Do AI-cited pages actually mention the brand?

A reviewed citation is one AI answer, one brand named in it, and one page cited for that brand. The no-brand-mention rate is the share of reviewed citations where the cited page contained no matching mention of the attributed brand.

The observed no-brand-mention rate was 38.5%. A citation attached to a brand is not proof that the cited page names that brand.

Semrush defines a ghost citation as a cited domain whose own brand is not named in the AI answer. Its study reported that outcome for 62% of 3,981 times a domain was cited.

This report tests the other side of the link. It asks whether the cited page contains the separate brand attached to that citation. The cited page contained the attributed brand in 61.5% of reviewed citations and did not contain it in 38.5%.

Google AI Mode had a 45.0% no-brand-mention rate across 142,848 reviewed citations. ChatGPT Search had a 24.7% rate across 66,268 reviewed citations.

This is a descriptive engine split. Each engine selected a different set of pages, so the result includes both source selection and page content.

Source type is the editorial category assigned to a cited domain. Among the eight largest source types, reference pages had the highest no-brand-mention rate at 49.7%. Social pages followed at 45.6%.

General web pages had the lowest rate in this group at 34.9%, and directories were close at 36.4%. No large source type eliminates the page-to-brand gap.

Among the 20 domains with the most reviewed citations, wrike.com had the highest no-brand-mention rate at 65.6%. medium.com followed at 51.7%.

Sheltera.ai was at 4.0%, dynatrace.com was at 4.3%, and soar.sh was at 6.8%. The wide range makes a domain-level review more useful than the overall average.

The eight brands with the most reviewed citations all had a material no-brand-mention rate. Dynatrace was lowest at 13.9%. AiDash was highest at 32.5%.

The difference is not a brand quality score. It describes the pages attached to each brand in this window.

Among the 10 largest primary brand-industry samples, Professional Networking had the lowest no-brand-mention rate at 22.7%. Semantic Search had the highest rate at 40.3%.

These are brand-industry labels, not prompt categories. Compare a brand with others in its industry before treating its rate as unusual.

A page-brand pair is one cited page matched to one brand under its standard name. The sample contained 25,491 pairs checked more than once.

Only 164 repeated pairs produced both outcomes. The mixed-outcome rate was 0.64%. The result is usually a stable property of the page-brand pair, not something that changes from answer to answer.

The set of reviewed citations expanded during the seven-week window, so raw weekly rates do not form a like-for-like trend. We tested the same 8,265 page-brand pairs in two three-week windows instead.

The no-brand-mention rate was 21.8% from June 1 to June 21 and 21.3% from June 22 to July 12. The same set of pairs moved by 0.55 percentage points.

We excluded 66,860 citations whose page we had not fetched yet, 43,896 where the page fetch failed, and 2,414 citations with no page URL.

We reviewed 68.3% of ChatGPT Search citations and 63.4% of Google AI Mode citations. The headline describes reviewed citations only. It should not be extended to pages that could not be evaluated.

A reviewed citation passes when the fetched page contains matching context for the brand under its standard name. It fails when the page contains no matching brand context.

This does not test whether every sentence in the AI answer is factually supported. A page can name a brand and still disagree with the answer. A page can omit the brand and still support a category-level statement.

First, open the cited page and look for the brand and relevant product aliases. Second, locate the sentence in the AI answer that the citation appears to support. Third, compare the specific price, capability or comparison claim with the page text. Passing the first check does not guarantee the other two.

For example, a page can mention a vendor in a navigation menu without supporting an answer that recommends its enterprise plan. Conversely, a general explanation of a technology may support background context without mentioning any vendor. The 38.5% result measures absent brand mentions in checked pages; it is not a finding that 38.5% of answers are fabricated.