Parse traced 743,998 brand mentions across ChatGPT Search and Google AI Mode in June 2026. Although 72.7% held some recommendation position, only 35.7% reached the top three and 12.9% ranked first. Another 35.7% had a cited source. Mention rate, recommendation rank, and citation support describe different outcomes and should be reported separately.
Named, recommended, cited: three rates, not one
When AI answers a buyer question, it usually names several brands in passing, ranks a few of them as recommendations, and backs only some with a source. These are three different events, and they happen at very different rates. We measured all three on the same set of answers.
Of 743,998 brand mentions, 540,868 (72.7%) sat somewhere in the answer's ranked recommendation list, 265,895 (35.7%) made the top three, and just 95,723 (12.9%) were ranked first. Separately, 265,897 (35.7%) were backed by a citation that the model attributed to a specific source. So roughly three-quarters of named brands earn some recommendation footing, but the top slot and a real source are each scarce: about one named brand in eight is the pick, and only about one in three has a source behind it.
- Of every brand AI named, 72.7% held some recommendation rank, 35.7% made the top three, and only 12.9% were ranked first (Parse, 743,998 mentions, ChatGPT Search and Google AI Mode, June 2026).
- More than a quarter of named brands (27.3%) never earned any recommendation slot at all: they were mentioned and then passed over.
- Only 35.7% of named brands were backed by a source the model cited. Most brands AI names are recalled from memory, with nothing behind them.
- A typical answer names about 5 brands but ranks only 2 in its top three and puts exactly 1 first (Parse's data on how many brands a typical AI answer names shows the same shortlist). Naming is crowded; the pick is singular.
- Across 30 industries the named-to-top-three rate barely moves (29.6% to 42.0%), so the gap between being named and being recommended is structural, not a quirk of any one category.
How we measured the funnel
We used Parse's response-mention layer, which records, for each AI answer, every brand named in the text, whether the answer ranked that brand as a recommendation (and at what position), and whether a cited source was attributed to it. The window is June 2026, the period where the recommendation rank is computed for both live engines, and it covers ChatGPT Search and Google AI Mode, the two surfaces Parse monitors on the current web-search collection. The panel holds 132,642 answers, 743,998 brand-naming events, and 105,673 distinct named brands.
We define the three rungs precisely. "Named" means the brand was directly mentioned in the answer text. "Recommended" means the answer assigned it a recommendation position (rank 1, 2, 3, and so on in the list the model actually put forward). "Cited" means a source attribution linked that brand to a specific domain in that answer. Every figure is aggregate across the monitored panel, with no single customer identifiable. We report rates over brand-naming events, so a brand named in two answers counts twice, which is the right unit for "of all the times AI names a brand, how often does the name become a recommendation."
Naming is nearly automatic; the top slot is scarce
The clearest way to see the gap is per answer. A typical answer (the median) names 5 brands, places 4 of them somewhere in its ranked recommendation, keeps 2 in the top three, and, by construction, puts exactly 1 first. The average answer names 5.45 brands. So the model is generous with names and stingy with the top slot: it will happily list half a dozen options, but the question a buyer actually cares about, who is first, has one answer.
| Rung in a typical answer | Brands (median) | What it means |
|---|---|---|
| Named | 5 | Brands mentioned in the answer text |
| Recommended (any rank) | 4 | Brands the answer put in its ranked list |
| Top three | 2 | Brands in the first three positions |
| Ranked first | 1 | The single brand AI leads with |
This is why a "mention" count flatters. If your dashboard tells you AI mentioned your brand, you are one of about five, and four times out of five you are not the brand it led with. The distinct-brand view says the same thing from the other direction: of the 105,673 brands named across the panel, 72,186 (68.3%) earned at least one recommendation slot at some point, but only 20,820 (19.7%) ever held the number-one position in any answer. Four out of five named brands never once got to be the pick.
If you want to see how AI engines describe your own brand, run a free brand check — it takes a minute.
Most named brands have nothing behind them
The third rung is the one buyers should care about most, and it is the weakest. Only 35.7% of named brands were backed by a source the model cited. The majority of the time, AI names a brand from its trained memory, with no live source attached. That matters because a recommendation grounded in a citation is one a buyer can check, and one that other content can influence; a recommendation pulled from parametric memory is neither. It also means the "earned media drives AI mentions" story is only half true: earned media clearly shapes the model's memory over time, but in the moment of answering, most named brands are recalled, not retrieved.
The two engines split sharply here. Google AI Mode backed 45.3% of its named brands with a citation; ChatGPT Search backed only 28.1%. AI Mode leans on live retrieval and shows its work more often; ChatGPT Search leans more on memory. For a brand, that is a practical fork: the work that moves AI Mode (fresh, well-structured, citable sources) is different from the work that moves ChatGPT Search (durable third-party presence that becomes part of what the model knows).
