How to optimise your site for AI citations

Key takeaways

• AI citation optimisation — often called Generative Engine Optimisation (GEO) — refers to the practice of making your content findable and quotable by AI platforms like ChatGPT, Gemini, and Perplexity, each of which uses distinct retrieval logic.
• Only 11% of cited domains appear across all three major AI platforms; optimising for one platform alone surrenders the other two.
• Adding inline citations to factual claims is the single highest-leverage on-page change, lifting AI visibility by up to 40% according to the Princeton/Georgia Tech GEO benchmark (KDD 2024).
• AI crawlers cannot execute JavaScript; serving clean, server-rendered HTML via edge interception solves the parse-quality problem at the source.
• Citation share shifts in weeks, not months — Reddit's ChatGPT citation share fell from ~60% to ~10% in six weeks after a single parameter change in late 2025.

What is the AI citation gap?

The AI citation gap refers to the disconnect between a site's organic search presence and its visibility inside AI-generated answers. Most sites have no measurable AI citation presence, even when they rank well on Google. The traffic numbers, however, are hard to ignore. According to a 12-month, 94-brand GA4 study by Visibility Labs, ChatGPT referral sessions convert 31% higher than non-branded organic — a premium that compounds quickly for high-consideration SaaS purchases. We've seen LLM visitors perform at 4.4x the organic conversion rate on comparable traffic bases. The channel is early (AI referrals are still below 0.2% of total traffic for most sites), but it arrives pre-qualified in a way organic search rarely does.

The concentration problem is more alarming. The 5WPR AI Platform Citation Source Index 2026, synthesising 680 million citations, found the top 15 domains capture 68% of all consolidated AI citation share. Google's PageRank distribution, notorious for its long-tail inequality, was never this steep. A small number of domains have hardened their positions across all three platforms. The gap between them and everyone else is widening.

Therefore, the actionable reading of this data is clear: there is no monolithic AI search to optimise for. There are three platforms, each with distinct retrieval logic, each pulling from partially non-overlapping source pools. This playbook maps both the platform-specific authority work that gets you into the retrieval pool, and the on-page technical layer that converts a crawl visit into a citation.

What do ChatGPT, Gemini, and Perplexity trust as sources?

Each AI platform trusts different source types: Gemini favours brand-owned websites, ChatGPT leans on third-party directories, and Perplexity rewards freshness and named authorities. The Yext analysis of 6.8 million citations (2025) provides the clearest per-platform breakdown available.

Gemini: structured, schema-marked first-party content

Gemini behaves most like traditional search. According to the Yext study, 52.15% of its citations come from brand-owned websites. It favours structured, schema-marked, first-party content and rewards the same editorial signals that have mattered in Google ranking for a decade. If you have strong organic SEO, Gemini is your lowest-friction citation target.

ChatGPT: third-party directories and Wikipedia

ChatGPT is different in two important ways. First, it leans heavily on third-party and directory sources: 48.73% of its citations come from sites like Yelp, TripAdvisor, and similar aggregators, per the same Yext study. Additionally, 28.3% of ChatGPT's most-cited pages have zero organic visibility in Google — the two rankings are genuinely partially decoupled. Wikipedia dominates its top-10 cited sources at between 26% and 48% share depending on query category, per the 5WPR Index. For B2B SaaS, the implication is clear: Wikipedia presence and third-party directory coverage are retrieval infrastructure, not vanity metrics.

Perplexity: freshness and primary-source authority

In contrast, Perplexity rewards freshness and primary-source authority more aggressively than the other two. Established publishers capture 22–35% of Perplexity citations, and the platform skews toward named B2B authorities and recently updated content. Unlike Gemini, which is patient with well-structured evergreen content, Perplexity will deprioritise pages that haven't been touched recently regardless of their original quality.

One cross-platform finding worth holding: a page at position 1 in Google has a 58% probability of being cited by AI; by position 10, that drops to 14% (Growth Memo, April 2026). SEO foundation is not sufficient for AI citation — ChatGPT's non-Google-visible pages prove that — but it remains necessary. Think of strong organic rankings as a prerequisite that raises the ceiling, not a guarantee of citation.

How do you give AI crawlers access to your site?

The first step is to explicitly allow each AI crawler in your robots.txt file, then verify through server logs that those bots are actually visiting. Every content and authority strategy is moot if AI crawlers cannot reach and parse your pages — and this fails more often than most teams realise.

