GEO: How to Get Cited by ChatGPT, Perplexity and Google AI Overviews

Your prospects can now ask an assistant instead of a search engine: "which firm should I hire for…", "how much does… cost". The answer cites a handful of sources; if your website is not one of them, you are not part of the conversation. This guide reflects the state of play on 28 September 2026 and relies only on primary sources: official vendor documentation, the paper that coined the term, the llms.txt proposal and RFC 9309 (listed at the end).

What Is GEO?

The term comes from the research paper "GEO: Generative Engine Optimization" (Pranjal Aggarwal and five co-authors), posted on arXiv in November 2023 and presented at KDD 2024. It means optimising content to be more visible in the answers of "generative engines", tools that retrieve pages and then write a synthesis citing some of them. On their benchmark (GEO-bench), the authors tested nine rewriting methods:

• Citing sources, adding quotations and adding statistics worked best, with a 30 to 40% relative gain in visibility (word count of the answer sentences that cite the site, weighted by citation position).
• Keyword stuffing brought little to no improvement, and even lost 10% visibility on Perplexity.ai.
• Lower-ranked websites can gain a lot: +115.1% visibility with "cite sources" for sites ranked fifth. On Perplexity.ai, adding quotations brought +22%.

These are lab results on 2023-era engines: keep the direction (verifiable, sourced facts get reused more), not the percentages.

GEO vs SEO

GEO does not replace SEO. Google states that no special optimisation is needed to appear in AI Overviews or AI Mode: the page must be indexed and eligible to show a snippet. What changes is the unit of visibility: a passage reused inside an answer, rather than a position in a list of links.

How AI Engines Choose Their Sources

An assistant answers either from its internal knowledge, learned in training from pages collected by crawlers such as GPTBot or ClaudeBot (frozen at a date, usually without a cited source, updated only with the next model), or through a live web search: it queries an index, reads a few pages and cites them. GEO works on that second mechanism in the short term. ChatGPT, Claude and Perplexity run their own search crawlers; AI Overviews and AI Mode rely on the Google Search index, which the Gemini app can also use to ground its answers. Google adds that its AI features may run several related searches across subtopics ("query fan-out"), so covering a topic's sub-questions gives you more chances to be picked. Bing Webmaster Tools now reports your citations in Copilot.

On-demand agents (ChatGPT-User, Claude-User, Perplexity-User) visit a page when a user asks. OpenAI says robots.txt rules may not apply to ChatGPT-User, Perplexity says Perplexity-User generally ignores them, and Anthropic says Claude-User honours them.

Practical Levers for an SMB

1. Let AI Search Crawlers In

Check your /robots.txt. Under the standard (RFC 9309), a crawler follows the group that names it (ignoring the User-agent: * group) or, if there is none, the User-agent: * group, so a rule for GPTBot does not affect OAI-SearchBot. This example opts out of model training by OpenAI, Anthropic, Google and Apple without leaving ChatGPT, Claude, Perplexity or Google Search (training crawlers from other companies, not listed here, remain allowed):

User-agent: GPTBot
User-agent: ClaudeBot
User-agent: Google-Extended
User-agent: Applebot-Extended
Disallow: /

Check your CDN or firewall too: bot protection can contradict your robots.txt. On Google, avoid accidental nosnippet, max-snippet or noindex directives, which limit a page's use in AI features. For Bing and Copilot, Microsoft recommends IndexNow to flag every page added, updated or removed. And OpenAI says it can take about 24 hours for a robots.txt change to be reflected in ChatGPT search results.

2. Write Passages That Are Easy to Quote

An answer engine extracts a passage, not a whole page. Write for that (see also our SEO copywriting guide):

• A 40 to 60-word answer paragraph at the top of each key page: who, what, for whom, how much, how long, where.
• Sourced figures and precise, dated facts: citing sources and adding statistics were among the most effective methods in the GEO paper, and Microsoft notes that examples, data and cited sources build trust.
• Clear headings, tables and FAQs, which Microsoft lists as easier for AI systems to reference accurately.
• One page per question that matters (pricing, timelines, comparisons), updated regularly.

3. Present One Consistent Entity

An AI must recognise a business before recommending it. Keep your name, address, phone, prices, founder and description identical on your website, Google Business Profile, LinkedIn and directories: an old price left on a profile can be repeated as is. Express these facts as structured data: Organization with sameAs links to your profiles (Google says this helps it understand an organisation and disambiguate it), plus Person, Service and FAQPage, all matching the visible text. Google also states that no special schema.org markup is required for its AI features: it is a clarity tool, not a switch. More on structured data and Google Business Profile.

