What evidence explains an AI answer's choice of source?
Broadcastwell measures and fixes whether AI search engines name B2B software companies.
Run the free checkCapterra.com was a cited host in 34 of 150 construction-software answers in our September Index, collected on 22 September 2026 across five engines. That supports a host-level citation observation; explaining a particular page’s selection requires its returned URL and answer, and still does not reveal the engine’s internal reason.
For a particular page-level example, our 22 September 2026 Google AI Overviews record about replacing CRM spreadsheets contains the fragment “HubSpot CRM or Pipedrive to transition away from manual spreadsheets” and records https://www.hubspot.com/products/crm among its citation destinations. The complete URL identifies a page that the host-level total cannot. The record does not establish which passage was consumed, whether the page supported every surrounding claim or why it was selected.
What we measured
Measured. Under the Index citation method, the construction category pools ten buyer questions, three repeats and five engines. A host is counted once per answer in which it appears. The thirty-four host appearances do not identify thirty-four distinct Capterra pages, nor establish that every engine selected the same URL.
Reported. Google's Search Central guidance, read on 29 October 2026, says a supporting page must be indexed and eligible for a Search snippet. It also says that satisfying requirements does not guarantee indexing or serving. Eligibility and an observed selection are therefore different claims, even within Google's own named features.
Reasoned. The returned citation is direct evidence that an answer linked a source. The source page can then be read to check what it actually supports. Neither the presence of the link nor a generally eligible page establishes which sentence, formatting choice or reputation signal caused the selection.
What this means for you
Read the host names. The construction export keeps capterra.com, g2.com and getapp.com separate, with 34, 21 and six appearances respectively in the same 150-answer base. A Capterra host total cannot be converted into the URL of a particular review page. It also cannot absorb the G2 or GetApp observations simply because all three publish software information.
Procore's own record illustrates a second distinction in that export. It has 109 named appearances but 30 own-domain citations across the pooled 150 construction answers, and its cited-domain list names procore.com and learn.procore.com. The domains identify a vendor source family; the differing name and citation counts show why a brand-mention record cannot stand in for a returned page link. An assertion about one linked document still needs that document's URL, not an invented page under either host.
Also asked as
What actually gets a page cited in AI answers? Trying to separate what Google confirms from what people are seeing
Related
- Construction citation evidence
- Google supporting-page requirements
- Construction source-host counts
- The Manual’s evidence boundary
Next step
Capture a separate public answer to examine its named sources with the Free 10-question check (one engine). No email is needed; runs are limited per network and per day, and by the site's daily allowance.
For one sourced factual error, Fix one answer, $490. Retain the answer and the document that supports the correction.
Every figure comes from the published method, v1.1.
Prepared by the Broadcastwell team.
Page owner: Rohan Iyer, AI Engineer. Corrections: hello@broadcastwell.com
Related evidence
- How can you correct a documented gap in an AI answer?
- How can I check whether an AI citation supports its claim?
- All fourteen category records
- Absence Index method and data definitions
Published . Last updated .