The word "instead" hides a counting problem. In the October 2026 sample, an answer recommended a mean of 2.3 brands while the answers to one category named a mean of 9.2, so most of a category's brands were missing from most of its answers, and a rival you saw in one answer may be missing from the next. The checklist below first separates a real gap from a lucky draw, then walks the places a real gap can sit.
Is the gap real?
Start with how crowded the answers are. In the October 2026 edition, excluding the alternative-to-a-brand question (1,271 answers), an answer recommended a mean of 2.3 brands. OpenAI's API with web search (gpt-6-luna) averaged 2.2 across 640 answers, and Perplexity, through its Agent API with the Sonar model (perplexity/sonar), averaged 2.5 across 631 answers.
Across the 160 categories, again excluding that question, the answers to a category named a mean of 9.2 distinct brands (median 8), with about 8 answers per category in the October 2026 edition. The most recommended brand in a category appeared in a mean of 62.3% of that category's answers, and a mean of 48.4% of a category's brands were named in a single answer only. These are patterns over 160 categories, not facts about any one category. They describe a familiar shape: a few names repeated, and a long tail that comes and goes.
So a rival named in your three test answers may be in the repeated group, or may be a one-off. A count settles it only with enough answers. The 95% Wilson intervals below use illustrative counts, with a rival named about half the time and your brand named rarely:
| Answers asked | Rival named | Your brand named | Rival's plausible true rate | Yours |
|---|---|---|---|---|
| 8 | 4 | 1 | 21.5% to 78.5% | 2.2% to 47.1% |
| 20 | 10 | 3 | 29.9% to 70.1% | 5.2% to 36.0% |
| 40 | 20 | 5 | 35.2% to 64.8% | 5.5% to 26.1% |
As a conservative rule of thumb, overlapping intervals mean you cannot yet call the gap real. Here the ranges separate only at 40 answers. Ask each question at least 20 times per engine for a first reading, and use 40 when you need to compare. The post on how many answers to trust explains the method.
Which question type does the gap sit in?
Run all five question types, not only "best overall": best overall, a budget with a price ceiling, premium, a specific use case, and an "alternative to" question.
The October data says the shortlists differ by type. Per category, the brands in the budget answers and the premium answers overlapped by a mean of 8.2% (brands named in both the budget and premium answers, divided by brands named in either, averaged over categories; usually two budget and two premium answers, 155 of 160 categories), and 70% of categories had no brand in common between the two. The brands in the best-overall answers overlapped those in the budget answers by a mean of 23.8% and those in the premium answers by 26.8%, with the same measure (brands in both lists divided by brands in either, averaged over categories). That is observed shortlist overlap, not proof that price wording causes anything.
What it means for you: a rival that is named often in your "best overall" test may be absent from "budget". Record the grid, so you know which cell to work on. The budget vs premium post has the detail.
"Alternative to [brand]" questions are a different game. In the October 2026 edition, the alternative-to-a-brand question returned 5.0 brands per answer across 316 answers (both engines together), and the named brand was itself recommended in 21 of the 296 answers where we found it. If the question names your rival, expect an answer that rarely recommends the rival and lists many others. If the question names your own brand, an answer full of other brands is what the question asks for, so test the "alternative to" question with your rival's name instead. The alternatives post covers it.
Which engine is it?
Measure each engine on its own. Excluding the alternative question, the two engines' brand lists overlapped by a mean of 47.0% per category (median 46.45%): brands both recommended divided by brands either recommended, averaged over 160 categories with equal weight. It is not answer-by-answer agreement. In 28 of 160 categories the overlap was 25% or lower, in 19 it was 75% or higher, and in 1 the engines shared no brand.
If your rival shows up on one engine and you are fine on the other, the diagnosis is engine-specific, and so are the sources to check next. The overlap post explains why one engine tells you little about the other.
Which source pages sit behind the answers?
For each answer where the rival is named and you are not, list the source URLs and group them by domain. The two engines record sources differently, so keep them apart:
- OpenAI's source list is the pages its answer links: 2.2 per answer on average (800 answers), with 2.0 distinct domains.
