A new brand has two worries: that assistants only recommend names they have seen many times, and that a handful of reviews is too few to count. Our data cannot settle the second. It can say something about the first, because it records how many different brands each category's answers named and how many of them turned up only once. This post reads that long tail honestly, then lays out where a small brand can plausibly start and how to tell whether it is working.
Is there room in the answers for a brand nobody knows?
In the October 2026 edition, OpenAI's API with web search (gpt-6-luna) and Perplexity's Agent API with the Sonar model (perplexity/sonar) produced 1,587 answers across 160 categories. This section leaves out the "alternative to a named brand" question, because it named a brand in the prompt and lists about twice as many brands as the others. That leaves about 8 answers per category in that edition.
Excluding that question, a category's answers named a mean of 9.2 distinct brands (median 8). On average, 48.4% of those brands appeared in one answer only. That is the long tail: names that came up once among about 8 answers. At the same time, the most recommended brand in a category appeared in a mean of 62.3% of that category's answers, and in 80% of categories it appeared in at least half of them. Both things are true together.
How they combine differs by category. Of the 160 categories:
| Half or more of brands named in one answer only | Fewer than half | |
|---|---|---|
| Most recommended brand in at least half of the answers | 61 categories | 67 categories |
| Most recommended brand in fewer than half | 27 categories | 5 categories |
October 2026 edition, excluding the alternative-to-a-brand question; OpenAI and Perplexity answers together, 160 categories, each with about 8 answers. Counts are categories.
Read the table as a spread, not a verdict. In 88 of 160 categories (61 plus 27), or 55%, at least half the brands appeared in a single answer, and in the other 72 fewer than half did. In 11 categories one brand appeared in every answer, and in 5 categories the top brand was in fewer than half of the answers and most brands recurred. Two real examples from the October 2026 edition, each with its count: sunglasses had 8 answers that recommended 18 different brands, 13 of them in one answer only; gas grills had 8 answers that recommended 3 brands, none in one answer only, with one brand in all 8. These are single categories with a handful of answers, not findings about those markets.
Three limits apply. We did not record how old or how large any brand is, so "named once" does not mean "new". Brands named once may be established names that happened to come up once. A category with 8 answers moves 12.5 points with each answer. And this is the API, not the apps.
One more fact bears on strategy. Per category, a mean of 5.3 brands were recommended by only one of the two engines and 3.9 by both. A new brand may be recommended by one engine and not the other, so track the engines separately. The overlap post explains the measure.
What does a first appearance look like in numbers?
A brand recommended in 1 of 8 answers has a measured share of 12.5%, and the 95% Wilson interval around that runs from 2.2% to 47.1%. That is not a stable position; it is an appearance. The table shows what small counts can and cannot support, at the 8 answers per category of our October edition and at the 40 answers per category that the 40 monthly updated US categories use from November 2026.
| Recommended in | Share | 95% Wilson interval |
|---|---|---|
| 1 of 8 | 12.5% | 2.2% to 47.1% |
| 2 of 8 | 25.0% | 7.1% to 59.1% |
| 0 of 40 | 0.0% | 0.0% to 8.8% |
| 1 of 40 | 2.5% | 0.4% to 12.9% |
| 4 of 40 | 10.0% | 4.0% to 23.1% |
Illustrative counts computed with the Wilson score interval; not Dopbot measurements.
Use these as milestones. At 40 answers, zero recommendations cannot rule out a true share as high as 8.8%, so "not visible yet" needs a large sample. One recommendation at 40 answers already gives an interval above zero (0.4% to 12.9%), but that interval is too wide to read as a position. It narrows only with more recommendations. A change from 1 to 2 recommendations at either size says nothing. The share of answer post has the method, and the experiments post covers how to test a change.
Where can a new brand start?
None of the steps below is proven to produce a recommendation. They are in the order we would start, not ranked by proof.
1. Be readable by the search crawlers. OpenAI says sites opted out of OAI-SearchBot will not be shown in ChatGPT search answers, though they can still appear as navigational links. Perplexity recommends allowing PerplexityBot in robots.txt to ensure a site appears in its search results. Google says a page must be indexed and snippet-eligible to be a supporting link in its AI features. A new brand that blocks these bots by accident, or whose pages are not indexed, is ruled out of those products' search answers or supporting links as of October 2026, so check this first.
