An AI answer to a business owner cites 8.7 sources - and most of the seats are held by platforms and forums, not vendors.

Key findings

• AI answers to business owners' questions cited a mean of 8.7 sources each, with a median of 9 and a maximum of 27.
• Across 80 owner-voiced questions, the most frequently cited domains were YouTube (47 answers), Reddit (34) and Salesforce (32).
• Of 117 questions where the filter phrase genuinely appears in the question, 80 were asked in an owner's voice and none in a marketer's.
• Of those 80 owner questions, none named a trade, one mentioned AI at all, and 29 explicitly asked for something free or cheapest.

How this was measured

When

Measured 31 August 2026 against a corpus of 1,935 captured answers covering 1,819 distinct questions.

Instrument

A commercial corpus of AI assistant questions and answers, with the cited source list preserved per answer.

Surface

Both surfaces present in the corpus. 88.0% of these answers were produced by a web search rather than from the model's memory.

Sample

The 117 records where the filter phrase genuinely appears in the question, of which 80 are voiced by a business owner.

Pre-registration

None - this is descriptive. The correction below was made before any number here was believed.

A correction that has to come first

The corpus filter matches the whole record, and a record is mostly answer text. Filtering on a phrase therefore returns answers that contain it, not questions that contain it - and in this data the phrase was genuinely in the question only about 6% of the time.

That is a large error and it would have contaminated everything below. So every figure on this page is computed on the clean subsample: the 117 records where the phrase really does appear in the question.

We are publishing the correction alongside the result because the uncorrected version of this analysis produced a confidently wrong answer, and anyone reading a similar number elsewhere should ask which one they are looking at.

Who holds the seats

Counted across the 80 owner-voiced questions. Video and forum content, not vendor service pages, hold the top two positions.

9 seats, and who they go to

The shape of the opportunity is in the arithmetic. An answer cites about 9 sources. 2 of the most reliable seats go to a video platform and a forum. Several more go to the operator of the platform being discussed - Google's own business documentation, Nextdoor's, Facebook's.

What is left is a handful of seats occupied by companies that published something substantial and factual: a large software vendor's business guides, an insurer's small-business library, a bank's. None of them won a seat with a service page. They won it with a reference library.

That is the same pattern our other work found from a different direction: the page cited is the page that owns the specific fact the answer needs.

What those owners are actually asking

The 80 owner questions are worth reading carefully, because they are not what a vendor in this category would expect.

None of them names a trade. Not 1 owner wrote “my roofing company” or “my massage studio” in a way that made the trade part of the question. Exactly 1 of the 80 mentioned AI or an assistant at all. And 29 of them - more than a third - explicitly asked for something free, cheapest, or achievable with no money.

Their real questions are plain: how to advertise a business, the best way to advertise, how to grow. That is the vocabulary, and it is a long way from the vocabulary this industry uses about itself.

The subsample also settles a related question: of these 117, 80 were voiced by an owner and none by a marketing practitioner. The plain-language business owner in this data is real. He is simply not shopping for this category.

What this means if you are buying

• There are roughly 9 seats in an answer and most are spoken for. Getting one is not a matter of being slightly better than a competitor's service page - it is a matter of owning a fact none of the current occupants owns.
• The companies holding vendor seats did it with reference libraries, not sales pages. That is the template worth copying, and it is what the pages in this section are.
• If you sell to local business owners, the language finding matters more than the citation finding: they do not use this category's words, do not attach their trade to their problem, and a large minority are looking for something free. Search is not how you reach them.

Honest limits of this measurement

• n=117 questions, of which 80 are owner-voiced. Directionally strong, not precise.
• The corpus is a sample of captured questions, not the whole answer space. A domain showing zero records here has not been shown to be absent from AI answers - we have seen companies named in our own runs that have no records at all in this corpus.
• The choice of filter phrases shaped everything. 14 phrases are not the whole of business language.
• Voice classification is automatic, based on phrasing patterns. It is good enough for a split this lopsided and would not be good enough for a close one.

Questions about this research

More measurements from the same instrument

• People ask about AI visibility on Google, not inside ChatGPT
• We asked an AI assistant the same question twice, 10 minutes apart
• All seven measurements →

We run this measurement for a living. If you want it run on your own category and your own city, the free scan is the same instrument on one question, and the report is the full set. Nothing here is behind a form.

Every category we measure has its own page, with the questions customers put to an assistant in that category and what a reading costs. See all industries we measure →

Published 31 August 2026. Every figure on this page is recomputed from raw observation files by a single script, so any of them can be traced back to the readings behind it. If you are named here and believe a reading is wrong, tell us: we will run the measurement again and publish what it returns, whichever way it goes.