The two engines disagree on how much to recommend
The engines are not symmetric on the recommendation rungs either, though they converge on the top slot. ChatGPT Search named more brands per answer overall, but Google AI Mode converted naming into a top-three slot more often (40.2% versus 32.2%) and into a citation far more often. Both, however, put a brand first at almost the same rate, around one named brand in seven to eight.
| Funnel rung | ChatGPT Search | Google AI Mode |
|---|---|---|
| Named (events) | 414,663 | 329,335 |
| Recommended, any rank | 71.1% | 74.7% |
| Top three | 32.2% | 40.2% |
| Ranked first | 11.5% | 14.6% |
| Backed by a source | 28.1% | 45.3% |
The reading: ChatGPT Search keeps a longer named list with a looser tie to sources, so a mention there is cheaper and a citation rarer. Google AI Mode runs a tighter, more source-anchored list, so a mention there is closer to a real recommendation and far more likely to rest on something a brand can influence. If you only watch one number, you will misread both engines: ChatGPT looks generous (lots of mentions) but is thin (few citations), and AI Mode looks stricter but is the one where a mention is actually worth more.
The gap is structural, not a category quirk
A reasonable objection: maybe the mention-to-recommendation gap is just a feature of crowded consumer categories, and in tight B2B verticals a mention really does mean a recommendation. It does not. We split the named-to-top-three conversion across 30 industries with at least 3,000 named events each. The rate barely moves: from 42.0% in Apps and 41.9% in Gaming at the top, to 31.7% in Consumer Electronics and 29.6% in Clothing and Apparel at the bottom. Every industry sits in a narrow band where roughly a third of named brands reach the top three. The source-backing rate is similarly flat (33% to 42%).
| Industry | Named-to-top-three | Backed by a source |
|---|---|---|
| Apps | 42.0% | 35.4% |
| Sales and Marketing | 41.5% | 40.0% |
| Financial Services | 40.0% | 37.2% |
| Software | 39.8% | 33.7% |
| Health Care | 37.7% | 37.8% |
| Commerce and Shopping | 35.3% | 34.5% |
| Consumer Electronics | 31.7% | 38.2% |
| Clothing and Apparel | 29.6% | 33.1% |
The narrowness is the finding. If the gap between being named and being recommended were a category effect, you would expect it to swing widely; instead it holds within about 12 points across everything from gaming to apparel. That tells you the gap is a property of how AI answers are built, not of any single market. Wherever your brand competes, expect to be named roughly two to three times as often as you are put in the top three, and treat the two as different goals.
What this changes about measuring AI visibility
The practical move is to stop reporting one "AI visibility" number and start reporting the funnel. Ask three questions of your own data. First, how often is your brand merely named versus actually recommended? A high mention count with a low top-three rate is a brand AI knows about but does not pick, which is a content and positioning problem, not a coverage one. Second, how often does a recommendation of your brand rest on a cited source versus memory? Source-backed recommendations are the ones you can move; memory-based ones require the slower work of becoming better known. Third, are your competitors converting mentions to the top slot at a higher rate than you, in the same answers? That is the gap that actually costs deals, and it is invisible to any tool that counts mentions and stops there.
None of this means mentions are worthless. Being named is the entry ticket; you cannot be recommended without it, and the distinct-brand data shows two-thirds of named brands do eventually earn some recommendation footing. But the report that treats a mention as a recommendation is overstating your position by a wide margin, and hiding the one number that maps to buyer behavior: whether AI puts you first.
How Parse measures the mention-to-recommendation gap
Parse tracks AI visibility across the engines buyers actually use, separating each brand's mentions, recommendations, and citations rather than collapsing them into one score. The public index spans more than 4.7 million AI responses, 603,000 brands, and 57 million citations. The pattern in this study is the market-wide picture; your brand has its own funnel. Parse's Brand Lookup shows how often AI names your brand, how often that name becomes a top-three recommendation, and how often a real source sits behind it, so you can see where you leak between being mentioned and being picked. For the related splits, see single-engine brands and why most AI winners depend on one platform, how ChatGPT and Google AI brand recommendations differ, and why one AI visibility score is misleading.
Is a brand mention the same as an AI recommendation?
No. In Parse's data, of 743,998 brand mentions across ChatGPT Search and Google AI Mode in June 2026, 72.7% held some recommendation rank but only 12.9% were the brand AI ranked first. A typical answer names about 5 brands and puts just 1 first, so a mention usually means you were listed, not picked.
What percent of brands AI mentions actually get recommended?
About 73% of named brands earned some position in the answer's ranked recommendation, and 35.7% made the top three. But more than a quarter (27.3%) were named and then never recommended at all, and only 12.9% were ranked first.
How often is a brand named by AI but not backed by a source?
Most of the time. Only 35.7% of named brands had a cited source attributed to them; the rest were recalled from the model's memory. The rate is much higher on Google AI Mode (45.3%) than on ChatGPT Search (28.1%), because AI Mode leans more on live retrieval.
Does the mention-to-recommendation gap depend on my industry?
Barely. Across 30 industries the share of named brands that reach the top three ranges only from 29.6% to 42.0%. The gap between being named and being recommended is a property of how AI builds answers, not of any single category, so every brand should plan for it.
Why does this matter for AI visibility tracking?
Because counting mentions overstates your position. A tool that logs "AI mentioned your brand" cannot tell you whether you were the pick (about 1 in 8 named brands) or one of five also-rans. Tracking mentions, recommendations, and citations as separate rates shows where you actually stand and where you leak between them.