How AI search engines choose who to cite —
and how to get on the list.

How AI search actually works — the part most guides skip

When someone types a question into Google, Perplexity, or ChatGPT in 2026, they no longer just see a list of links. They see a synthesised answer — generated by a large language model that has read, processed, and condensed information from across the web. That answer often cites sources. Sometimes it doesn't. Understanding why requires understanding how each engine is built.

There are two fundamentally different types of AI search response:

• Retrieval-augmented generation (RAG) — the AI retrieves live web pages at query time, reads them, and synthesises an answer. This is how Perplexity, ChatGPT with browsing, and Google AI Overviews work. Fresh pages can be cited almost immediately after indexation.
• Parametric knowledge — the AI draws on information baked into its model during training. This is how ChatGPT (without browsing) and Claude answer questions. You cannot influence these answers in real time — only through long-term brand and citation building that shapes future training data.

For most UK businesses trying to get cited in AI answers today, RAG-based systems are the primary target. That means Google AI Overviews, Perplexity, and ChatGPT's browsing mode — all of which index and retrieve live web content.

The 7 signals AI search engines use to select citations

1. Existing Google rankings

For Google AI Overviews, the single strongest predictor of being cited is already ranking in the top 10 organic results for the query. Google AI Overviews overwhelmingly pull from pages already on page one — because those pages have already passed Google's relevance and authority assessments. If you are not ranking organically, your path to AI Overview citation runs through improving your organic position first.

2. Direct question-answer structure

AI systems are optimised to find and extract direct answers to specific questions. Pages that clearly state a question and then immediately answer it — in the first sentence, not buried in paragraph five — are significantly more likely to be extracted and cited. The inverted pyramid structure (most important information first) that good journalism uses is also optimal for AI citation.

3. Schema markup

FAQPage schema tells an AI system: "this section contains questions and answers." HowTo schema tells it: "this section contains step-by-step instructions." Article schema provides authorship, publication date, and topic context. These structured data signals reduce the computational effort required for an AI to parse and use your content — making citation more likely. Pages without schema are not excluded from citations, but pages with well-implemented schema have a meaningful advantage.

4. E-E-A-T signals

Experience, Expertise, Authoritativeness, and Trustworthiness — Google's quality evaluation framework — is applied to AI citations as much as organic rankings. AI systems favour content from organisations or individuals that demonstrate genuine expertise through: named authors with credentials, first-hand experience claims, citations from other authoritative sources, and consistent publishing history in a subject area. A page on dental care written by a named dentist outcompetes a generic health content farm — in both organic and AI results.

5. Topical authority depth

A single well-optimised page can win a citation for one query. But consistent citation across a topic area requires topical authority — a body of content that covers a subject comprehensively, with internal links between related pages, and a track record of being referenced by other sites. AI systems are increasingly good at identifying "domain experts" based on the breadth and depth of coverage across an entire site, not just individual pages.

6. Content freshness and accuracy

AI systems — especially RAG-based ones — prioritise recently updated content for fast-changing topics. If your competitor updated their guide on AI search last month and yours is from 2022, theirs will be preferred. For evergreen topics, freshness matters less than depth. For regulatory, statistical, or fast-moving subjects (tech, finance, law), regular content updates are essential for maintaining AI citations.

7. Crawlability and technical accessibility

An AI system cannot cite what it cannot read. Pages blocked by robots.txt, behind login walls, with broken canonical tags, or with very slow load times are at a disadvantage. Google's AI Overviews crawler and Bing's AI-serving infrastructure both require clean, crawlable, fast-loading pages to reliably include content in generated answers.

Engine-by-engine: how each platform selects citations

Google AI Overviews

AI Overviews appear above organic results for a growing proportion of UK queries — predominantly informational, how-to, and comparison queries. Google's system pulls primarily from pages already in its index that rank highly for the query. Key optimisation levers: organic ranking improvement, FAQPage schema, clear H2/H3 question-based headings, and concise paragraph-opening answers. AI Overviews tend to synthesise from 3–8 sources per query.

Perplexity

Perplexity is a RAG-first engine — it retrieves live web pages for every query. It has its own crawler (PerplexityBot) but also draws from Bing's index. Perplexity tends to cite more sources than Google AI Overviews (often 5–15 per answer) and favours pages that are authoritative, well-structured, and on high-quality domains. Optimisation approach: ensure PerplexityBot is not blocked in robots.txt, ensure Bing indexation, and focus on structured, cited content (pages that reference statistics and studies).

ChatGPT with web browsing

When ChatGPT browses the web, it uses Bing's search index as its primary source. This means that Bing SEO — traditionally neglected by most UK marketers — is now a direct input to ChatGPT citation frequency. Pages well-optimised for Bing (which shares most ranking signals with Google but has meaningful differences) are far more likely to appear in ChatGPT browsing responses. Submit your sitemap to Bing Webmaster Tools, check your Bing coverage, and ensure GPTBot is not blocked.

Microsoft Copilot

Copilot is built on Bing's index and OpenAI's models. It has high penetration through Windows 11 and Microsoft Edge — making it disproportionately important for queries from PC users, a significant share of UK B2B search. Same optimisation path as ChatGPT: Bing indexation is the prerequisite.

