Most advice about getting recommended by AI assumes ChatGPT searches first and decides second. A small study suggests the order is often reversed. Research published by Suganthan Mohanadasan on Search Engine Journal on 14 August 2026 found that in 21 of 27 tested conversations, ChatGPT wrote brand names into its own search query that the user had never mentioned. The brands it named that way made the final answer most of the time. The brands it only found while searching almost never did. If you're trying to show up in AI answers, that gap tells you where to spend your effort.
What actually happens when you ask ChatGPT for a recommendation
When ChatGPT decides to search the web, it doesn't send your question as-is. The model rewrites your question into search queries (the fan-out), retrieves documents, and synthesizes an answer. We explained the mechanics of that rewriting step in our query fan-out explainer.
The new detail is what goes into the very first rewritten query. Ask for "the best project management tools for a small agency," and the query ChatGPT actually sends may already include specific product names. Because these were first-round queries, those names could not have been learned from an earlier search result in the same conversation. They came from what the model already knew.
The method was simple. The research is based on observable behavior in the ChatGPT interface, where search queries are visible in browser developer tools under a key currently named "search_queries."
The 68.9% vs 2.1% finding
Mohanadasan split every brand that came up into two groups and checked how often each group made the final answer. Brands named in ChatGPT's own query appeared in the final answer 68.9% of the time (68 of 119 cases). Brands that were fetched during search but never named in a query appeared in the final answer only 2.1% of the time (11 of 515 cases). That's roughly a 33x gap.
The search step mattered less than you might expect, too. The author also identified 86 cases where a brand was recommended in the final answer without its website being fetched at all during that conversation, indicating that some mentions come from the model's training data rather than retrieved content.
Search Engine Land's summary of what this means: "Retrieval determines who gets cited from that shortlist." The model decides who's in contention. The search then decides which of those contenders get a source link.
PPC Land's point matters for anyone paying for AI-visibility services: most of what is currently sold as generative engine optimization operates on retrieval, which is the 2.1 percent column.
Why the model "remembers" some brands and not others
A separate, larger study points the same way. Models search for brands they already remember at 3.2 times the rate of brands they don't. geoSurge measured this across nine industries, 66 US buyer prompts, 3,960 responses and 13,281 fan-out queries between May 29 and June 9, 2026.
Most fan-out queries don't name brands at all, but when one does, the familiar names dominate. Only 31% named a brand. But 63% of those brand-named queries pointed at one of the model's five most remembered brands. And the gap held in all nine industries, with remembered brands searched 41% to 82% of the time and unremembered brands 9% to 23%.
So what makes a model remember a brand? The best evidence so far comes from Ahrefs. Ahrefs' study of 75,000 brands found that branded web mentions had a 0.664 correlation with AI Overviews visibility, compared with 0.218 for referring domains (the classic backlink metric). A follow-up study covering ChatGPT, AI Mode, and AI Overviews found YouTube mentions had the strongest correlation, at 0.737.
In plain terms: being talked about across the web, on video, and in third-party coverage tracks with AI visibility much more closely than backlinks or page count do.
How to read this without overreacting
These numbers are useful, but they aren't laws, and some of the sources are open about their limits.
The 68.9% study is small. Lily Ray points out that the author states his methodology himself: a single account, based in Dubai, sampled in July 2026, weighted toward software and AI tools, with the brand-injection pattern measured across 27 initial queries. There's also a counting detail: the 68.9 against 2.1 figures use whole conversations, so a brand learned in turn one and queried in turn two counts as pre-known.
The brand-mention data is correlation. Kevin Indig, writing about the Ahrefs findings, warns that brand size, existing demand and category could influence both, so the finding doesn't prove that increasing mentions alone will increase AI visibility. Big brands get mentioned more *and* get recommended more. That doesn't prove the mentions caused the recommendations.
Retrieval still counts. Being on the shortlist gets you into the answer. Being retrievable is still how you get the citation link and the click. These are two separate jobs, and the research suggests most teams are working hard on only one of them.
What to actually do with this
If the model's memory sets the shortlist, your job is to become a brand it remembers, and to stay easy to retrieve once it goes looking for you.
- Test your own category the way the study did. Ask ChatGPT the recommendation questions your customers would ask. If you know how, open the browser developer tools and look at the search queries it sends. Are you named in the first query, or only fetched later, or missing entirely?
- Put third-party mentions ahead of more on-site pages. Reviews, roundups, podcast and YouTube appearances, comparison articles, and community discussions are the off-site signals that track with AI visibility.
- Make your category obvious. A model can't shortlist you for "best X for Y" if the web doesn't clearly connect your name to X and Y. Describe what you are the same way everywhere.
- Keep the retrieval basics. Clear, current, well-structured pages still decide whether you get cited once you're on the list.
- Track mentions and citations separately. Being named in an answer and being linked as a source are different results. Measure both.
None of this is a quick technical fix, and that's the point. The 2.1% column is where the quick fixes happen. The 68.9% column comes from slower work: getting known in your category.
FAQ
Does ChatGPT really decide which brands to recommend before it searches?
Often, according to the available evidence. In one study, ChatGPT added brand names the user never typed into its first search query in 21 of 27 conversations, and those brands made the final answer 68.9% of the time. The sample was small and leaned toward software, so read it as a strong signal, not a universal rule.
Do backlinks still matter for AI search?
Yes, for traditional rankings and for getting your pages retrieved. But in Ahrefs' 75,000-brand study, backlinks correlated far more weakly with AI visibility (0.218) than branded web mentions did (0.664). Links help you get found. Mentions seem to matter more for being remembered.
How can a small brand get onto the AI shortlist?
Focus on a narrow category where you can realistically become one of the names people mention. Earn reviews, comparison-article placements, video coverage, and community discussion that consistently tie your brand to that category. Then check regularly whether ChatGPT names you on its own when someone asks the questions your customers ask.