Every category in the October 2026 edition was put to both engines with the same questions, which makes it possible to count how many brands the two engines have in common. The engines are OpenAI's API with web search (gpt-6-luna) and Perplexity's Agent API with the Sonar model (perplexity/sonar) and its web search tool; below they are "OpenAI" and "Perplexity". The numbers describe answers to prompts the index wrote and collected through the APIs. They do not describe what anyone saw in a consumer app, which can differ.
How is the overlap measured?
For each category, take the set of brands OpenAI's answers recommended and the set Perplexity's answers recommended. The overlap is the number of brands in both sets divided by the number of brands in either set (a Jaccard index). Each category counts equally, whatever its size. It compares brand lists, not individual answers: two engines can recommend the same brands in different answers and still score 100%.
Two real categories show the range, with their answer counts (examples, not findings). In espresso machines, 8 answers, OpenAI recommended 7 brands and Perplexity 9; 6 brands were in both and 10 in either, so the overlap was 60.0%. In bed frames, 8 answers, OpenAI recommended 7 and Perplexity 8; only 2 were in both and 13 in either, so the overlap was 15.4%.
All figures here exclude the alternative-to-a-brand question. In the October 2026 edition that question names a brand in the prompt, and its answers recommend 5.0 brands on average, against about 2.3 for the other four questions, so mixing it in changes what is being counted. That leaves 1,271 of the 1,587 collected answers: about 8 per category, 4 from each engine (9 of the 160 categories have 7 because a Perplexity answer failed). With the alternative question included, the mean overlap reads 53.6% over about 10 answers per category. Read that figure only next to the 47.0% below.
What does the overlap look like?
Across the 160 categories (US 100, Mexico 30, Canada 30), the mean overlap was 47.0% and the median 46.45%. In the average category, 3.9 brands were recommended by both engines and 5.3 by only one of them, 9.2 distinct brands in all.
The average hides a wide spread. One category shared no brand at all, 159 shared at least one and 135 shared three or more. Six categories had identical brand lists.
| Overlap in the category | Categories (of 160) |
|---|---|
| Under 25% | 19 |
| 25% to under 50% | 63 |
| 50% to under 75% | 59 |
| 75% or more | 19 |
OpenAI and Perplexity, October 2026 edition, excluding the alternative-to-a-brand question: 160 categories, about 8 answers each.
By market, the means are close:
| Market | Categories | Mean overlap | Brands in both | Brands in one only |
|---|---|---|---|---|
| United States | 100 | 47.5% | 3.8 | 5.3 |
| Mexico | 30 | 46.3% | 3.7 | 5.3 |
| Canada | 30 | 45.9% | 4.0 | 5.2 |
Mean per category, October 2026 edition, excluding the alternative-to-a-brand question.
The markets use different categories and prompts, and Mexico's are in Spanish, so this table is descriptive. A gap of one or two points is small beside the spread between categories.
If a brand shows up in one engine, how often does it show up in the other?
Count each brand once per category. OpenAI recommended 991 brand-category pairs in these answers and Perplexity 1,092; 617 pairs were in both. So 62.3% of OpenAI's pairs also appeared in Perplexity's answers, and 56.5% of Perplexity's pairs also appeared in OpenAI's. Perplexity recommended slightly more distinct brands per category: 6.8 against 6.2 for OpenAI, on a similar number of answers.
Seen the other way, 617 of the 1,466 pairs that either engine recommended (42.1%) were in both. Roughly three in five of one engine's pairs turned up in the other engine's answers, and about two in five did not. That is a pattern across 160 categories. It is not a probability for any single brand: the data counts brand-category pairs and does not separate brands that many answers recommend from brands recommended once.
Does overlap depend on how many brands a category has?
In this data, overlap falls as the number of brands in a category rises.
| Brands recommended in the category (either engine) | Categories | Mean overlap |
|---|---|---|
| Up to 5 | 20 | 72.0% |
| 6 to 7 | 43 | 54.4% |
| 8 to 9 | 34 | 45.0% |
| 10 to 11 | 25 | 37.7% |
| 12 or more | 38 | 33.3% |
October 2026 edition, excluding the alternative-to-a-brand question: 160 categories, about 8 answers each.
The groups are formed by the number of brands either engine recommended, which is also the denominator of the overlap, so the two move together by construction. Part of this is arithmetic. When the answers recommend only four brands in total, two engines that each answer four questions will recommend most of them. When many brands are credible, four answers per engine sample only a few. So a low overlap in a crowded category says little by itself about the engines. It is the pattern to expect.
Are the engines' source lists comparable?
Not directly. The index records sources differently for each engine: OpenAI's list is the pages its answer links (2.2 pages per answer on average across its 800 answers, 2.0 distinct domains); Perplexity's is the pages its search retrieved (27.3 URLs per answer on average across its 787 retrieved source lists, 15.2 distinct domains), not all of which the answer uses. Every question type is included in these source figures, unlike the overlap figures above. Because the two lists measure different things, they cannot show whether the engines drew on the same pages, or whether sources explain any of the overlap. The post on the websites behind these answers lists each engine's most frequent domains on its own denominator.
What overlap does not tell you
- It is not agreement between answers. In the October 2026 edition each engine has about 4 answers per category, and one answer is a single draw. This edition has no repeat runs of the same engine on the same prompt, so it cannot say how much of the brand lists that were not shared comes from the engines and how much from chance.
- It is not accuracy. A low overlap does not mean one engine is wrong.
- It is the API, not the app. Apps can add personalization, memory, location and their own product interfaces. See API answers vs app answers.
- It depends on the brand lists. An AI model extracts the brands each answer recommends, and a hand-made merge list joins name variants (methodology). A variant nobody has reviewed splits a brand and lowers overlap.
- It is one sample. 160 categories, asked once in October 2026, with 13 Perplexity answers failing and left out.
What should a brand do with this?
- Track each engine on its own. A brand in 3 of 4 OpenAI answers can be in none of Perplexity's. Never average the two into one share.
- Ask who else is in the category. In the 20 categories where the answers recommended five brands or fewer, the two engines' brand lists overlapped by 72.0% on average (about 8 answers each). In a crowded category, expect less overlap, so one engine's shortlist is a thin guide.
- Compute your own overlap for a category you care about:
- Write the same four questions for each engine: best overall, a budget with a price ceiling, premium, and a specific use case.
- Ask each question in each engine and list the brands each answer recommends.
- Count the brands in both engines' lists and the brands in either.
- Divide the first by the second. Repeat the questions on another day; one pass is a draw.
- Look at sources per engine, not together. If you appear in one engine and not the other, compare each engine's source list before concluding anything. A domain in a list does not show that it influenced the recommendation.
For how many answers make a difference trustworthy, see how many AI answers you need and how to measure your brand's AI visibility. The edition's other headline numbers are in What 1,587 AI shopping answers show, and the full tables are in the October 2026 report.
Frequently asked questions
Can I use one engine as a stand-in for the other?
Not on this data. In the October 2026 edition, 62.3% of OpenAI's brand-category pairs also appeared in Perplexity's answers and 56.5% of Perplexity's pairs also appeared in OpenAI's, across 160 categories, so a shortlist from one engine would miss roughly two in five of the other engine's pairs. Measure your brand in each engine.
Should I focus on one engine first?
The data cannot answer that, because it covers answers, not traffic or sales. Measure your own share in each engine first, then decide from the size of the gap and where your customers actually ask.
How we write: drafted with AI tools from Dopbot Index data and public sources; every number and outside fact is traced to its source in a claims record and checked before publishing. More about how we write; corrections to hello@dopbot.com.