Ask an AI for the best studio, the right tool or where to stay, and it answers in one paragraph that names a few sources. Being one of those names is now a discipline with measurable rules. Here is what the research shows actually works, and what is hype.
The short answer
To get cited by AI, structure each page so a model can lift a confident, self-contained claim: answer the question in the first two sentences under every heading, back claims with named statistics and expert quotations, and earn authority the model has already seen elsewhere. A 2024 academic study (Aggarwal et al., presented at KDD) measured these exact tactics and found they lift visibility in generative engines by up to roughly 40%.
Why being the cited answer matters now
- Zero-click is the norm: in 2026, about 68% of Google searches in the US end without a single click to the open web (SparkToro, on Similarweb panel data).
- AI answers are eating the results page: Google's AI Overviews grew from roughly 6% of tracked queries in early 2025 to between a quarter and nearly half by 2026, depending on the study (Semrush; Ahrefs/BrightEdge).
- When an AI Overview appears, the click roughly halves: Pew Research measured click-through to results falling from about 15% to about 8%.
- The audience is already massive: ChatGPT alone reported around 900 million weekly active users in early 2026.
What actually moves AI citations
The most rigorous public evidence is the GEO study (Aggarwal et al., KDD 2024), which tested optimization tactics across 10,000 queries and validated them live on Perplexity. The pattern is clear: models reward content that is quotable, specific and well-sourced, not keyword-stuffed.
| Tactic | Measured visibility uplift |
|---|---|
| Add quotations from credible experts | up to +43% |
| Add relevant, specific statistics | up to +33% |
| Improve fluency and clarity | up to +29% |
| Cite named, credible sources | up to +28% |
These are uplifts from a baseline, not guarantees, and they compound when combined. Tested live on Perplexity, the same study saw a 37% gain in how favorably the engine presented the source.
The llms.txt myth, told straight
You will be told an llms.txt file makes AI crawlers read your site. The honest 2026 picture: Anthropic (Claude) and Perplexity document support for it, but Google has publicly said it does not use llms.txt and has no plans to (Gary Illyes, 2025), and OpenAI has never confirmed it. Independent measurement puts real llms.txt fetches at a tiny fraction of AI-bot traffic. Ship the file, it is cheap and harmless, but do not treat it as the thing that earns citations. Structure, clarity and authority do.
The GEO playbook we actually run
- Audit first: see exactly how ChatGPT, Perplexity, Gemini and Google AI Overviews describe you today, and for which queries.
- Answer-first structure: every key page leads with a direct, self-contained answer, then the detail.
- Quotable facts: named statistics, expert quotations and clear definitions a model can lift verbatim.
- Machine-readable proof: Schema.org structured data and FAQ markup so facts are unambiguous.
- Crawler access: explicitly allow GPTBot, ClaudeBot and PerplexityBot, and serve real server-rendered HTML, since some AI crawlers do not run JavaScript.
- Authority over time: a consistent, cited presence the models learn to trust.
How to measure it
Track whether and how often each engine cites you for your priority questions, across ChatGPT, Perplexity, Gemini and AI Overviews, and watch the trend as foundations and content ship. GEO is a 6 to 12 month compounding curve: technical fixes land fast, but durable citations are earned, not bought, and no honest provider can guarantee a specific mention.
FAQ
- Do the +40% numbers mean guaranteed results?
- No. They are average uplifts from a controlled study (Aggarwal et al., KDD 2024), not promises. They show which tactics reliably help; actual outcomes vary by engine, query and competition.
- Is llms.txt worth adding?
- Yes, as a cheap baseline: Anthropic and Perplexity use it. But Google has said it does not, and OpenAI has not confirmed, so it is not a primary lever. Structure, quotable facts and authority do the heavy lifting.
- How long before AI engines cite me?
- Expect a 6 to 12 month compounding curve. Technical foundations land in weeks; consistent citations across engines take sustained content and authority work.