Google Gemini

Gemini draws from Google's index and is increasingly integrated into Google Search results. The citation signals overlap heavily with AI Overviews. Focus on the same foundations: organic ranking quality, schema markup, E-E-A-T, and direct question-answer content structure.

Schema markup for AI citation: a practical checklist

Schema markup tells AI systems exactly what your content is, who produced it, and what questions it answers. The following schema types have the highest impact on AI citation frequency:

• FAQPage — mark up every Q&A section. Each question becomes a discrete, extractable unit for AI systems. This is the highest-impact schema for AI citation.
• HowTo — for any step-by-step content. AI systems extract and display HowTo steps directly in answers.
• Article — include author (with Person schema and credentials), datePublished, dateModified, and headline. Signals freshness and authority.
• Organization — site-wide schema declaring who you are, your URL, social profiles, and contact. Builds entity recognition across AI systems.
• Service — for service pages, declares what you offer, to whom, and at what price. Improves commercial query citation.
• BreadcrumbList — clarifies page hierarchy and topic context.
• SpeakableSpecification — marks content as appropriate for voice and AI reading. Signals to Google which parts of your page are most citeable.

Content structure that maximises AI citability

The way you structure content has direct impact on whether AI systems can extract and cite it cleanly. High-citation content shares these structural characteristics:

• Question-as-heading, answer-as-first-sentence — use H2 or H3 headings that are literal questions ("How does X work?"), then open the paragraph with a direct, concise answer before expanding with context.
• Short answer first, detail second — the inverted pyramid. AI systems extract the first sentence or two of a paragraph most reliably.
• Numbered or bulleted lists for multi-part answers — lists are easier for AI to extract and reformat than prose paragraphs for enumerative content.
• Defined terms and clear entities — name the specific tools, platforms, regulations, and organisations you reference. AI systems anchor citations to named entities.
• Statistics with clear attribution — AI systems cite pages that contain cited, verifiable data. Include statistics with their source year and origin.
• Summary sections and TL;DR boxes — some AI systems preferentially extract clearly labelled summary content.

Building topical authority for sustained AI citations

A single well-optimised page can win a citation. Sustained citation across a topic area — the kind that builds brand visibility at scale — requires topical authority. This means:

• Content clustering — a pillar page covering a topic broadly, supported by multiple cluster pages going deeper on specific sub-topics. Internal links between all cluster pages signal topic comprehensiveness to both search engines and AI systems.
• Consistent publishing cadence — AI systems tracking freshness reward sites that regularly update and expand content in a topic area. Monthly content publication in your target topic cluster is a minimum baseline.
• External citations and brand mentions — when other authoritative sites link to or mention your content, AI systems register your content as more trustworthy. Digital PR, guest content, and industry directory listings all contribute.
• Author entity building — linking author bios to structured data, LinkedIn profiles, and published content elsewhere builds the author entity that AI systems use to assess expertise.

Technical foundations: what must be in order before anything else

No amount of content optimisation will drive AI citations from pages that AI crawlers cannot reliably access. Audit these technical points first:

• robots.txt — verify GPTBot, PerplexityBot, Google-Extended, and ClaudeBot are not blocked unless you have a specific reason to do so. Check with: yoursite.com/robots.txt
• Sitemap submission — submit your XML sitemap to both Google Search Console and Bing Webmaster Tools. Bing submission is critical for ChatGPT and Copilot citations.
• Page speed — pages loading in over 3 seconds are crawled less frequently and indexed less reliably. Aim for Core Web Vitals pass on mobile and desktop.
• Canonical tags — ensure self-referencing canonicals on all key pages. Duplicate content without clear canonicals confuses AI systems about which version to cite.
• HTTPS — non-HTTPS pages are not cited by major AI systems. Ensure valid SSL across all pages.
• Mobile-first rendering — Google and Bing both crawl primarily via mobile user-agent. Ensure your pages render fully on mobile without content hidden behind JavaScript that crawlers can't execute.

How to track and measure AI citation frequency

Measuring AI citations is less mature than traditional rank tracking, but improving rapidly. Current approaches:

• Manual monitoring — search your key target queries in Google (with AI Overviews enabled), Perplexity, and ChatGPT. Record whether you appear and which pages are cited. Weekly spot-checks build a citation frequency picture over time.
• Google Search Console — AI Overview impressions are increasingly included in GSC data as a distinct appearance type. Monitor impressions and clicks for queries where you know AI Overviews trigger.
• Third-party tools — tools including SE Ranking, Semrush, and BrightEdge now offer AI visibility tracking. These monitor your citation frequency across AI engines for tracked keyword sets.
• Direct traffic and brand search trends — AI citations drive brand awareness. An increase in direct traffic or branded search volume often correlates with increased AI citation presence, even in the absence of direct click data.

The businesses winning AI citations in 2026 are not the ones with the biggest budgets — they are the ones with the clearest, most structured, most authoritative content on the specific questions their audience is asking. Start with a content audit, identify your highest-value query clusters, implement schema, and build out your topic authority. The AI citation layer is early enough that consistent effort over 6–12 months can establish a durable competitive position.