How to get cited by AI assistants: ChatGPT, Perplexity, Claude, Gemini

An AI assistant cites a page when it can reach it, finds something there it cannot get elsewhere, can read it without effort, and trusts the source. The four steps below follow that order. The work is AI search optimization, also called GEO (generative engine optimization) or AEO (answer engine optimization), and Google's own guide treats it as SEO applied to its AI features (Google).

#Step 1: be reachable

If an engine cannot fetch your page, nothing else counts. Each provider runs separate crawlers, controlled by separate tokens in robots.txt: one builds its search index, one fetches a page when a user asks a question, and one collects training data. The search crawler puts you in the index the answers draw on, and the user fetcher reaches your page live when someone asks. Training is a separate choice. OpenAI puts it plainly: a site that opts out of its search crawler "will not be shown in ChatGPT search answers" (OpenAI).

Each row comes from the provider's own page: OpenAI, Anthropic, Perplexity and Google.

Four details change what you do:

• OpenAI and Perplexity say a robots.txt change can take about 24 hours to show in their search results, so judge the effect the next day.
• A crawler looks for its own token in robots.txt, matched without regard to case, as a whole word (RFC 9309). Spell each token out.
• Blocking Google-Extended keeps your pages out of Gemini's grounding as well as its training, and leaves Google Search untouched. For AI Overviews and AI Mode, Google also requires your site to be included in its generative AI features in Search Console.
• For training, a newer signal is arriving: Anthropic honors a machine-readable "no AI training" preference since 30 October 2026, based on an IETF draft, and Google plans to support it in early 2027.

#Step 2: say what no other page says

A model composes one answer from many sources. A page that repeats what ten others say gives it nothing new to cite. A page with something only you can publish does: your numbers, your tests, your customers' words, an honest comparison from someone who has used both products.

Google says it plainly: content people find unique and useful will likely shape your presence in its AI features more than any other suggestion in its guide, and it advises against recycling what others have said or what a model could easily write (Google).

Then write the answer first. Lead each page, and each section, with the direct answer in one or two sentences, and put the depth underneath. A reader who skims has it, and a model has a clean sentence to quote. Keep claims specific and checkable: "faster onboarding" says little, "onboarding in one form, with the first audit back the same day" can be quoted as it stands.

Google adds two things you can skip: cutting pages into small chunks, and rewriting your style for AI. Its systems understand synonyms and pages that cover several topics.

#Step 3: keep it clear

Structure lets a model read a page with confidence about what each part means.

1. Give each question or topic a clear heading, so the page mirrors the questions people ask.
2. Keep one topic per page.
3. Put the answer in the visible text. A test by searchVIU, reported by Ahrefs, found that ChatGPT, Claude, Perplexity, Gemini and Google AI Mode read only the visible HTML when they fetched a page live (Ahrefs).
4. Write a real FAQ, in your buyers' words, for your readers.

Schema markup still earns rich results in Google Search, and Google recommends it for that. Once a page is cited, it is no lever for more citations: Google says no special schema is needed for its AI features, and Ahrefs found that adding JSON-LD to 1,885 pages AI already cited did not raise their citations on AI Mode or ChatGPT (May 2026). Whether it helps a page no engine cites yet, that study cannot say. Google also stopped showing FAQ rich results on 7 May 2026 (Google). An FAQ still helps your readers.

#Step 4: be trusted

The last step is the slowest and the most durable. Models lean on sources they already trust, and they read your brand across the web. A real reference on an independent site carries more weight than another page on your own domain.

That means earning it: being included in a roundup a model reads, answering in the communities the engines index, being the source other people cite when they explain your category. Shortcuts fail here. Google warns that chasing inauthentic mentions is less effective than it looks, because its ranking systems favor quality and other systems block spam.

#How to know it worked

Ranking position is the wrong metric in AI search. Track presence instead:

• Mention rate: how often the engines name you for the questions that matter.
• Citation rate: how often they link you as a source.
• AI share of voice: your slice of the answers against your competitors.
• Position in the answer: where you sit in the order the model wrote.

Check each engine apart, because presence on ChatGPT says nothing about Perplexity, Claude or Gemini. For Google, Search Console now reports how your pages perform in AI Overviews and AI Mode. Measuring AI search visibility defines each number. Then run it as a loop: publish, verify, adjust, repeat.

#Put the four steps to work

Obi, an AI Search Employee, does this work for founders and GTM teams, from the agent harness they already use. It checks that the engines can reach your pages and proposes the fixes. It finds the missing pieces only you can publish and drafts those pages with you in your AI assistant, from your data, for you to review and publish. Off your site, it proposes a Reddit reply or a media pitch, for you to post or send. On a paid pack, it reads your answers on ChatGPT, Gemini and Google AI Overviews every week, so you see what moved.