How often an AI app names a startup tracks how many different websites it cites a page about that startup from, more closely than it tracks how much money the startup raised. In Presspool's record of 11,131 answers from seven AI apps to questions about AI startup markets, 172 sourced AI startups were sorted by the number of different websites that had a page naming them among the sources cited in answers to one random half of the questions. Their naming was then counted in the answers to the other half. The 68 startups with no such website were named in 1.3 percent of those answers; the 50 with one or two, 8.2 percent; the 30 with three to nine, 20.7 percent; the 24 with ten or more, 36.4 percent.
The rank correlation between the number of websites and the naming rate was 0.73 (0.65 to 0.80). Between funding size and the naming rate it was 0.49, among the 132 startups with a stated amount, where the website count scored 0.74. The website count tracked naming inside every funding band and held when the two halves were swapped (0.66).
What was counted #
- Answers. 11,131 answers that returned a result, from ChatGPT, Claude, Perplexity, Gemini, Copilot, Google AI Mode and Google AI Overviews, collected by Presspool, a Bramwell company, between 2026-09-27 and 2026-11-19 to 1,145 questions about AI coding, AI infrastructure, legal AI, physical AI, voice AI, world models, AI security and humanoid robots, plus a funding-roster question set. The questions are Presspool's own pinned and discovery questions, not measured buyer demand.
- Companies. The 227 companies on Presspool's niche maps, each sourced to its own site and an independent source, less 55 whose names are ordinary English words (a page title containing the word cannot be told from a page about the company). 172 remain; each had at least 91 answers in its niche in the held-out half.
- Two halves. Questions were split into two halves by a fixed hash of the question text, so no question is in both. In the first half, a company's coverage is the number of different websites, other than its own and other than directories, profile and social sites, that had a cited page whose title or web address names it. In the second half, its naming rate is the share of answers in its niche that name it, in the text or by citing its own site.
- Result. The mean naming rate of the companies in each group, with intervals from resampling companies, 2,000 draws, fixed seed.
The result #
| Websites with a cited page naming the company | Companies | Mean share of held-out answers naming it | Interval | Named in at least one answer |
|---|---|---|---|---|
| None | 68 | 1.3% | 0.9 to 1.8 | 52 |
| 1 or 2 | 50 | 8.2% | 4.9 to 12.4 | 44 |
| 3 to 9 | 30 | 20.7% | 14.1 to 28.3 | 29 |
| 10 or more | 24 | 36.4% | 26.4 to 47.6 | 24 |
The median company in the top group was named in 27.9 percent of its niche's held-out answers; the median company with none, in 0.5 percent.
It is not only funding size #
Bigger rounds go with more coverage, so the same table is split by the amount of the company's latest round. Cells with fewer than five companies are left blank.
| Latest round | None | 1 or 2 | 3 to 9 | 10 or more | Rank correlation, websites against naming |
|---|---|---|---|---|---|
| Under $20 million | 0.7% (18) | 1.8% (6) | 0.67 (28 companies) | ||
| $20 to $100 million | 1.8% (20) | 17.1% (8) | 20.0% (8) | 0.61 (40) | |
| $100 million or more | 1.8% (11) | 5.6% (23) | 26.0% (14) | 35.0% (16) | 0.73 (64) |
| No amount stated | 1.1% (19) | 10.3% (13) | 15.0% (5) | 0.70 (40) |
Companies in brackets. Within each funding band the more widely covered companies were named more.
Does it matter which website it is? #
Not in a way this sample can separate. 79 of the 963 company-and-website pairs, 8.2 percent, were a tier-one business or tech outlet (TechCrunch, Reuters, Business Insider, Bloomberg, Forbes and similar). At the same breadth, companies with at least one such outlet were named at 4.0 against 9.4 percent (1 or 2 websites; 11 and 39 companies), 21.1 against 20.4 (3 to 9; 16 and 14) and 35.8 against 42.8 (10 or more; 22 and 2). Swap the halves and the same cells read 7.8 against 8.4, 21.1 against 22.5 and 33.3 against 11.1, with as few as 3 companies on one side of a cell. The direction differs between cells and between the two splits, so no tier-one advantage at equal breadth shows up here.
The pages are mostly not news. Of 1,298 distinct company pages, 320 (24.7 percent) had a best-of, comparison, review or roundup title by a simple rule, and 34 (2.6 percent) were wire releases. OpenAI's own site was the website with the most company pages (89 of the 963 pairs). Leaving out openai.com, google.com and anthropic.com pages lowers the rank correlation to 0.68 (0.59 to 0.76).