- Perplexity's source list is the pages its search retrieved, not all of which the answer uses: 27.3 URLs per answer on average (787 answers), with 15.2 distinct domains.
Then ask three questions. Is the rival's own site in the list and yours is not? An own-site match is a domain whose first label equals a recommended brand's name; it misses country domains and other store names. The October figures cover all recommended brands, not the rival or you in particular: OpenAI linked an own-site match in 46.6% of answers and Perplexity's retrieved lists held one in 62.0%. These source figures cover all 800 OpenAI and 787 Perplexity answers, including every question type and all three markets. Does the list include an editorial review of the rival that you do not have? A Wirecutter page was among the retrieved sources in 333 of 787 Perplexity answers, and 0 of 800 OpenAI answers linked nytimes.com. Is a retailer page in the list? Perplexity's retrieved lists held a retailer domain in 76.7% of answers, and OpenAI's linked pages in 17.6%, using a hand-made list of 49 retailer domains, so it understates retailer pages.
A domain in a list is not proof that it influenced the recommendation. Treat gaps as places to look, not as causes.
Is your name getting split?
Search your own answers for variants of your brand name: a shortened form, a longer legal name, a product-line name. A split name looks like a smaller share than you have. Count the variants together before you compare yourself with a rival.
What do the platforms say about price, reviews and retail?
As of October 2026, OpenAI, Perplexity and Shopify describe how products are chosen or scored in their shopping features. These are descriptions of the consumer apps and Shopify's channels, which we do not sample, but they list points you can check against a rival.
- OpenAI's shopping help says ChatGPT may consider available options, price, reviews and ease of use, and any context the shopper provides, and that merchants are ranked by factors such as availability, price, quality and whether they make or primarily sell the item.
- Perplexity says that, as it ranks sources by authority and relevance, it ranks product listings with similar criteria, and that it is more likely to recommend merchants who provide deeper product details such as availability, reviews, pricing and specifications.
- Shopify's listing quality indicators for AI channels include description length, images, review rating and count from verified sources, variant data with in-stock availability, and shop policies. Shopify adds that search relevance also depends on signals such as popularity, customer engagement and brand recognition, which accumulate over time.
Compare your product page, retailer listings and review profile with the rival's on exactly those points.
What should you change for each finding?
| Finding | Where to work | Caution |
|---|---|---|
| Absent from budget answers only | A page and retailer listings built around the price ceiling and the use | A budget answer may name only two brands |
| Absent on Perplexity only | Editorial review pages and retailer pages, the types we counted in its retrieved lists | Retrieved is not the same as used |
| Absent on OpenAI only | Your own product pages, crawl access, and being linked | OpenAI links about 2 pages per answer |
| Named under variants | Consistent naming across pages and feeds | Merge variants before counting |
| Rival has reviews you lack | Earn reviews and editorial coverage within each site's rules | No review count is a guarantee |
| Absent from "alternative to [rival]" answers | A fair comparison page that answers that question | If the question names your own brand, other brands are the expected result; in the October 2026 prompts the named brand was recommended in 21 of 296 answers |
Change one thing at a time, then repeat the same questions with the same settings and compare the intervals. Nothing here guarantees a recommendation, and the data shows association between pages and answers, not cause.
What can this data not tell you?
- It is the API, not the app. The apps can differ in personalization, location and product interfaces. API vs app answers explains the gap.
- Per-category counts are small. About 8 answers per category in October 2026, from prompts we wrote.
- One edition. The October 2026 numbers are one sample, and we make no claim about change over time.
Frequently asked questions
Can a competitor pay to be recommended?
As of October 2026, OpenAI says its product results are selected independently, are not ads and are not influenced by its partnerships, and that ads are separate. Perplexity says advertisers cannot pay to list products in its related products section; that statement covers only that section. Dopbot did not test for paid placement; the index records which brands the answers recommended, not why.
Should I just ask the assistant why it picked my rival?
You can, but treat the reply as more generated text, not as a record of what the system did. Use it to find pages to check, then verify those pages yourself.
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.