2. Make your own site the clearest source. In 46.6% of OpenAI's answers (800), a recommended brand's own domain was among the linked pages; in 62.0% of Perplexity's answers (787), it was among the retrieved sources. Both figures cover all five question types of the October edition, including the alternative-to-a-brand question, and count a domain whose first label equals a recommended brand's name, so they miss country domains, parent companies and other store names. They say a brand's own domain is a regular neighbor of recommendations, not that it caused them. The brand website post goes into the rule and what to keep current.
3. Write for narrow questions (our judgment, not a measured result). Answers to the four question types other than the alternative question recommended between 2.2 and 2.6 brands on average: 2.2 brands for use-case answers (318 answers) and 2.4 brands for best-overall answers (318 answers; both engines and all three markets). A narrow question does not offer more slots. Across questions, brand lists differ: in the 155 categories with two budget and two premium answers, the brands in the budget and premium answers overlapped by a mean of 8.2%, and best-overall against budget by 23.8% and against premium by 26.8%. Overlap here is computed per category as the brands recommended in both answer sets divided by the brands recommended in either (Jaccard), then averaged over categories with equal weight; it is not answer-by-answer agreement. Those are observed overlaps of short lists, from few answers per category. Our hypothesis, which the data do not test, is that a brand missing from a broad question may still appear in a narrower one. The buying guides post shows how to write for those.
4. Make sure retailers list your products well. A retailer domain on a hand-made list of 49 retailer domains was among the pages linked in 17.6% of OpenAI's 800 answers and among the retrieved sources in 76.7% of Perplexity's 787 answers. That does not show that a retail listing, or a Vine review, gets a brand recommended. If you sell on Amazon with few reviews, Amazon's Vine program is aimed at that situation. As of October 2026, Amazon says existing products can enroll with fewer than 30 reviews on the detail page, and the seller needs a professional selling account and a brand representative or reseller role for a brand in its Brand Registry, among other requirements. Amazon's Vine page also says enrollment carries a cost per parent product and requires eligible Fulfillment by Amazon offers, so check the current terms before you plan around it. Amazon's Vine page says it connects your products to a network of reviewers that Amazon has invited as Vine Voices. See the retail listings post for the rest.
5. Collect real reviews, by the rules. The FTC's guidance says the rule exempts generalized solicitations to purchasers to post reviews about their experiences, and that the rule prohibits compensation or other incentives conditioned on a review expressing a particular sentiment, while it does not prohibit giving incentives for reviews as such. Ask every buyer, offer nothing in exchange for a positive review, read the FTC guidance for what is allowed, and see the reviews post for the full list.
6. Get covered by review sites. Review sites differ in how they source what they test; the review sites post covers what each says and how to approach them. Whatever the route, the first job is retail availability and verifiable specifications.
What would a sensible sequence look like?
- Month one: check crawl access and indexing, fix product facts on your own site, and take a baseline of 40 answers per prompt per engine on three or four narrow prompts (your own sampling: 40 repeats of one prompt on one engine, which is not the same as the index's 40 answers per category across five prompts). Our judgment, not a measured result: start with narrow questions, not the broad "best" one.
- Month two: get listed where the buyers are, start the review request flow, and publish guides for the situations your product fits.
- Month three: contact the sites that suit your category as the review sites post describes, then resample with the same prompts and the same number of answers, and compare counts and intervals, treating overlapping intervals as no evidence of change.
If a new brand does start to be named, our judgment is that it would first show as occasional mentions in one engine and one narrow question, but we have no data on this. Treat a first mention as a reason to keep sampling.
What do we not know?
- Whether review count or brand age matters. We did not classify brands by age or size, and we have no experiment on either.
- Whether any single step works. Each step above rests on a documented platform behavior or on a pattern in a sample, not on a measured effect.
- How the apps behave. They can differ from the APIs in personalization and product features.
Frequently asked questions
How many reviews does a brand need before assistants recommend it?
We found no published number. Amazon's Vine page uses "fewer than 30 reviews" as the line for enrolling an existing product, but that is the rule for that program, not a visibility threshold in any assistant.
Is being named in one answer out of eight a success?
It is a start and not a result. At 8 answers, 1 mention is consistent with a true share anywhere from 2.2% to 47.1%. Treat it as a reason to keep sampling, and compare it with your own later samples, not with a competitor's single count.
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.