How to get recommended by ChatGPT and Perplexity
What answer engines quote, what to publish so they quote you, and how to verify it independently.
A growing share of buyers now form their first impression of a category inside a chat window. 6sense found 94% of B2B buyers using large language models somewhere in the buying process during 2025.
Those systems answer in prose and cite a handful of sources. Getting quoted is a different discipline from ranking, and it rewards different writing.
How answer engines choose what to cite
A model assembling an answer needs passages it can lift with the meaning intact. That favours writing with specific properties, most of which run against the habits of marketing copy.
- A clean definition in one sentence. Subject, verb, meaning, placed early.
- Answers directly under their questions. Heading as question, first sentence as the complete answer, elaboration afterwards.
- Attributed numbers. A figure with a named source and a year survives extraction. A bare percentage gets dropped as unverifiable.
- Self-contained paragraphs. A passage that depends on the three paragraphs above it becomes wrong when quoted alone.
- Plain nouns. Invented category names and internal jargon break the match between a buyer's question and the text.
What to publish
Definitions of the terms your category argues about
Definitional queries are the most commonly answered kind, and in most categories the existing definitions are thin. Writing the clearest one available is the single highest-return page in this discipline.
Comparisons, written honestly
Buyers ask models to compare options constantly. A comparison that concedes where an alternative wins gets treated as a credible source. A comparison where one option wins every row reads as promotional and gets skipped.
Original numbers
Models cite sources for facts. A company that publishes its own measured data becomes a citable primary source. Everyone else is restating someone else's research.
Direct answers to the awkward questions
Pricing, limitations, who the product suits poorly. These questions get asked in every evaluation, and most vendor sites leave them blank, which leaves the answer to third parties.
The technical minimum
- Allow the crawlers that decide whether an engine can answer from you: OAI-SearchBot for ChatGPT search, PerplexityBot for Perplexity, and Claude-SearchBot with Claude-User for Claude. GPTBot and ClaudeBot are the training crawlers, so those two are a separate choice.
- Google's AI Overviews and AI Mode are part of Search and follow Googlebot and the snippet controls, nosnippet and max-snippet. Google-Extended covers Gemini Apps and Vertex AI grounding, and Google states it has no effect on inclusion in Search.
- Serve content in the initial HTML, since crawlers that skip JavaScript rendering see an empty page otherwise.
- Mark up FAQ, Article and Organization schema, which supplies the explicit question-to-answer mapping.
- Keep one clear topic per URL. A page covering six subjects matches none of them precisely.
How to check whether it worked
Visibility here is measurable, with more effort than a rank tracker requires.
- Ask directly. Put the twenty questions a buyer would ask to each major model monthly and record which sources get named. Slow, and the only ground truth available.
- Watch referral traffic. Visits arriving from chat interfaces appear in analytics as their own referrers.
- Watch branded search. A recommendation inside a chat window frequently produces a search for the name afterwards, which shows up as branded volume with no matching campaign behind it.
I have taken a twenty-person software company to first position across ChatGPT, Gemini and Perplexity for its category term in about two months, and repeated the same approach since at an AI automation platform.
The method was unglamorous. Establish what the buyers actually type. Audit what the existing answers leave out. Write one page that answers the core question in its opening sentence, then supporting pieces that each answer one adjacent question completely. Mark it all up so the question-to-answer mapping is explicit. Then wait, because indexing takes weeks.
No trick, and no budget required. The advantage came from being the clearest source available on a question that mattered, at a moment when most of the category was still writing brochures.
The part that gets missed
Optimising for answer engines and writing genuinely useful material have converged into the same activity. The writing that gets quoted is clear, specific, sourced and willing to concede a point.
Which describes good writing, and always did.
Where this becomes work
Sold as a project, this is AI search visibility: six to ten weeks to build the topic architecture, then ongoing. The industry also calls it answer engine optimisation and generative engine optimisation, AEO and GEO.
Keep reading: Content marketing wins the deal before the first call and What is curiosity marketing?.
Questions people ask
How do you get cited by ChatGPT and Perplexity?
Publish passages that can be lifted with the meaning intact: a clean one-sentence definition, answers placed directly under their questions, numbers with a named source and year, and self-contained paragraphs that stay true when quoted alone.
Do answer engines use different signals from Google?
They overlap but reward different writing. Search ranks pages, answer engines quote passages, so structure and extractability matter more than page-level authority alone.
Which crawlers should a site allow for AI visibility?
OAI-SearchBot for ChatGPT search and GPTBot for OpenAI training, PerplexityBot for Perplexity, Claude-SearchBot and Claude-User for Claude answers with ClaudeBot for training, and Google-Extended for Gemini Apps and Vertex AI grounding. Google's AI Overviews and AI Mode follow Googlebot and the snippet controls, since Google-Extended has no effect on inclusion in Search. All are controlled through robots.txt.
How do you measure visibility in answer engines?
Ask each major model the twenty questions a buyer would ask, monthly, and record which sources it names. Support that with referral traffic from chat interfaces and with branded search volume that has no campaign behind it.
Curiosity works both ways.
Analytics shows me which pages hold a reader and which lose one. It counts pages and stops there. What it stores.