4. Be Present Where AI Engines Look

For each question you track, note which sites the engines cite: directories, comparison sites, trade media, local press, professional bodies. If a comparison site keeps coming up and you are not on it, getting listed with accurate details may matter more than another blog post. Bing's new "Citation Share" metric (your share of all citations shown for a given grounding query, the query the AI used to retrieve its sources) reflects this logic. Earn mentions like good links: guest articles, partners, recognised directories, interviews.

5. llms.txt: A Bonus, Not a Guarantee

The /llms.txt file is a proposal published in September 2024 by Jeremy Howard and revised (v2) in August 2026: a Markdown file placed at the site root or in a subfolder it describes (a required title, then an optional summary and links to key pages) meant to help AI agents use a website. It is not an official standard. Google says no AI-specific text file is needed for its AI features, and the OpenAI, Anthropic and Perplexity documentation does not say they use it to choose sources. Lighthouse (Chrome) does check it in its experimental "Agentic Browsing" category. It is cheap; just keep it current. How-to: llms.txt, what it is and how to create one.

How to Measure Your AI Visibility

No single tool covers every engine. The most reliable method is manual: list 20 to 40 real prospect questions, ask them every month with the same wording in ChatGPT, Perplexity, Gemini, Copilot, Claude and Google, and record whether your site is cited, whether your brand is named, whether the facts are right and which sources appear. Archive the raw answers: they vary between sessions, so only the trend over months counts.

• GA4: in Traffic acquisition, filter the session source on chatgpt, perplexity, gemini, copilot or claude, or create an "AI" channel group. Visits without a referrer land in direct, so treat the figure as a floor (see our GA4 guide).
• Search Console: AI Overviews and AI Mode traffic is counted in the Performance report ("Web" search type), with the rest of your Google traffic.
• Bing Webmaster Tools: the AI Performance report, launched in public preview in February 2026, counts your citations in Copilot and Bing's AI summaries and lists the queries used to retrieve your pages; it measures neither clicks nor position. Since June 2026, Microsoft has been rolling out intents, topics and citation share in it, in preview.

Mistakes to Avoid

• Blocking AI search crawlers by accident (over-broad robots.txt rule, CDN bot protection).
• Confusing training and search, and blocking OAI-SearchBot to avoid training.
• Betting on llms.txt or "AI-specific" markup, which Google says its AI features do not require.
• Vague copy with no price, date or source, or details that contradict each other across profiles.
• Keyword stuffing, which brought no gain in the GEO paper (and -10% on Perplexity.ai).
• Believing in guaranteed citations: no vendor documents a way to secure one.

10-Point GEO Checklist

• OAI-SearchBot, Claude-SearchBot, PerplexityBot, Googlebot and Bingbot not blocked in robots.txt.
• The decision on GPTBot, ClaudeBot, Google-Extended and Applebot-Extended is deliberate.
• No CDN or firewall rule blocks the search crawlers.
• Key pages are indexed in Google and Bing, with no accidental noindex or nosnippet; IndexNow is on.
• Every key page opens with a 40 to 60-word answer paragraph.
• Prices, lead times, service area and conditions are written out, with an update date.
• Pricing and service pages include a table and an FAQ; every figure links to a source.
• Name, address, phone, prices and founder are identical everywhere.
• Organization, Person, Service and FAQPage markup is validated and matches the visible text; llms.txt is current.
• A 20 to 40-question panel is tested monthly, alongside GA4 and Bing AI Performance.

FAQ

Sources

Consulted on 28 September 2026.

• Aggarwal et al., “GEO: Generative Engine Optimization”, arXiv 2311.09735 (KDD 2024)
• OpenAI — crawlers
• Anthropic — Claude crawlers
• Perplexity — crawlers
• Google — AI features in Search
• Google — common crawlers and Google-Extended
• Google — Organization structured data
• Apple — Applebot and Applebot-Extended
• Bing — AI Performance report (February 2026)
• Bing — intents, topics, citation share (June 2026)
• llmstxt.org — the llms.txt proposal
• Chrome for Developers — Lighthouse llms.txt audit
• Chrome for Developers — Lighthouse Agentic Browsing category (experimental)
• IETF — RFC 9309 (robots.txt)