Is your brand on the AI buyer's shortlist before anyone even searches?
- A disclosed-methodology analysis found that a brand named inside ChatGPT's own self-generated search query reached the final answer 68.9% of the time, versus just 2.1% for a brand that was fetched during the search step but never named in the query itself. Roughly a 33x gap.
- The honest caveat, stated by the analysis's own author: this comes from one account, 57 conversations, and 3,554 retrieved pages, weighted toward software and AI tools, "a direction rather than a measurement," not a controlled study. This piece treats it accordingly.
- The mechanism this points to: ChatGPT sometimes already has a working list of relevant brands before it retrieves anything, likely shaped by training-data presence and category association, and what gets fetched live during search struggles to break into that list.
- This differs from most "get your brand on the AI shortlist" content circulating right now, which gives generic visibility advice without explaining or sourcing the actual mechanism behind why some brands make the cut and others don't.
Search "is your brand on the AI shortlist" and you'll find no shortage of confident advice. What's harder to find is anyone explaining the actual mechanism: what determines whether a brand is even in the running before a generative engine starts retrieving anything. This piece is built around one real, disclosed-methodology attempt to measure that mechanism directly, reported with its actual limitations intact.
What did the analysis actually find?
Analyst Suganthan Mohanadasan, co-founder at Snippet Digital, ran a self-disclosed analysis sorting brands mentioned across ChatGPT conversations into two groups: brands that appeared in a search query ChatGPT generated for itself, and brands that were only fetched during that search without ever being named in the query. He then checked how often each group made it into the model's final answer.
The gap was large: brands named in ChatGPT's own query reached the final answer 68.9% of the time. Brands only fetched during the search, but never named in the query, reached it just 2.1% of the time, roughly a 33-times difference. Across the full dataset (57 conversations, 3,554 retrieved pages), only about 3.1% of retrieved pages were cited in a final answer at all, and the majority of first-query tests (21 of 27) already contained an unnamed brand baked into the query before any retrieval happened.
What does this mean, mechanically?
The implication is that ChatGPT sometimes isn't starting from a blank slate and then searching fairly across whatever it finds. It's sometimes already carrying a working shortlist, encoded in the query it writes for itself, before retrieval even runs. A brand that's part of that pre-search shortlist gets a large structural advantage; a brand that only shows up because a live web search happened to surface it is fighting for a much smaller share of outcomes.
That pre-search shortlist most plausibly reflects what's strongly associated with a category in the model's training data: the brands that show up repeatedly, consistently, and in context across the web content the model learned from. This lines up with, and gives a concrete mechanism for, the broader finding already covered on this site: identical prompts return the same list of companies less than 1% of the time across roughly 3,000 runs (SparkToro), and 82-89% of AI citations trace back to earned, third-party media rather than owned content (Muck Rack). See why earned media outperforms owned content. A brand's odds of landing on the pre-search shortlist are plausibly being set by the same off-site presence that already drives both of those findings, rather than by anything a brand controls at query time.
How reliable is this finding, honestly?
This is not a controlled, peer-reviewed study, and the author says so directly: "all of this comes off one account, so every percentage below is a direction rather than a measurement," and the sample is explicitly weighted toward software and AI tools rather than a representative cross-section of categories. A single account's conversation history, run a few hundred times, is a real and transparent piece of evidence, but it's one data point, not a settled law about how ChatGPT works across every category, account, or model version. Treat the 68.9%/2.1% gap as a credible signal that pre-search brand recall matters a great deal, worth checking for your own category with your own prompts, not as a number to quote as if it were a controlled experiment.
How does this differ from the rest of the "AI shortlist" advice already out there?
Searching this exact framing turns up a real amount of existing content, most of it generic "improve your brand visibility in ChatGPT" advice with no disclosed mechanism or methodology behind it. This piece exists because that mechanism is checkable, and because knowing why pre-search recall dominates changes what's worth doing about it: building the off-site, category-defining presence that shapes a model's default associations is a fundamentally different (and slower) project than optimizing a website for a live search index to retrieve. The five stages for winning in the AI search era, and specifically the fractional AEO/GEO/SEO methodology built around them, treat that off-site, compounding work as the priority precisely because a one-time on-site fix can't touch what a model already "knows" to look for before it ever runs a search.
Frequently asked questions
Does ChatGPT decide which brands to recommend before it even searches the web?
Sometimes, yes. One disclosed-methodology analysis found that a brand named inside ChatGPT's own generated search query reached the final answer 68.9% of the time, versus only 2.1% for a brand that was fetched during the search but never named in the query itself. That's evidence that pre-search brand recall (what the model already 'knows' to look for) matters more than what gets retrieved fresh.
How reliable is the 68.9% vs 2.1% statistic?
It comes from one analyst's single account, across 57 conversations and 3,554 retrieved pages: a real, disclosed methodology, but a small, single-account sample weighted toward software and AI tools, self-described by its author as 'a direction rather than a measurement.' Treat it as a credible signal worth investigating for your own category, not a universal law.
Why does it matter if a brand is named in the model's own query versus just fetched?
Because it suggests two very different paths to visibility. A brand that shows up in the query the model generates for itself likely has strong presence in the model's training data or established category association. A brand only fetched during live retrieval is competing purely on real-time search relevance, a weaker position, since the analysis found brands in that group made it into the final answer at roughly 1/33rd the rate.
What can a brand actually do about this?
Build the kind of off-site, third-party presence that shapes what a model already associates with a category before any specific query is asked (earned media, consistent entity information, and category-defining content) rather than only optimizing pages for a search index to retrieve. That's a slower, compounding fix, not a one-time technical change. Before investing in it, it's worth checking where you currently stand. An AI Shortlist Scan runs real buying-stage prompts against your brand and named competitors across four engines to show whether you're already inside that pre-search shortlist or not.
Sources referenced
- Suganthan Mohanadasan / Snippet Digital, via Search Engine Journal — "ChatGPT Already Knows Who It'll Recommend Before It Searches" (57 conversations, 3,554 retrieved pages, single-account analysis)
- SparkToro — brand consistency across repeated identical AI prompts (~3,000 runs)
- Muck Rack — Generative Pulse research on AI citation sourcing (25M+ links analyzed)
For the broader off-site strategy this finding supports, see why earned media outperforms owned content and the five stages for winning in the AI search era. To check where your own brand currently stands, see the AI Shortlist Scan.