Why crawler access matters now

According to Cloudflare's 2025 data, GPTBot's share of all AI crawler traffic surged from 5% to 30% between May 2024 and May 2025, with overall AI crawling up 32% year-on-year. The volume is significant enough to appear in server logs without hunting for it. Roughly 80% of AI crawler activity is model training, 18% is search indexing, and 2% is user-initiated retrieval. The citation-relevant bots — OAI-SearchBot, PerplexityBot, Google-Extended — fall into that indexing slice.

How to configure robots.txt for AI crawlers

Start with robots.txt. Check that GPTBot, OAI-SearchBot, ChatGPT-User, ClaudeBot, PerplexityBot, and Google-Extended are each explicitly listed as Allow. They are independently controllable, and a blanket disallow on AI crawlers (common in older security-first configs) blocks all of them at once. Next, review server logs to confirm which bots have visited recently.

Solving the parse-quality problem

The parse-quality problem is separate from access. AI agents cannot execute JavaScript. As a result, JS-heavy pages, cookie consent walls that render before content, and navigation menus that load via client-side fetch all degrade what a crawler extracts — even when the bot reaches the page. Carousels, tabbed content, and dynamically loaded testimonials are invisible to the retrieval layer. The practical fix is to serve clean, server-rendered HTML to AI user-agents via edge interception or static alternatives.

How to measure AI referral traffic

On measurement: since June 2025, ChatGPT appends utm_source=chatgpt.com to citation links. Without a dedicated GA4 channel grouping, this traffic disappears into the generic Referral bucket. Set up a custom AI Traffic channel using regex across chatgpt.com, perplexity.ai, claude.ai, gemini.google.com, and openai.com. The data you cannot see you cannot improve.

How should you write content for AI retrieval?

The most effective way to write for AI retrieval is to add inline citations to every factual claim, structure content with clear headings and short paragraphs, and update pages frequently. The Princeton and Georgia Tech GEO study — a 10,000-query benchmark published at KDD 2024 — documented that specific textual optimisations can lift AI visibility by up to 40%.

More precisely, pages ranked fifth in traditional search saw a 115.1% visibility lift in AI responses after proper inline citations were added. Adding sourced claims — not just making them — is the single highest-leverage on-page change documented in the academic literature.

"AI search shows systematic and overwhelming bias towards earned media over brand-owned content."

The practical read of that finding: a single authoritative mention in a recognised trade publication likely outweighs ten well-optimised blog posts on your own domain. Content strategy and PR are no longer separate budgets.

Why inline citations matter

The mechanism is straightforward. AI retrievers pattern-match to trustworthy content. A page that cites its sources inline, uses authoritative named references, and structures claims with evidence is a closer match to the training distribution of trusted documents. Therefore, write factual claims with their source linked inline — not at the end of the page in a bibliography nobody reads.

Why freshness matters more for AI than traditional SEO

Freshness matters more for AI retrieval than for traditional SEO. According to Position Digital's analysis, AI platforms cite content that is 25.7% fresher than what traditional search surfaces, and 76.4% of ChatGPT's most-cited pages were updated within the last 30 days. A well-structured page untouched for eight months is at a structural disadvantage against a thinner page updated last week.

Format for scanners, not just readers

AI retrievers extract meaning the same way a fast human skim does: H2 and H3 headings, numbered lists, short paragraphs, defined terms. A page structured as flowing prose with no subheadings requires more inference from the retrieval model. Give it the structure instead.

What are machine-readable parallel pages?

Machine-readable parallel pages are stripped, server-rendered versions of your key product, comparison, and use-case pages — served to AI crawlers via edge interception instead of your standard JavaScript-heavy front end. They are not separate websites. They are the same content, rendered without JS dependencies, cookie walls, or navigation chrome, and routed to AI user-agents at the CDN layer.

This approach solves what most optimisation guides miss. Instead of hoping crawlers can extract meaning from your existing markup, you control exactly what they retrieve. A clean, distraction-free version of your core pages increases citation probability because the retriever has better signal to work with. No guessing about whether JavaScript loaded in time. You serve the content in the form that matters most.

For teams thinking beyond citation to AI agent action — the scenario where ChatGPT or a browser agent navigates your site on behalf of a user to compare plans or start a trial — the next layer is agentic page formatting. Agentic page formatting refers to the practice of adding labelled actions, structured metadata, and machine-interpretable CTAs that a browsing agent can act on without human supervision.

What are five AI citation fixes you can make this week?

The five highest-impact, lowest-effort changes are: auditing your robots.txt, setting up AI traffic tracking, adding inline citations, planning static alternatives for JS-heavy pages, and refreshing your top pages. A focused team can move through all five in five working days.

1. First, audit robots.txt today. Confirm GPTBot, OAI-SearchBot, ChatGPT-User, ClaudeBot, PerplexityBot, and Google-Extended are each explicitly listed as Allow. Then pull server logs for the last 30 days and check whether each has actually visited. The gap between "not blocked" and "actively crawling" is worth knowing.

