- Nike led with a 94% appearance rate across 261 answers, followed by On (89%) and Adidas (86%).
- Under Armour was named in zero answers. Reebok managed 3%. Brand size did not predict AI presence.
- On, founded in 2010, out-appeared Adidas, founded in 1949, but only on two of the three engines.
- The leader changed by question. On the walking-comfort prompt, Hoka hit 100% and pushed Nike down to 82%.
How did we measure which brands AI recommends?
We ran 5 buyer-intent sneaker prompts through ChatGPT, Gemini, and Perplexity once a day for 19 days, from 26 June to 23 July 2026. That produced 261 usable answers, 87 per engine, and we recorded every brand named in every one.
The metric is appearance rate. Appearance rate is the share of answers that name a brand at least once.
What did we track, and for how long?
Five prompts, three engines, 19 consecutive days. We deliberately used full buyer questions rather than keywords, because that is how people actually talk to an AI assistant.
The prompts covered streetwear styling, light running, all-day walking comfort, a general "best brands" question, and one head-to-head comparison naming Nike, Adidas, and On directly.
What counts as a brand being recommended?
A brand counts once per answer if the engine names it anywhere in the response. We count answers, not raw mentions, so a brand repeated eight times in one paragraph does not outrank a brand named once in eight separate answers.
Note: appearing in an answer is not the same as being cited. A mention is your name in the text; a citation is a link to your page as the source. We measured the first, and covered the gap between them in a separate study on what sources AI engines cite.
How did we stop product names from inflating the results?
This mattered more than expected. The raw extraction returned 726 distinct entities, but most were not brands: product models like Vomero and Samba, cushioning technologies like CloudTec and Boost, and even cited publishers such as tomsguide.com.
We verified each high-frequency entity against the actual answer text and folded models into their parent brand. Nike's Vomero line, for example, was confirmed from a tracked Gemini answer describing the "Nike Ava Rover" as offering "advanced ReactX cushioning".
One correction worth naming: the entity "Ellipse" looked like a Nike model but the answers were explicit, calling it "the New Balance Ellipse" with a "Fresh Foam X midsole". It counts for New Balance. Guessing would have overstated Nike.
Which brands does AI recommend most often?
Nike topped the list at 94%, appearing in 245 of 261 answers. On reached 89% and Adidas 86%. After the top three there is a clear step down: New Balance at 64%, ASICS at 62%, and Hoka at 57%.
Below that, presence thins fast. Salomon managed 30%, Puma 22%, and the bottom two barely registered.
Why does the top brand only reach 94%, not 100%?
Because AI answers are not a ranking, they are a shortlist assembled per question. Nike missed 16 of 261 answers, mostly on the walking-comfort prompt where specialist brands crowded it out.
No brand hit 100% overall. Even the three brands named directly in one of our prompts only reached 100% on that specific prompt.
Which household-name brands does AI ignore?
Under Armour was named in zero of 261 answers. Reebok appeared in 9, a 3% rate. Both are globally recognised sportswear brands with decades of history, and both were effectively invisible in the AI answers our prompts produced.
Puma, by contrast, still managed 22%. So this is not simply "big brands lose to specialists".
What does zero mentions actually mean for a brand that size?
It means that for these five buying questions, across three engines and 19 days, an AI assistant never once put Under Armour in front of a shopper. Not ranked low. Absent.
Reach matters here because AI is now a real research channel. The Pew Research Center found 34% of US adults had used ChatGPT by mid-2025, roughly double two years earlier (Pew Research Center, 2025).
Is a 3% mention rate better than zero?
Barely, and not in a way you can build on. Reebok's 9 appearances were spread thinly: 1 on ChatGPT, 5 on Gemini, 3 on Perplexity, out of 87 answers each.
At that rate a brand cannot tell a real gain from noise, which is the same problem we found when measuring how much AI citations change from day to day.
Can a newer brand out-appear an established giant?
Yes. On, founded in Switzerland in 2010, appeared in 231 answers against Adidas's 225. That is 89% versus 86%, a narrow but consistent lead for a brand roughly 60 years younger than its rival.
The gap is small enough to treat as a tie in absolute terms. What makes it interesting is that it happened at all.
Where does the challenger's lead come from?
Editorial and review coverage, mostly. On appeared in 95% of ChatGPT answers and 97% of Gemini answers, the two engines that leaned hardest on publisher content in our earlier tracking.
This fits published GEO research. A Princeton-led study found that clear structure, statistics, and citations lifted a source's visibility in AI answers by up to 40% (Aggarwal et al., Princeton, 2023). Brands covered heavily by structured review sites inherit that advantage.
Where does it break down?
On Perplexity. On's appearance rate fell to 74% there, its weakest engine by 21 points, while Adidas held 84%.
So the challenger's win is not durable across the board. Pick a different engine and the leaderboard changes.
Do all three AI engines recommend the same brands?
