How AI assistants decide which sources to use
Assistants build answers in two ways. Some of what they say comes from what the model learned during training, which reflects how often and how consistently a brand appeared across the web. The rest increasingly comes from live retrieval: the assistant runs searches, reads a handful of pages and summarises them, often with citations. ChatGPT search, Perplexity, Copilot and Google's AI features all lean heavily on this retrieval step for commercial questions.
That retrieval step behaves a lot like a search engine with a very short results page. The assistant favours pages it can fetch quickly, parse cleanly and quote without ambiguity, and it tends to trust sources that other sources agree with. A brand that is described the same way on its own site, in reviews, on comparison lists and in press coverage is easier for a model to recommend than one with scattered or conflicting information.
This is why GEO is not a trick or a hidden setting. It is the work of making your brand easy to find, easy to understand and well corroborated, specifically for the prompts your buyers use. We study which pages each assistant actually cites for your topics, then work on getting your content and your mentions into that pool.