2. Next, add a custom AI Traffic channel in GA4. Use a regex source filter across chatgpt.com, perplexity.ai, claude.ai, gemini.google.com, and openai.com. Without this, every ChatGPT citation click after June 2025 lands in your generic Referral bucket and the channel is unmeasurable.

3. Then, add inline citations to your three highest-traffic pages. For every factual claim, link the source inline — not in a footnote, not in a "Sources" section. The GEO benchmark (KDD 2024) documented up to a 40% visibility lift from this change alone. It requires no engineering, no redesign, and takes a single afternoon.

4. Additionally, map your most JS-heavy commercial pages and plan static alternatives. Identify which product and comparison pages return empty or degraded content when fetched without JavaScript. These are your silent citation killers. If you run Cloudflare, an edge worker can intercept AI user-agents and serve a stripped parallel page without touching the existing stack.

5. Finally, refresh your top 10 pages with recent updates. Pull your analytics for pages that receive AI traffic today. Timestamp each with a recent edit — even a minor update to a date, a new example, or a refined explanation. The signal matters more than the scale of the change.

Where is AI citation share heading?

AI citation share is heading toward greater volatility in the short term and a structural shift from citation to agent-driven action in the medium term. The instability is the part most teams underestimate. According to the 5WPR Citation Source Index, Reddit's citation share in ChatGPT fell from roughly 60% to roughly 10% in six weeks in late 2025 following a single parameter change. Citation share is now measured in weeks, not months.

The decline in traditional search

The macro trend running beneath the volatility is not ambiguous. Gartner forecast in February 2024 that traditional search engine volume would drop 25% by 2026 due to AI chatbots — and that forecast appears conservative given that Google organic CTRs fell 41% across the board in 2025, according to Position Digital's data. Whether the precise number is 25% or more, the direction is not reversible.

The shift from citation to agent action

The next structural shift is from AI citation to AI action. Browser agents that navigate, compare, and transact on behalf of users are already in early deployment. A site with machine-readable agentic pages — structured metadata, labelled actions, content that an agent can parse and act on — will be navigable by those agents. In contrast, a site built only for human browsers will be skipped in the same way a JS-heavy page is skipped by today's citation crawlers.

As a result, the brands compounding now in AI citation share are not winning on infrastructure alone. The durable leaders are winning on genuine topical authority, consistent external mentions, and structured content that retrieval models can trust. The technical work is the floor — the prerequisite for being in the retrieval pool at all. What determines whether you stay there is whether you've earned the kind of external, third-party attestation that no robots.txt configuration can manufacture.

Frequently asked questions

What is the difference between GEO and traditional SEO?

Generative Engine Optimisation (GEO) refers to making content citable by AI platforms like ChatGPT, Gemini, and Perplexity, while traditional SEO focuses on ranking in Google's link-based results. According to the Princeton/Georgia Tech GEO benchmark, 28.3% of ChatGPT's most-cited pages have zero Google organic visibility — which means the two ranking systems are partially decoupled and require distinct strategies.

Do I need to optimise separately for each AI platform?

Yes. Only 11% of cited domains appear across ChatGPT, Gemini, and Perplexity simultaneously. Gemini favours brand-owned, schema-marked pages (52.15% of its citations). ChatGPT pulls 48.73% of citations from third-party directories. Perplexity rewards freshness and primary-source authority. A single approach covers, at best, one platform.

How often should I update pages to maintain AI citation eligibility?

Aim for updates at least every 30 days on high-priority pages. Position Digital's data shows 76.4% of ChatGPT's most-cited pages were updated within the last 30 days, and AI platforms cite content that is 25.7% fresher than what traditional search surfaces. Even minor edits — a revised date, a new example — send a freshness signal.

Can a JavaScript-heavy site still earn AI citations?

Not reliably. AI crawlers like GPTBot and PerplexityBot cannot execute JavaScript, so carousels, tabbed content panels, and dynamically loaded sections are invisible to them. The fix is to serve clean, server-rendered HTML to AI user-agents via edge interception (for example, a Cloudflare edge worker) so the retriever gets a complete document rather than a partially rendered DOM.

Is earning a mention in a trade publication more valuable than optimising my own blog?

For AI citation purposes, yes. The September 2025 arXiv GEO paper found AI search shows "systematic and overwhelming bias towards earned media over brand-owned content." A single authoritative mention in a recognised trade publication likely outweighs ten well-optimised blog posts on your own domain, making PR and content strategy inseparable for AI visibility.