How AI engines work
How does ChatGPT choose which brands to recommend?
Two different mechanisms, two different fixes
When ChatGPT browses, the answer is assembled from pages retrieved in that moment. Being named depends on whether your brand appears in the sources it happens to pull — review sites, comparison articles, forums, and your own pages. This is fixable in weeks, because it is a retrieval problem.
When ChatGPT answers from memory alone, it draws on statistical patterns in its training data. That is far slower to influence and depends on the breadth and consistency of everything published about you before the training cutoff. There is no way to edit it directly — only to be described consistently enough, in enough places, that the pattern forms.
What tips a mention into a recommendation
Being mentioned and being recommended are different outcomes. A mention needs your name to appear in a retrieved source. A recommendation needs the model to have a reason to prefer you — a stated strength, a specific audience fit, a price point, a third-party rating it can point at.
The brands that win recommendations tend to have specific, quotable claims attached to them in independent sources: 'best for small teams', 'cheapest per seat', 'the one with SOC 2'. Vague positioning gives a model nothing to justify a recommendation with.
See where you actually stand
We ask AI engines the questions your buyers ask and show you whether you were named, who was named instead, and which sources they cited. Free.
Related questions
Perplexity searches the live web for every query, retrieves a shortlist of pages, and cites the ones whose passages most directly answer the question. It favours pages with a clear, extractable answer near the top, recent publication or update dates, and topical focus — a narrow page that answers precisely often beats a broader, higher-authority one.
AI Overviews run on Google's core ranking and quality systems, then select passages through query fan-out — breaking one question into several sub-queries and retrieving a passage for each. Because selection happens at passage level, most cited pages do not rank in the organic top ten for the original query.
Usually one of four reasons: your site blocks AI crawlers, your content only renders in JavaScript, nothing independent has been published about you, or your brand name is ambiguous enough that the model cannot resolve which company you are. The first two are technical and fixable in a day.
Query fan-out is when an AI engine breaks a single question into several related sub-queries, searches each one separately, and synthesises the results into one answer. Asking for 'the best CRM for a small agency' might fan out into searches for best CRM 2026, CRM pricing for small teams, and agency CRM reviews.
More on how ai engines work
Last reviewed 2026-07-30.