ChatGPT's recommendations come down to five things: how relevant you are to the exact question asked, whether your site carries structured data it can read cleanly, whether your information is current, how strong your review signals are, and how well your content matches the intent behind the question rather than its keywords.
It's synthesis, not ranking
The important shift: ChatGPT isn't picking the top ten results and handing you the list. It's assembling one answer and deciding which specific names are safe to put in it.
That changes what winning looks like. In classic search, being marginally more relevant than the next site moves you up a position. In a synthesized answer there are no positions — there's a short list of names, usually three to five, and everyone else is absent. Being sixth-best is the same as being invisible.
It also changes what counts as good content. A model needs a passage it can lift and stand behind. Long preambles, vague claims, and pages that bury the answer under a story give it nothing to work with, however well written they are.
The five inputs, concretely
Relevance to the specific question. "Roofer near me" and "who repairs tile roofs in Winter Park" are different questions with different answers. Pages built around one broad service term compete for the broad question and lose every specific one.
Structured data. JSON-LD stating your business type, service area, services, and identity in machine-readable form. This is the difference between a model inferring what you do and knowing it.
Currency. Hours, phone number, service area, pricing posture. Stale facts are worse than missing ones — a model that gets burned on outdated information about you has a reason to stop citing you.
Review signals. Volume, recency, and consistency across platforms. Reviews are third-party corroboration, weighted more heavily than self-description.
Intent match. Somebody asking "how much does a roof replacement cost in Florida" wants a number and a range, not a contact form. Answer the question they asked; the business case follows.
Where most contractor sites fall down
In practice the failure is rarely all five at once. It's usually one of these:
- Schema exists, but it describes the website instead of the business.
- Name, address, or phone differs between the site footer, the Google Business Profile, and the licensing record.
- Service pages describe the service generically instead of the business performing it in a specific place.
- Reviews are strong on Google and absent everywhere else, so there's only one corroborating source.
Any one of those is enough to keep you out of the answer while your Google rankings look perfectly healthy.
The order to fix things in
Entity consistency first — it's free and it's foundational. Then schema, because it's a one-time build that keeps paying. Then rewrite your highest-intent pages answer-first. Reviews are a continuous habit rather than a project, so start that early and let it compound while the rest of the work lands.
None of this is exotic. It's the same fundamentals that make a site easy for any search engine to trust, applied to a system that writes a paragraph instead of listing ten links.
Go deeper: How to get cited by AI is the step-by-step version. Schema and llms.txt: the AEO edge covers the technical layer. If you'd rather see where you stand first, the AI visibility and AEO service opens with a free six-engine audit.
Go deeper
Full breakdown: How to Get Cited by AI