- On was the most-mentioned brand we tracked, with 3,646 mentions, yet it was described positively just 64% of the time, the lowest of the four.
- Nike had far fewer mentions (1,672) but the warmest treatment of the big brands at 82% positive.
- Mention count and sentiment did not move together. Being named often is not the same as being recommended.
- The same brands scored 87% positive on Gemini but only 70% on ChatGPT. Tone depends heavily on the engine.
Does being mentioned more often mean AI recommends you more?
No. Across 258 AI answers about running and lifestyle sneakers, mention volume and positive sentiment pulled in opposite directions.
On was named far more often than Nike, 3,646 mentions to 1,672, yet On was described positively only 64% of the time versus Nike's 82%. More mentions did not buy warmer treatment.
Sentiment is the tone an AI engine takes when it names your brand: positive when it reads as a recommendation, neutral when it is a plain factual mention, and negative when it warns the reader off. Here is how the four tracked brands compared over 23 days of daily scans.
| Brand | Answers mentioned (of 258) | Total mentions | Positive tone |
|---|---|---|---|
| On | 228 | 3,646 | 64% |
| Adidas | 222 | 3,078 | 84% |
| Nike | 242 | 1,672 | 82% |
| Puma | 56 | 326 | 80% |
Read the mentions column and On is the runaway leader. Read the sentiment column and On is dead last.
That mismatch is the whole point. If you only counted how often each brand appeared, you would crown the brand the engines were actually least enthusiastic about (leadrescue.app tracking, June to July 2026).
What is the mention-sentiment gap?
The mention-sentiment gap is the difference between how often an AI engine names your brand and how positively it describes you. A brand can top the mention count and still sit at the bottom for warmth.
On is the clearest example in our data. Of its 228 mentioned answers, 36% were neutral, meaning the engine listed On factually without endorsing it.
On ChatGPT specifically, On's mentions were positive only 56% of the time, with the rest reading as plain description. The engines kept slotting On into "rising brand" and technical explainer passages: accurate, frequent, and flat.
Reading the recorded reasons, the neutral mentions were factual accounts of On's design and market rise rather than any recommendation. Nike and Adidas, by contrast, got pulled into "best of" and "most comfortable" framing far more often, which the engines scored as positive.
The lesson for a smaller brand is uncomfortable. You can earn a flood of AI mentions and still be described in a tone that does not sell anything, which is why AI brand visibility has to include tone, not just presence.
Which AI engine describes brands most positively?
Gemini was the warmest by a wide margin. Across the same tracked brands, Gemini rated 87% of mentions positive, while ChatGPT sat at just 70% and Perplexity at 72%.
That is a 17-point spread on identical brands answering identical prompts. The engine you check changes the tone you see.
Gemini is not a small sample either. Google reported that its Gemini app surpassed 750 million monthly active users by February 2026, so its warmer framing reaches a very large audience.
Negative sentiment was almost nonexistent across every engine. Out of hundreds of tracked brand mentions, only three were scored negative, so the real contest is positive versus neutral, not good versus bad.
What makes AI describe one brand more positively than another?
AI engines describe you using the third-party material they can find about you. Where that material is full of praise, the tone comes out positive. Where it is mostly specs and factual coverage, the tone comes out neutral.
Third-party citations are mentions of your brand on sites you do not own, like reviews, forums, and news. Brands with lots of independent "best running shoe" coverage get summarized warmly, because the engine is paraphrasing an existing recommendation.
This is where structure matters too. Researchers Aggarwal and colleagues, who coined the term "Generative Engine Optimization" in their 2023 study, found that content which is structured, cited, and quotable can lift a brand's visibility in AI answers by up to 40%.
The same clean, quotable material also gives the engine better sentences to lift when it describes you, which nudges tone in your favour. If a rival keeps getting the warmer write-up, our guide on why ChatGPT recommends your competitor covers the usual causes.
How do you track AI sentiment over time?
You track it by running the same buyer prompts across ChatGPT, Gemini, and Perplexity on a schedule, then logging not just whether you appear but the tone of every mention. A single check is a snapshot, and sentiment shifts as coverage changes.
Our numbers here came from 45 prompts run daily across all three engines for 23 days, which is what makes the mention-sentiment gap visible instead of a one-day fluke.
Lead Rescue runs those daily scans across all three engines and records how each one describes you, so you can watch tone move, not just presence. If an engine starts publishing wrong or cold descriptions, our guide on fixing wrong information AI engines say about your brand walks through the fix.
Is a neutral AI mention bad for my brand?
Not bad, but weaker than it looks. A neutral mention means the engine named you factually without recommending you, so a buyer reading that answer sees you exist but gets no reason to pick you. In our data, On collected thousands of mentions yet 36% were neutral, which blunts the value of all that visibility.
Can a small brand get warmer AI sentiment than a market leader?
Yes. Sentiment tracks the tone of your coverage, not your company size, so a focused brand with glowing niche reviews can outscore a giant with mostly factual coverage. Adidas edged out Nike on positive tone in our tracking despite fewer total mentions, showing the ceiling is set by how you are written about, not how big you are.
Which matters more for GEO, how often AI mentions you or how it describes you?
Both, in sequence. Mentions get you into the answer, and sentiment decides whether that appearance helps or just fills space. Chasing mention count alone can leave you widely named but coldly described, which is exactly the trap On fell into across 258 answers, so treat them as two metrics rather than one.