AI search optimization in 2026: how to get cited in ChatGPT, Perplexity, Claude and Google AI Overviews

Why AI search is different from SEO

Classic SEO optimises a page to rank. AI search optimisation (AISO, or GEO — generative engine optimisation) optimises a passage to be quoted. The unit isn't the page any more — it's the paragraph or the bullet list that answers the question directly. Rank a page and you compete with nine other blue links. Get quoted and you're the only answer the user sees.

The four AI engines that matter and how they source

Each engine sources differently. Optimise for the pattern, not the brand — because the pattern is stable even as the leaderboard shuffles.

• 01ChatGPT search — retrieves from the live web, favours authoritative pages with clean HTML and clear structured answers. Cites 3–6 sources per answer.
• 02Perplexity — retrieves aggressively, cites 5–10 sources, favours pages with explicit factual claims and clear headings.
• 03Claude with web search — favours long-form authoritative content with visible expertise and specific data.
• 04Google AI Overviews — sources primarily from pages that already rank on page one, with a heavy weight on FAQ and HowTo schema.

The one paragraph that gets cited

Look at any AI answer and you'll see the same pattern in the passages it quotes: a direct question in a heading, followed by a direct answer in the first sentence, followed by 2–4 sentences of substantiation with specific numbers or names. That structure — question, direct answer, evidence — is the single highest-leverage change you can make. Rewrite the first paragraph under every H2 to answer the H2 as a question, and citations start appearing within weeks.

Structured data: FAQ, HowTo, Article, Product

AI engines lean heavily on structured data because it removes ambiguity. FAQPage JSON-LD tells the engine 'this is a question and this is the answer' — no parsing required. HowTo schema turns a process into a citable sequence. Article schema with named author, publish date and modified date signals authority. Product schema with price and availability makes commercial pages quotable in shopping-adjacent queries.

The E-E-A-T signals AI engines weight most heavily

Google's Experience, Expertise, Authoritativeness, Trust framework applies double to AI search. A page with no author, no date and no citations gets skipped in favour of one that has all three — even if the anonymous page ranks higher in classic search. Name the human. Show the credential. Cite the source. Date the content.

• 01A visible author byline with a real name and job title on every article.
• 02Author schema (Person) with URL back to a full about/author page.
• 03Published and modified dates in visible text AND in Article schema.
• 04External citations to primary sources — ONS data, gov.uk, industry bodies — with real links.
• 05Case studies with named clients and specific outcomes, not 'a leading UK firm'.

Question-led headings beat keyword-led headings

'H2: Website cost UK' is a keyword heading. 'H2: How much does a website cost in the UK in 2026?' is a question heading. AI engines pattern-match user queries against question headings before they even read the paragraph beneath. Rewriting your top ten H2s from keyword form to question form is a 30-minute job and it doubles citation rates on most sites we audit.

Specificity beats generality every time

AI engines are trained to prefer specific, verifiable claims over vague ones. 'Websites take a few weeks to build' is not citable. 'A five-page small business website typically takes 3–4 weeks from brief to launch in the UK in 2026' is citable. Numbers, dates, countries, currencies, named tools — every specific detail is a hook the engine can grab and quote.

llms.txt and the emerging AI crawler standard

A file at /llms.txt gives AI crawlers a curated tour of your site — the pages you want them to weight most heavily. It's the AI equivalent of a sitemap for humans. Not every engine reads it yet, but the ones that do (Perplexity, ChatGPT search) treat it as high-signal. It costs nothing to publish and takes fifteen minutes to write.

Speakable schema for voice AI

As voice-first assistants (Alexa, Siri, Gemini, ChatGPT voice) start pulling from web content, Speakable JSON-LD marks up the passages you want read aloud. Add it to your top FAQ answers and your H1+intro paragraph. It's a low-effort, high-optionality bet that costs nothing to add now and could pay off dramatically over the next two years.

The measurement problem — and how to solve it

Google Analytics doesn't tell you when ChatGPT cited you. But three signals are easy to track: direct traffic spikes on long-tail pages, referral traffic from perplexity.ai / chat.openai.com / claude.ai in your analytics, and manual query testing — literally typing the questions your customers ask into each engine and screenshotting the citations. Do the third one monthly and you'll see the shape of your AI visibility even without perfect analytics.

The 90-day AI search playbook

Month one: rewrite the top ten H2s across your site as questions, and rewrite the first paragraph under each to lead with the direct answer. Month two: add FAQPage schema to every service and pricing page, and add visible bylines with dates on every article. Month three: publish /llms.txt, add Speakable schema to your top pages, and start a monthly citation-testing routine across the four major engines. Ninety days is enough to see the pattern; six months is enough to dominate a niche.