plus the Google-Extended control, ChatGPT Search uses OAI-SearchBot for retrieval and GPTBot for training, and Perplexity uses PerplexityBot. Microsoft Copilot relies on Bingbot. To be eligible for citation, allow the search and retrieval bots in robots.txt even if you choose to disallow training crawlers like GPTBot.

The distinction matters. Blocking GPTBot stops your content being used to train models but does not block ChatGPT Search citation, which is handled by OAI-SearchBot; OpenAI documents this split in its overview of OpenAI bots. The robots.txt format itself is standardised, and the Robots Exclusion Protocol (RFC 9309) is worth a read before you edit yours.

Do llms.txt and brand mentions actually help?

llms.txt is an emerging convention: a plain-text file at your domain root that points AI tools to your most important, cleanest content. Adoption is early and no major engine guarantees support in 2026, so treat it as low-cost insurance rather than a ranking factor. Add it, keep it accurate, and do not expect it to replace solid HTML and schema.

Brand mentions across the web carry more weight. Models build a sense of which entities are credible from how often, and how consistently, your business is named on other reputable sites, directories and reviews, even without a link. Consistent mentions on the FSB directory, trade bodies, local press and review platforms strengthen the association the model makes between your brand and your specialism.

How GEO differs from classic SEO

GEO differs from classic SEO in its target and its unit. Classic SEO optimises a page to rank for a query and earn a click; GEO optimises a passage to be extracted and attributed inside an answer the user may never click through. The two share a technical foundation but reward different writing, structure and measurement.

What changes, and what stays the same?

What stays the same: crawlability, indexing, page speed, helpful original content and a credible domain. None of that goes away, because retrieval still runs on the search index. If your fundamentals are weak, no amount of GEO polish helps, which is why GEO and SEO budgets sit together rather than compete. Our breakdown of what SEO costs in the UK still frames the spend you should expect.

What changes: the writing leans harder on answer-first structure, self-contained passages and explicit facts, because the model extracts rather than ranks. Measurement changes too; you track citations, referral traffic from AI tools and branded query growth, not just position. The biggest shift is mindset, treating each section as a standalone answer the machine can quote, not a paragraph that only works as part of a whole.

Will AI search reduce your website traffic?

It can, and UK businesses should plan for it. When an AI Overview answers a question fully, some users never click. The defensive play is to be the cited source so your brand is at least named, and to build content for queries where users still want to act, compare or buy rather than simply learn a fact. The ONS publishes UK internet and device usage data in its IT and internet industry statistics, useful context when you model how behaviour is shifting.

A practical GEO checklist for a UK SMB

A practical GEO checklist for a UK SMB starts with access, moves to structure, then to entity signals. Work through it in that order, because there is no point polishing prose the engine cannot reach. Most UK small businesses can complete the first pass in a focused week, then refine page by page.

What should you do first, second and third?

First, fix access. Confirm robots.txt allows OAI-SearchBot, PerplexityBot and Bingbot, check your host or Cloudflare is not silently blocking them, and make sure pages render their content in HTML rather than only via JavaScript. Without this, nothing else counts because your pages never enter the candidate set the model reads.

Second, fix structure. Give each key page a question-style heading with a 40 to 60 word direct answer beneath it, break long sections into self-contained paragraphs, add a genuine FAQ, and use tables for prices, comparisons and specifications. Third, fix entity signals: add Organization, Article and FAQPage schema, name a real author, and make your business name and address identical everywhere it appears online, from your homepage to directories. If you would rather hand this to specialists, our content writing service builds pages to exactly this pattern.

If you want your business named in AI answers rather than buried beneath them, the work is specific and learnable: answer-first structure, clean schema, the right crawler access and content worth quoting. We help UK SMBs do exactly that through our content writing service, and you can talk through your site with us directly via our contact page.