How ChatGPT Chooses Sources?

AI names the businesses it can identify and trust

AI assistants answer a buyer's question with a specific recommendation, and they name the business they can most clearly identify and trust. Understanding how chatgpt chooses sources starts here: the model is not ranking pages, it is choosing an entity. AI visibility is the outcome of being that chosen entity.

AI surfaces businesses from training data and live search

Assistants draw on two things: what they learned in training, and what they retrieve live from the web when they answer. A business that is clear, consistent and well-cited across both is the one that gets named. ChatGPT, Gemini and Perplexity each weight these differently, but the principle holds across all of them.

Training data gives the model its baseline sense of who you are; live retrieval lets it confirm and update that in the moment. Businesses that are absent from one and thin in the other give the model too little to go on, so it recommends someone else.

Entity, citations and reviews are the signals that matter

A handful of signals do most of the work in deciding who AI recommends:

• Entity clarity: a business AI can identify without ambiguity, the job of entity optimisation
• Citations and mentions: trusted third-party pages that corroborate you, built through citation building
• Reviews: genuine, consistent reviews that signal reputation
• Quotable content: passages a model can lift and attribute to you

Each engine weights these signals differently

The signals are shared, but the emphasis is not. Google Gemini leans heavily on Google's index, Knowledge Graph and Maps; Perplexity prizes pages it can cite live; ChatGPT blends its training with web search. Optimising for only one engine leaves you exposed on the others, which is why real AI visibility is built across all of them at once.

Inconsistent details keep businesses invisible to AI

The common failure is not bad luck. It is mixed signals. Conflicting details across directories, no third-party mentions and thin content all tell a model it cannot safely recommend you. Why your business isn't in AI covers the usual culprits and the fixes.

GEO turns these signals into AI recommendations

Knowing how AI chooses is only useful if you act on it. We run the full process end to end, audit, entity, citations, content and monitoring, so AI moves from ignoring your business to recommending it. See how we do it for clients.