Gemini, by a wide margin. Across 60 buyer-intent prompts GetIntel tracks for its own brand, checked over the 7 days ending 26 August 2026, Gemini named a competitor instead of GetIntel on 91.7% of prompts. Google AI Overviews did it on 76.3%, ChatGPT on 71.7%, and Perplexity on just 18.3%, though Perplexity also named no specific brand at all on 66.7% of the same prompts, so its low rate is closer to a no-answer problem than a win. This is GetIntel's own tracked data, about GetIntel's own brand, published as a worked example of how to read this kind of number for any brand.
Every figure below comes from GetIntel's own prompt-tracking data for its own brand ("GetIntel Founders"), read on 26 August 2026 over the trailing 7-day scan window. This measures four engines: ChatGPT, Google AI Overviews, Perplexity, and Gemini. Claude is tracked separately on GetIntel's Growth plan and isn't part of this 60-prompt set, so it's excluded here rather than guessed at.
Which Engine Recommends a Competitor Over You Most Often?
Four numbers, all from the same 60 buyer-intent prompts, all read 26 August 2026.
| Engine | Named a competitor, not us | Named nobody | Named us |
|---|---|---|---|
| Gemini | 91.7% (55/60) | 6.7% (4/60) | 1.7% (1/60) |
| Google AI Overviews | 76.3% (45/59) | 20.3% (12/59) | 3.4% (2/59) |
| ChatGPT | 71.7% (43/60) | 13.3% (8/60) | 15.0% (9/60) |
| Perplexity | 18.3% (11/60) | 66.7% (40/60) | 15.0% (9/60) |
Computed from GetIntel's own 7-day prompt-matrix data, read 26 August 2026. "Named a competitor" means the engine's dominant state across recent scans of that prompt was naming another brand, not GetIntel. Google AI Overviews has 59 rather than 60 because one prompt returned no answer to read (a failed render, not a verdict) and was excluded rather than counted either way.
By this measure, Gemini is the clearest case: on more than 9 prompts in 10, when it names anyone, it names a rival instead of GetIntel. ChatGPT and Google AI Overviews sit closer together in the 70s. Perplexity looks like the outlier worth explaining, since a low "named a competitor" number sounds like good news and mostly isn't.
What Do "Named a Competitor," "Named Nobody," and "Named Us" Actually Mean?
Three distinct outcomes per prompt, and only one of them is good.
Named a competitor (the "absent" state in GetIntel's own tracking language) means the engine answered with at least one brand name, and GetIntel wasn't one of them. This is the state that matters for the buyer question this piece answers: it's the engine actively choosing a rival over you, not just failing to mention anyone.
Named nobody (the "open" state) means the engine answered the buying question without naming any specific brand, ours or a rival's, an open slot rather than a lost one. This is a different kind of miss: nobody currently owns it, which is a different problem to fix than dislodging an established rival.
Named us (the "mentioned" or "recommended" state) is the only outcome where GetIntel actually appears in the answer.
Why Does Perplexity Look So Different?
Because Perplexity mostly doesn't commit to naming anyone, not because it favors GetIntel.
Perplexity's "named a competitor" rate (18.3%) is the lowest of the four, but its "named nobody" rate (66.7%) is by far the highest. Two-thirds of the time, Perplexity's answer to one of these 60 buying-intent prompts doesn't confidently name a specific tool at all, ours or otherwise. That's a wide-open slot, not a won one: a low competitor-favoring rate on an engine that mostly answers vaguely is not the same finding as a low competitor-favoring rate on an engine that commits to a name and picks you. Reading Perplexity's 18.3% on its own, without the 66.7% next to it, would overstate how well GetIntel is doing there.
Why Is Gemini the Worst of the Four?
Gemini's 1.7% named-us rate here, 1 mention out of 60 tracked prompts, is consistent with a pattern documented in more depth elsewhere on this site: Gemini went 919 consecutive answers without naming GetIntel once before its first-ever mention on 10 August 2026, and even now sits at roughly 1 mention per 1,280 runs. This piece's 91.7% "named a competitor instead" figure and that piece's near-zero mention rate are two views of the same underlying gap, not two separate findings: an engine that almost never names you is, by construction, an engine that almost always names someone else when it names anyone.
What Do You Actually Do With This?
Treat the four engines as four separate problems, not one aggregate score.
Gemini's gap looks like a recognition problem: an engine that has essentially never encountered the brand in its answers isn't going to start recommending it without something changing in what Gemini has to work with. Google AI Overviews and ChatGPT, both in the 70s, look more like a competition problem: the engines know competitors well enough to name them by default on these prompts, and diagnosing why AI recommends a competitor over you is a different exercise from introducing a brand an engine has never seen. Perplexity looks like an opportunity: two-thirds of its answers name nobody, which is the closest thing to a genuinely open slot in this data, and it's worth checking which of your own high-intent prompts land in that 66.7% before assuming Perplexity is a lost cause.
None of this generalizes to another brand's numbers without measuring that brand's own prompts the same way. The point of publishing the method alongside GetIntel's own figures is that the same four-state read (competitor, nobody, us, and the runs that failed to answer at all) applies to any brand's tracked prompts, and the answer for a different brand in a different category could easily look nothing like this one.
How Would You Compute This for Your Own Brand?
Three things, per prompt, per engine, over a real window rather than a single check.
Run the same set of buying-intent questions your actual buyers would ask, on a fixed schedule, through each engine's real interface rather than its API where possible. For each run, record one of exactly three outcomes: your brand was named, a competitor was named and you weren't, or nobody specific was named. Then roll that up per engine across at least a week of runs, not a single day, since any one run can land on an atypical answer. The per-engine split is the point: an aggregate "we're visible 40% of the time" number, blended across engines, would have hidden that Gemini and Perplexity are almost opposite problems for GetIntel specifically, and the same is likely true for most brands measuring this for the first time.