You get recommended by ChatGPT — and Perplexity, Gemini, and Google AI Overviews — by making three things true at once: the model can form a clear, unambiguous picture of who you are, it can cite evidence that answers the buyer's question, and that picture is corroborated by third parties the model already trusts. There is no prompt hack, no hidden tag. Being recommended is the compounding result of clarity, evidence, and reputation.
The good news: most brands have done none of this deliberately — a gap that the benchmark methodology behind my audits is built to measure. The work below is what closes it.
Key takeaways
AI recommendations come from entity clarity, citable evidence, and third-party corroboration — not prompt tricks.
Publish content that answers a question better than the question was asked: research, benchmarks, definitions, how-tos.
Third-party mentions in publications, reviews and directories act as citations for models that never visit your site.
Write the first 100–150 words of key pages as a complete, extractable answer.
Measure monthly against a fixed prompt set and let the measurements steer what you publish next.
The five steps
Make your entity unambiguous
AI engines answer from a coherent picture of who you are. A clear, consistent identity — name, category, differentiators, proof — across your site, schema, and profiles is the foundation everything else stands on.
Publish citable evidence
LLMs cite content that answers a question better than the prompt needed to ask it: original research, benchmarks, definitions, clear how-to guides. Thin marketing copy gets cited by nobody.
Earn mentions where AI reads
Answer engines lean on third-party sources — industry publications, review sites, directories, expert roundups. Mentions there function as citations for the AI models that never visit your site.
Shape answers, not just pages
Write the first 100–150 words of key pages as a direct, self-contained answer. Structure with clear headings, facts, and lists. You are optimizing for extraction, not just ranking.
Measure and iterate
Track your share of voice, citation presence and description accuracy monthly, against a fixed prompt set. What the measurements say should decide what you publish next.
What does not work
Keyword-stuffing your pages: answer engines extract meaning, not keyword density.
Publishing AI-generated content about your brand in bulk: models weigh distinctive, corroborated sources, not volume.
Trying to "optimize one prompt": recommendations are patterns across the corpus, not a single slot you win once.
Where to start this week
Run the diagnostic first: ask ChatGPT, Perplexity and Gemini the ten questions your buyers actually ask, and note where you appear, where competitors appear, and what the answers get wrong about you. That list — not a content calendar — is your roadmap. The method for doing it rigorously is in How to Measure AI Visibility, and a done-for-you version is the AI Visibility Audit.
Done consistently, the effect compounds: each piece of citable evidence makes the next recommendation more likely, and the brands that start early become the defaults the engines reach for. That is what makes AI visibility a growth channel, not a gimmick — the numbers behind it are here.