The brand came up somewhere 185 times, and everywhere 60 times
If ChatGPT names you for a question, does Perplexity? Most tracking assumes the answer is roughly yes, because the per engine rates usually look alike. We checked it directly.
Between 30 September and 4 October 2026 our sampler put the same unbranded buyer questions to four engines on the same day: our ChatGPT engine on OpenAI's API, Google AI Overviews, Perplexity and Claude. That gave us 576 question days, meaning one question answered by all four engines on one day, across 292 questions and 14 brand setups in 13 categories. For each answer we recorded which tracked names it mentioned: the brand and the 5 to 11 competitors it tracks. Read every number here as ours.
- 391 of the 576 question days, 67.9%, had no engine name the brand at all.
- 60, 10.4%, had all four name it. Interval 8.2% to 13.2%.
- 125, 21.7%, were split: one, two or three engines named the brand and the rest didn't.
- So of the 185 question days where at least one engine named the brand, all four agreed on 60. That's 32.4%, with a 95% interval of 26.1% to 39.5%.
Three engines scored 22% to 23% and still disagreed
Here's why a per engine score misleads. ChatGPT named the brand in 23.3% of these question days, AI Overviews in 23.1% and Claude in 22.2%. Perplexity was lower at 18.1%. Put those side by side on a dashboard and three of them look interchangeable.
They aren't. Those similar rates came from different questions. If you watched only one engine, this is how many of the 185 brand mentions you'd have seen:
- ChatGPT: 134, 72.4%. Interval 65.6% to 78.4%.
- AI Overviews: 133, 71.9%.
- Claude: 128, 69.2%.
- Perplexity: 104, 56.2%. Interval 49.0% to 63.2%.
- The best pair, AI Overviews with Claude, caught 167, or 90.3%. No set of three caught more than 178, 96.2%.
30% of the names we tracked came from one engine only
The brand is one name. Widen it to every tracked name, brand and competitors together, and the pattern holds. Across the 576 question days a tracked name was mentioned 1,818 times, counting each name once per question day.
- 545 of those, 30.0%, were named by exactly one engine. Interval 27.9% to 32.1%.
- 570, 31.4%, were named by all four.
- The other 38.6% were named by two or three.
- Only 105 question days, 18.2%, had all four engines name exactly the same set of tracked names, with at least one name in it.
Each engine agreed with itself more than with the others
The fair objection is that these engines are random, so of course two answers differ. SparkToro and Gumshoe showed in January 2026 that the same engine almost never repeats itself: they put the chance of ChatGPT or Google's AI giving the same list of brands twice at under 1 in 100. So we need a baseline.
We used each engine against itself a day later. For every question and engine we compared the tracked names on one day with the next day, and measured overlap as shared names divided by all names either answer gave.
- Same engine, next day: overlap between 0.62 (Perplexity) and 0.69 (ChatGPT). The yes or no on the brand matched between 87.6% and 92.7% of the time.
- Two different engines, same day: overlap between 0.50 (AI Overviews with Perplexity) and 0.57 (ChatGPT with AI Overviews).
- When three engines named the brand and one didn't, the odd one out was Perplexity 22 times out of 46.
What we'd actually do
Cheapest first.
- Ask for the overlap as well as the rate. A report that says you're at 23% on ChatGPT and 22% on Claude should also say how many questions both of them named you on. If it can't, it can't tell you whether you have one audience or two.
- Track at least the two engines your buyers use most. In our sample one engine alone missed 28% to 44% of the brand's appearances.
- Treat a split question as the work list. A question where three engines name you and one doesn't is the cheapest win you have, because the content clearly exists and one retrieval path isn't finding it. Read that engine's cited sources for that question first.
- Don't average engines into one score before you've looked at them apart. BrightEdge found in August 2025 that ChatGPT, AI Overviews and AI Mode gave the same brands on only 17% of queries. A blended number assumes the opposite.
Where this could be wrong
Five days is short. Perplexity and Claude only started returning answers we can score on 30 September, and most question days come from 3 and 4 October.
One category dominates. Mattress retail accounts for 129 of the 185 question days where the brand was named. Take it out and all four agreed on 15 of 56, 26.8%, with an interval of 17.0% to 39.6%. Lower, so it isn't propping up the headline, but the interval is wide.
We only count tracked names, the brand and its listed competitors. An engine that recommends ten other companies registers as naming nobody, so overlap among all brands mentioned could be higher or lower than ours.
Our ChatGPT and Claude answers come through the providers' APIs, not the consumer apps. SparkToro's data suggests API and app variance may be similar, but they flag it as an open question and so do we.
Name matching is automated. A missed alias would count as disagreement that isn't real, though it would hit every engine alike.
