Generative Engine Optimization: How to Get Cited by AI Answer Engines

Generative engine optimization (GEO) is the work of making your pages easy for AI answer engines, such as ChatGPT search, Perplexity, Claude, Gemini and Google’s AI Overviews, to find, understand and cite. Most of it is solid SEO: crawlable pages, clear answers and visible expertise. What’s specific to GEO is checking that AI crawlers can actually reach you, and writing passages that still make sense when quoted on their own.

How AI answers pick their sources

There are two ways a language model can “know” about your site, and only one of them sends visitors.

• Training data. Text collected in the past and baked into the model. It shapes general knowledge, but the model usually can’t point to where a fact came from, so there’s no link.
• Live retrieval. When a question needs current or specific information, the system runs searches, fetches pages, reads the most relevant passages and writes an answer with citations. This is where GEO pays off.

The retrieval step works much like a search engine, because it usually is one. Google says its AI Overviews and AI Mode may use a “query fan-out” technique, issuing multiple related searches across subtopics and data sources to build a response. That’s described in its documentation on AI features and your website. OpenAI states that sites which block OAI-SearchBot won’t be shown in ChatGPT search answers. Perplexity runs its own index built by PerplexityBot.

So the first filter is the same as in classic SEO: the page must be in the index the AI system searches. The second filter is passage quality. Systems pick passages that answer one part of the question directly and credibly.

Technical access: let the right AI crawlers in

AI companies separate their bots by purpose, so blocking “AI” in general is rarely what you want. The table below follows each company’s documentation: OpenAI, Anthropic, Perplexity and Google.

Google and Apple work differently. AI Overviews and AI Mode draw on the regular Googlebot index, while the Google-Extended robots.txt token controls whether Gemini may use your content for training and grounding. Apple’s Applebot crawls for products such as Siri and Spotlight, and Applebot-Extended is an opt-out from training.

A common setup allows search and user fetches and blocks training:

# AI search and user-requested fetches: allowed
User-agent: OAI-SearchBot
User-agent: ChatGPT-User
User-agent: Claude-SearchBot
User-agent: Claude-User
User-agent: PerplexityBot
Allow: /

# Model training: blocked
User-agent: GPTBot
User-agent: ClaudeBot
User-agent: Applebot-Extended
User-agent: CCBot
Disallow: /

Three points to keep in mind:

• Google-Extended doesn’t affect Google Search. Blocking it doesn’t remove you from AI Overviews or change rankings. It does limit how Gemini uses your content, including for grounding answers, so decide deliberately.
• User fetchers may ignore robots.txt. OpenAI says robots.txt rules may not apply to ChatGPT-User, and Perplexity says the same of Perplexity-User, because a person requested the page.
• Firewalls override robots.txt. Some CDNs and security plugins offer one-click AI bot blocking. A 403 at the firewall blocks a bot your robots.txt allows. The AI crawler checker catches this by requesting your page with real AI user agents.

Finally, make sure the content is in the server HTML. Vercel’s study of AI crawler traffic found that none of the major AI crawlers it measured rendered JavaScript. A page that builds its text in the browser can look empty to them. Our guide on blocking or allowing AI crawlers in robots.txt goes deeper into each bot.

Structure content so it can be quoted

An AI system rarely cites a whole page. It lifts a passage, sometimes a single sentence, that answers one sub-question. Write so that the best passages are easy to find and safe to quote.

Answer first, then explain

Open each section with the direct answer in one or two sentences, then add detail, exceptions and examples. Compare:

Hard to quote: “There are many factors to think about when it comes to how often you should water a fiddle-leaf fig, and the answer may surprise you…”

Easy to quote: “Water a fiddle-leaf fig when the top 5 cm of soil is dry, usually every 7–10 days indoors. Water less in winter, when growth slows.”

Make passages self-contained

A passage quoted out of context should still be correct. Name the subject instead of writing “it” or “this method”, and avoid answers that depend on a table three screens up. If a sentence only makes sense after reading the previous section, rewrite it.

Use headings that match real questions

Phrase H2 and H3 headings the way people ask: “How much does a heat pump cost to run?” rather than “Running costs overview”. Query fan-out breaks a question into sub-questions, and clear headings help a system map each one to the right section.

Be specific and show your sources

Numbers, dates, named methods and clear definitions are what answer engines cite. Link to the primary source when you state a fact. The academic paper that coined the term, “GEO: Generative Engine Optimization”, found that adding citations, quotations and statistics made sources more visible in generated answers in its test setup. That’s a research benchmark, not proof of how any production engine ranks sources, but it fits common sense: verifiable content is safer to quote.

