Learn · AEO, measured
Is AI visibility worth paying for yet, or is SEO enough?
Pay for AI visibility once your buyers ask AI before they shortlist; the seed-stage B2B companies we measured were in under 5% of the answers. In 50 baseline reports we ran in September 2026, Claude and GPT-6 Sol named the company in 81 of 1,679 answers to questions that left out its name, and 18 of the 50 were never named at all. SEO stays the floor either way, because engines with web search fetch from the same web your SEO work is for; it is just not enough on its own.
How often do AI engines name a young B2B company today?
AI engines rarely name young B2B companies: across 1,679 answers that left out the company’s name, 50 companies were named 81 times. In September 2026 we produced free baseline reports for 50 seed-stage B2B companies, all from Y Combinator’s 2026 batches: 20 buyer questions each, written for each company’s own category, answered once by Claude Opus 5 and once by OpenAI’s GPT-6 Sol, both with web search on (one report used Gemini in place of GPT-6 Sol; only its Claude answers are counted). Counting only the 849 questions that named neither the company nor its product, at least one engine named the company on 65 of them. We first published these figures in why AI engines can find your brand name but not your category, with a count we corrected on October 4, 2026.
| Measure | Count |
|---|---|
| Companies | 50 |
| Questions naming neither the company nor its product | 849 |
| Answers to those questions | 1,679 |
| Answers that named the company | 81 |
| Companies never named | 18 |
| Questions where at least one engine named the company | 65 |
| Citations in those answers | 12,728 |
| Citations to the company’s own website | 89 |
| Citations to the company’s own Y Combinator profile | 28 |
| Citations to the company’s homepage | 47 |
If we already do SEO, are we covered in AI answers?
SEO alone does not cover it: in 1,679 AI answers about 50 startups’ categories, only 89 of 12,728 citations went to the startup’s own site. Engines with web search build each answer from a handful of fetched pages, and for these companies those pages were mostly someone else’s: vendor documentation, comparison posts written by competitors, GitHub repositories, government and standards pages, and press. The companies’ own Y Combinator profiles were cited 28 times and their homepages 47 times. SEO still matters, because a page that cannot be crawled or indexed cannot be fetched at all (see AI crawlers don’t run JavaScript). It is necessary and not sufficient: ranking your own site does not put your name on the third-party pages an engine trusts for your category.
Does a well-indexed site show up for buyer questions?
Our own well-indexed site did not: Google reported one category-search impression in 28 days, and AI answers named us 0 times in 200. Google had 165 of the 185 pages in our sitemap in its index. From September 2 to 29, 2026, it showed claw-world.app 1,680 times; of the 111 pages that got impressions, 96 were posts about AI news, and of the queries Google names, all but one were about those posts’ topics or our own name. The exception asked about AEO tools. On September 28, Claude and GPT-6 Sol named us in 0 of 200 answers to 20 buyer questions about our category.
| Our own site | When | Result |
|---|---|---|
| Google impressions, all queries | Sep 2–29, 2026 | 1,680 |
| Google impressions on a category query | Sep 2–29, 2026 | 1 |
| Sitemap pages in Google’s index | Oct 3, 2026 | 165 of 185 |
| Claude and GPT-6 Sol answers that named us | Sep 28, 2026 | 0 of 200 |
Google names the query behind only some impressions and withholds low-volume ones, so the category count is a floor. What the comparison shows is narrow: on a site with no indexing problem, being indexed and getting impressions did not make us visible for buyer questions in either channel.
When is AI visibility worth paying for?
AI visibility is worth paying for when your buyers ask an AI engine before they shortlist and the answers name someone else. Three signals are quick to check: recent customers say they found or vetted you through ChatGPT, Claude, Perplexity or Gemini; your analytics show visits referred by those engines; and when you ask your category’s questions yourself, the answers name competitors, so there is a list to be on. If your pipeline comes from referrals and outbound and no deal mentions AI, the honest answer is probably not yet: spend on the pages and crawlability both channels need, and check again in a quarter.
How can I check before paying anyone?
Ask an engine your category question without your name, five times in fresh sessions, and write down every company it names and every URL it cites. Five times, because answers change from one run to the next. The cited URLs are your to-do list: they are the pages the engine trusts for your category, and getting onto them is most of the work. Then fetch your most important page the way a crawler does and check that the first sentence under each heading is in the raw HTML and short enough to quote. Our sample baseline report shows a fuller version of this check, with the company’s name taken out.
What can a provider honestly promise, and at what cost?
Nobody can honestly promise a ranking, a citation rate or traffic from AI answers; a provider can promise a baseline, the work and a re-measurement. Engines change their answers between runs and between model releases, and crawling, indexing and retrieval each add days to weeks, so honest results are measured against a baseline, with a frozen question set and control brands, rather than promised by a date; our method is on our methodology page. For reference, ClawWorld runs a free audit and pilot sprints from $500 with no monthly retainer, with the scope in writing before each sprint starts.
What can these numbers not tell you?
These numbers describe 50 young companies on two engines in one week of September 2026: a starting point, not your odds. The companies were chosen as outreach prospects, not at random, and all went through Y Combinator in 2026, so part of their absence may simply be that they are new. Whether an answer named the company was read by hand in each report, with same-name businesses left out. Each question was asked once per engine, not five times as our method asks, so one company’s result is noisy even where the total is not. Two engines run through developer tools are not every AI product your buyers use, and our Google figures are for one site, ours. None of this measures whether being named turns into pipeline; we have no data on that yet. Our own numbers, round by round, are on the ClawWorld Lab.
What decides whether an engine names you?
Being named by Claude mostly follows being cited for your own pages; GPT-6 names companies almost only on shortlist and price questions. The numbers behind that are in what makes ChatGPT or Claude name a company in an answer?