It holds when the halves are swapped #
Counting coverage in the second half and naming in the first gives 2.0, 8.4, 21.9 and 27.8 percent across the same four groups, and a rank correlation of 0.66 (0.57 to 0.74). Within the funding bands the correlation was 0.36, 0.60 and 0.73 for the three bands with a stated amount, the weakest in the smallest band.
How this fits the other studies #
Cited sources are rarely named, and funded AI companies are mostly absent from the Index's buying-question answers. This study adds a within-niche split: among sourced AI startups, how many separate websites engines cite a page about is a better marker of which ones they name than the size of the round. The page-feature studies look at the cited page; this one looks at how many different pages about one company are in the retrieved sources.
What this changes #
- Count websites, not logos. For a company that wants an AI app to name it, the number of different sites with a page about it in what the engines retrieve is the figure that tracked naming here; one tier-one story did not separate companies at equal breadth.
- The first few websites separate the most. Startups with one or two were named six times as often as those with none (8.2 against 1.3 percent); those with three to nine, 20.7 percent.
Limits #
- It is an association among 172 companies, with some groups of 3 to 24, and it is not an experiment. Prominence, age, category and a company's own publishing can raise both the coverage and the naming. The two halves share niches and similar questions, so they are not independent samples.
- Coverage is read from page titles and web addresses. It misses pages that name a company only in the body, so some companies counted as none are covered, and a roundup that lists twenty companies counts for each of them the same as a feature story.
- The 55 companies dropped for ambiguous names are not measured. The companies are those Presspool maps, not every AI startup.
- Naming includes citing a company's own site. A company's rate is over answers in its niche, which differ in number by niche.
- The window is 2026-09-27 to 2026-11-19. The funding amount is the latest round's, which may postdate the first answers.
Method note #
Answers: Presspool's daily observations (one record per answer: app, question, time, answer text, cited sources and the companies named), 2026-09-27 to 2026-11-19, results that returned without error, 11,131 of 12,144. The text of answers is not published; this page carries derived counts only. Companies: presspool.ai/companies.json on 2026-11-19, entries with at least one niche, less those whose name is a single ordinary English word (55). Names and aliases of four characters or more were matched as whole words, case-insensitively, in a cited page's title and in its web address without the query string, so a tracking tag such as utm_source=openai cannot match; the company's own domains (its domain, its description-page host and any earlier domains) and the directory, profile and social domains are excluded (Crunchbase, LinkedIn, Wikipedia, PitchBook, G2, Facebook, X, Twitter, YouTube, Reddit, Instagram, Tracxn, ZoomInfo, Sacra, CB Insights, GitHub, Medium, DEV). Split: the question text hashed with MD5, parity decides the half. Rank correlation: Spearman with average ranks, interval by resampling companies, 1,000 draws, seed 20261119. Tier-one outlets: techcrunch, reuters, bloomberg, fortune, cnbc, forbes, businessinsider, wsj, axios, venturebeat, ft, theinformation, wired, nytimes, theverge, fastcompany, inc, washingtonpost, economist and siliconangle, on .com. Roundup-style title: a title with best, top, vs, versus, compar, review, alternatives, leading, landscape, guide, list of, companies or startups as a word. The script and the group-level output are retained in the Machine Relations Research repository with this analysis; the answer file itself is private.
Frequently asked questions #
How many websites does a startup need for AI to name it? #
There is no threshold in this data, but the naming rate rose with every step. Across 172 AI startups, those with no outside website naming them in cited sources were named in 1.3 percent of held-out answers, those with one or two in 8.2, with three to nine in 20.7 and with ten or more in 36.4.
Does funding size explain it? #
Only partly. Funding size alone scored a rank correlation of 0.49 with the naming rate among the 132 companies with a stated amount, against 0.74 for the website count, and the website count tracked naming inside every funding band (0.61 to 0.73 in the three largest).
Does a TechCrunch or Reuters story count for more? #
This sample cannot tell. At the same number of websites, companies with a tier-one outlet were named more in some groups and less in others, and the direction changed when the halves were swapped.
Does more coverage cause AI to name a company? #
This study does not show that. Companies engines already know attract pages that name them, so the coverage can follow the naming. It counts association among 172 companies and is not a test of earning coverage.
What kind of pages name these startups? #
A quarter had a roundup, comparison or review title, 2.6 percent were wire releases and 8.2 percent of company and website pairs were tier-one outlets. The rest were not classified.