Offer something others don’t

If ten pages say the same thing, the system only needs one of them. Original data, first-hand testing, photos, local detail and expert opinion give it a reason to choose you.

Entities and schema

AI systems try to work out who is speaking as well as what is said. Help them by using one consistent name for your organization, products and authors across your site and profiles, and by marking that up with structured data:

<script type="application/ld+json">
{
  "@context": "https://schema.org",
  "@type": "Article",
  "headline": "How to repot a fiddle-leaf fig",
  "datePublished": "2026-03-02",
  "dateModified": "2026-10-14",
  "author": {
    "@type": "Person",
    "name": "Dana Reyes",
    "url": "https://example.com/team/dana-reyes",
    "jobTitle": "Horticulturist"
  },
  "publisher": {
    "@type": "Organization",
    "name": "Example Plant Co.",
    "url": "https://example.com",
    "sameAs": ["https://www.linkedin.com/company/example-plant-co"]
  }
}
</script>

Be honest about what this does. Google says no special schema is needed for its AI features, and no AI platform has published evidence that structured data increases citations. Its proven value is clearer machine understanding and eligibility for rich results. Keep the markup consistent with what’s visible on the page. Our schema generator builds Article, Organization and Person markup from a form.

E-E-A-T signals that machines can see

Experience, expertise, authoritativeness and trust, or E-E-A-T, comes from Google’s quality rater guidelines. It isn’t a single score, but its signals are visible on the page, which makes them useful to any system judging whether a source is reliable:

• A named author with a bio page that states real credentials or experience.
• An about page, contact details and a clear owner of the site.
• Citations to primary sources for facts and figures.
• First-hand evidence: “we tested”, original photos, data you collected.
• An editorial or review policy for sensitive topics such as health and finance.

Freshness

For time-sensitive questions, answer engines tend to favor recent sources. Show a visible “last updated” date, set dateModified to match, and actually update the content: new prices, current versions, changed rules. Changing the date without changing the page is easy to spot and erodes trust. For evergreen topics, accuracy matters far more than recency.

Measuring AI referrals

There’s no single dashboard for AI visibility yet. Combine three sources.

Analytics

Visits from AI assistants usually arrive as referrals. In GA4, open Traffic acquisition, set the dimension to session source and filter with a regular expression like this:

chatgpt\.com|chat\.openai\.com|perplexity\.ai|claude\.ai|gemini\.google\.com|copilot\.microsoft\.com

ChatGPT often adds utm_source=chatgpt.com to the links it shows, which makes those visits easy to spot. Some AI traffic still arrives with no referrer and lands in “direct”.

Server logs

Logs show which AI bots crawl which pages, and how often. Verify the bots before trusting the numbers: OpenAI and Perplexity publish their IP ranges as JSON files, linked from their bot documentation.

Search Console and manual checks

Google includes AI Overviews and AI Mode traffic in the Performance report under the Web search type, not as a separate line. Add a monthly manual check: ask a few AI assistants the questions your pages answer and record whether you’re cited. Answers vary from run to run, so look for trends, not single results.

What’s unproven, and what to skip

• llms.txt. An optional proposal from llmstxt.org for a Markdown map of your site. As of 2026, no major AI platform has confirmed using it for ranking or citations, and Google has said it isn’t needed. It’s cheap to add, so add it if you like, but don’t expect results. Our llms.txt guide explains the format.
• Content shown only to AI bots. Serving different text to bots than to people is cloaking, and it’s a spam policy violation for Google.
• Statistic stuffing. Adding numbers without sources makes content less trustworthy, not more.
• Guaranteed AI rankings. Nobody controls which sources an AI system picks. Treat such promises as a red flag.
• Prompt-tracking scores as truth. Tools that sample AI answers are useful for trends, but a sample of prompts isn’t your real audience.

A short GEO checklist

1. Page is indexed in Google and Bing and returns 200.
2. robots.txt and firewall allow the AI search and user-fetch bots you want.
3. Main content is in the server HTML, with no snippet restrictions you didn’t intend.
4. Each section opens with a direct, self-contained answer.
5. Facts are specific, sourced and current.
6. Author, organization and dates are visible and marked up.
7. AI referrals are tracked in analytics and logs.

Google’s AI features have a few rules of their own. Our guide on how to appear in AI Overviews covers them.

Check your page’s AI visibility

Our free AI SEO checker gives any page an AI visibility score from 0 to 100. It checks AI crawler access in robots.txt, server-rendered text, snippet directives, answer-ready structure, entity schema, author and date signals, freshness, citations and llms.txt, and tells you what to fix first.