No, and the differences are large. Gemini named more brands more often than either rival, while ChatGPT was the most selective. Puma appeared in 32% of Gemini answers but only 9% of ChatGPT answers, a three-and-a-half-fold gap for the same brand on the same questions.
| Brand | ChatGPT | Gemini | Perplexity | Spread |
|---|---|---|---|---|
| Nike | 92% | 98% | 92% | 6 pts |
| On | 95% | 97% | 74% | 23 pts |
| Adidas | 79% | 95% | 84% | 16 pts |
| New Balance | 60% | 77% | 56% | 21 pts |
| ASICS | 56% | 77% | 53% | 24 pts |
| Hoka | 46% | 72% | 52% | 26 pts |
| Brooks | 43% | 37% | 55% | 18 pts |
| Salomon | 20% | 44% | 25% | 24 pts |
| Puma | 9% | 32% | 24% | 23 pts |
| Reebok | 1% | 6% | 3% | 5 pts |
| Under Armour | 0% | 0% | 0% | 0 pts |
Which engine names the most brands?
Gemini. It gave the highest appearance rate to 8 of the 11 brands we verified, including every brand outside the top three.
ChatGPT was the harshest on mid-tier brands, dropping Salomon to 20% and Puma to 9%. If you are a challenger brand, ChatGPT is the hardest room.
Which brands swing most between engines?
Hoka moved most, a 26-point spread from 46% on ChatGPT to 72% on Gemini. ASICS and Salomon both swung 24 points.
Only Brooks did better on Perplexity (55%) than on Gemini (37%), the single inverted result in the table. Checking one engine and assuming the rest match is how brands end up with a false read on their visibility, which is why we tracked whether ChatGPT, Perplexity, and Gemini recommend the same brands separately.
Does the leading brand win every question?
No. Nike led four of the five prompts but lost the fifth outright. On the question about shoes for walking 10,000 or more steps a day, Hoka appeared in 100% of answers while Nike fell to 82% and Adidas collapsed to 55%.
That single prompt reordered the entire leaderboard.
| Prompt theme | Nike | On | Adidas | Hoka |
|---|---|---|---|---|
| Compare Nike, Adidas, On (brands named) | 100% | 100% | 100% | 0% |
| Best brands for performance and fashion | 100% | 86% | 100% | 84% |
| Everyday streetwear | 94% | 80% | 84% | 45% |
| Light running plus smart casual | 93% | 88% | 91% | 54% |
| Walking 10,000+ steps a day | 82% | 88% | 55% | 100% |
Which prompt flipped the leaderboard?
The walking-comfort one. Hoka went from 0% on the comparison prompt to 100% on the comfort prompt, the widest move any brand made.
Brands do not own categories in AI search. They own use cases, and the use case is set by how the buyer phrases the question.
What does that mean for choosing your prompts?
Your measured visibility depends heavily on which questions you track. Pick only the flattering prompts and you will report a win that no real buyer experiences.
Watch out: a single prompt is not a baseline. Nike looked dominant on four prompts and mid-pack on the fifth, from the same 19 days of data.
How should you use appearance rate for your own brand?
Treat it as a rate, not an event. Pick 5 to 10 questions your buyers genuinely ask, run them on all three engines repeatedly, and track the percentage of answers that name you.
- Measure per engine, never blended. A 23-point spread, as On had, disappears entirely in a single average.
- Track the use case, not just the category. Hoka's 100% came from one specific intent, and that is where a smaller brand can realistically win.
- Treat zero as a finding, not an error. Under Armour's 0% held across 261 answers and 19 days. That is a signal, not a glitch.
Doing this by hand across three engines is slow, which is what Lead Rescue's share-of-voice and competitor tracking automates: it runs your prompts daily and reports the percentage of answers naming you against each rival, per engine.
Adjacent questions
Is appearance rate the same as share of voice?
Close, but stricter. Share of voice usually divides your mentions by all brand mentions, which rewards a brand repeated many times in one answer. Appearance rate asks a simpler question: in what percentage of answers do you show up at all? We use the second because it cannot be inflated by repetition. Our explainer on share of voice in AI search covers the difference.
Does a high appearance rate mean AI links to your website?
No, and the gap is usually large. In our earlier sneaker tracking the most-named brand had its own website cited just twice across 165 answers. Being named and being cited are separate outcomes with separate causes, so a brand can dominate the text of AI answers while its own pages are almost never used as the source.
Why would AI name a brand it has no source for?
Because a general-purpose model already carries brand knowledge from training, separate from anything it retrieves live. That is why a famous brand can be named confidently in an answer whose citations all point at review sites and publishers rather than the brand itself.
Would these results hold for software instead of sneakers?
The mechanism should, the numbers will not. Sneakers have unusually rich third-party review coverage, so appearance rates skew high. In a smaller software category expect lower rates overall, plus the same two patterns: big engine-to-engine spreads, and specialists winning specific use cases.
How often should you re-measure?
Weekly at minimum, over a rolling window. Single checks are unreliable because engine outputs shift day to day, so a rate calculated across many runs is the only figure worth acting on.
To learn the underlying metric, start with what AI brand visibility means, or see the method in how to track which competitors AI engines recommend.
Lead Rescue runs your buyer questions across ChatGPT, Gemini, and Perplexity every day and reports what share of answers name you and each competitor, engine by engine. See your AI recommendation share →