ChatGPT Already "Knows" Which Brand to Recommend Before It Even Searches

SEO specialist and Snippet Digital partner Suganthan Mohanadasan closely analyzed 60 ChatGPT conversations and uncovered a striking pattern: before ChatGPT ever fetches a single page from the web, it has already written specific brand names into its own internal search query. For instance, when a user asked about "the best AI note-taking apps," the search query ChatGPT generated already contained seven product names — Granola, Notion AI, Otter, Fireflies, Fathom, Mem, and Limitless — even though the user hadn't named a single one in their question.

Here's the mechanism: ChatGPT turns the user's question into a search query itself (stored in an internal field called "search_queries," renamed from "search_model_queries" in early August 2026), runs a search on that query, reads the results, and writes a response. This isn't a leak — the browser itself uses that data to render the page, and it's visible through browser DevTools.

The author's main objection was about timing: maybe ChatGPT searches once first, reviews the results, then writes a smarter second query? To test this, he compared the user's first message against the very first search query in each conversation — the point at which nothing had been fetched yet. Of 27 conversations, 21 already contained brand names in that very first query that the user had never written.

To check whether this was confined to one category, the author ran tests across 12 unrelated categories. For robot vacuums, ChatGPT volunteered specific, current model names unprompted — Saros, Dreame X50, Eufy S1 Pro. For electric SUVs, the behavior was different: instead of brands, it cited publication names (Car and Driver, Edmunds, Top Gear). On retesting, some categories stayed stable (language-learning apps), while others shifted sharply — in accounting software, for example, six brands collapsed into a single query aimed only at QuickBooks.

But two things held constant: brand names were "baked into" the query every single time, and categories with clear market leaders kept their names. The author also tested whether the word "best" mattered — it didn't; the key factor was whether ChatGPT had to "invent" the products itself or not. If a question draws on general knowledge (like "how does noise cancellation work") or the user is simply stating an opinion, no search runs at all. If the user names a brand themselves, ChatGPT goes straight to that brand's page; if the choice is left up to ChatGPT, it draws on its "memory" — that is, on data it already learned.

The practical takeaway for marketers and brands is clear: if your brand isn't in ChatGPT's "memory," it has already been crossed off the list before the web search even begins — and it will never "see" you, no matter how well-built your site is. The author's advice: ask ChatGPT a "best..." question in your category and read the internal search query it writes — the simplest way to find out whether your brand exists in the AI's "memory."