The pipeline, in five steps
When ChatGPT, Perplexity, or Gemini answers a query like ‘best dentist near St. Louis,’ five things happen behind the scenes — and your site can be present at any one of them.
- Index — the engine has crawled and stored a representation of your page (and the web’s pages, plus third-party text about you).
- Retrieve — when the query comes in, the engine pulls the small set of documents most likely to answer it. This is mostly vector similarity plus traditional ranking signals.
- Synthesize — the language model writes a paragraph using the retrieved documents as context.
- Cite — the synthesizer references one to three sources, picking those most likely to be authoritative for the specific claim.
- Render — the answer plus citations is shown to the user, sometimes with follow-up suggestions.
Where you can win
You can’t change the model. You can change what gets retrieved and what gets cited. That’s the entire game. Three levers:
- Be in the index. Crawlable, indexable, present in the corpora the engines train and retrieve from. This is the SEO basics — sitemap, robots, internal links, schema.
- Be retrievable. Your content matches the embedding the engine generates from the query. Specific, well-structured answers beat keyword-stuffed paragraphs.
- Be citable. Third-party text refers to you in plausibly authoritative ways — local press, podcasts, a Wikipedia page, citations across .gov and .edu domains.
“AI search rewards specificity in ways traditional SEO never did. ‘Best dentist’ is a keyword. ‘Best dentist for kids in Webster Groves on a Saturday’ is a citation magnet.”
What changed for local businesses
The single biggest shift: the ‘long tail’ got short. A user used to type ‘dentist near me’ and click through ten options. Now they ask the AI a fully-formed question, get one paragraph, click one option. The number of impressions per query went down; the value of being the cited option went up by an order of magnitude.
Where we focus the work
We focus 60% on FAQ schema and answer-pages tuned to retrieval, 30% on third-party citations and authority signals, 10% on the basics. The basics still matter — but most local sites have those covered already. The retrieval and citation work is the lever that’s rarely pulled.