ChatGPT often decides which vendors to consider before it ever fetches the web. That internal, pre-search shortlist — visible in the chat response under a key called search_queries — regularly names brands, model numbers and modifiers like “official” and “pricing.” If your brand isn’t on that list, the model may never run a site: probe against your domain, and your chances of being cited fall sharply.

What the tests found

An examination of dozens of ChatGPT conversations shows a clear, repeatable pattern. When asked to recommend products, the model first writes its own search queries. Those generated queries often contain brand names and specific terms the user did not supply.

  • In 21 of 27 conversations, the model’s first query included brands the user hadn’t typed.
  • First queries commonly add year and utility modifiers — for example, “2026,” “official” and “pricing.”
  • Across categories the model sometimes lists vendors (e.g., Duolingo, QuickBooks, Roborock) and in other cases reviewer sites (e.g., Car and Driver, Edmunds); the pattern varies by category and phrasing.
  • Some first queries even include specific product model numbers without prompting, indicating fine-grained product awareness.
  • Being mentioned in that pre-search query matters: the analysis found appearing in the model’s generated query increased the chance of being cited in the final answer by roughly 33× compared with merely being retrievable during the subsequent web fetch.

Why this matters for marketers and product teams

Two decisions in the model’s workflow now control brand visibility. First, the model assembles a shortlist internally before any web retrieval. If your brand isn’t on that shortlist, the model often never probes your site. Second, when it does run targeted site: probes, it applies a stricter filter that favors pages labeled “official” or “pricing.”

That means classical SEO (crawlability, content quality, links) remains necessary but is no longer sufficient by itself. You also need to be visible to the model at the shortlist stage — effectively an awareness problem, not just a technical one.

Practical checklist: what to test and fix first

These are short, testable actions every digital team can run today.

  • Inspect the model’s shortlist for your category. Ask ChatGPT the buyer-facing question your customers ask (for example, “best [your category]”). In the chat response JSON look for the search_queries key and record the brands and terms the model generated.
  • Repeat the query multiple times. Run the same prompt at least five times. Names that recur across runs are the model’s stable competitive set; names that fluctuate are contested opportunity areas.
  • Ensure site: probes survive. When ChatGPT names a brand it often follows with a site:yourdomain.com probe targeting pages like /pricing or /features. Make sure these pages exist, are canonical, and return clean, crawlable HTML that answers short comparison queries.
  • Treat shortlist absence as awareness work. If your brand never appears, prioritize signals the model consumes: consistent product naming, clear authoritative pages, and distribution that places your brand in the same contexts as known competitors.

These steps don’t assume you can change the model’s internal training. They assume you can adapt your public footprint so the model sees your brand as a candidate in the first place.

What to watch next

Model vendors may change probing behavior, and the sources that feed those shortlists could shift over time. For now, add a ChatGPT shortlist check to routine competitive audits: map recurring names, confirm your official and pricing pages survive site: probes, and decide whether to prioritize awareness campaigns or page-level fixes based on that result.

If you run the shortlist test and find your brand missing across runs, treat that as a strategic signal: you need to buy a place in the model’s head, not just in search engine indexes. The concrete next step is simple and repeatable — run the test, map the results, and act on whichever gap the test highlights (visibility or technical coverage).