AI Brand Mention

People read an AI answer as a shortlist. Asked which bed frame fits a small room, the answer names a few brands and leaves out the rest. Being named is the first step into that list, and it happens before any reader decides whether to click.

The reports the engines publish count links, not names. Google’s AI features surface links to supporting pages, and the sites they link are counted in the Performance report of Search Console with the rest of Web search^[1]. Bing’s AI Performance report shows the total number of citations displayed as sources in AI-generated answers, with the pages cited^[2]. A name in the text of an answer, with no link to the brand, gives these reports nothing to count, so a team that reads only them cannot see its mentions.

Research measures visibility the same way. The paper that named generative engine optimization scores a source by the words of the sentences that cite it and by where the citation sits in the response^[3]. A mention without a citation has no such sentence to count, so mentions need a record of their own.

Access to the brand’s pages is a separate question. In QueryRev’s study of the top sites, 896 of the 6,294 sites whose robots.txt could be read, 14.2%, block at least one AI crawler by name^[4]. A rule like that decides which of a site’s pages a crawler may fetch, and so which of them an answer can bring in as a source. It does not decide what other pages say about the brand, so the two counts can move apart.

Per prompt, AI surface and run: the text of the answer, the brands it names and the links it gives. A mention is recorded when the brand is named in the text, in any form the brand uses, whether or not a link follows. An AI citation is recorded when one of the brand’s pages is linked as a source. The mention rate is the share of tracked answers that name the brand.

Give the rate a denominator and keep it. Bing’s Citation Share divides the citations attributed to a site by all citations shown for the same grounding query, and Bing calls it an observational metric, not a ranking or a scoreboard^[5]. A mention share is built the same way from names instead of links: the answers that name the brand, divided by all answers recorded for the same prompts, surfaces and runs.

Read the sentence around each name and keep it with the record. An answer can name a brand as the recommendation, as a cheaper alternative or as a namesake from another business. Only the first is the mention the brand wants, and checking a name against the entity it refers to keeps the others out of the count.

Sample data: an invented site, to show the method.

A furniture shop tracks four buying questions on four AI surfaces: sixteen answers. The brand is named in seven of them. In two of the seven, the answer also links one of the shop’s pages, so those are cited. In the other five it is named with no link, and in nine answers it is absent.

The five are the work. For each, the shop notes what the answer says about it and which page the answer links instead, such as a review site or a retailer’s listing. Where the claim is the shop’s own to make, its sizes or delivery terms, the shop’s page should state it plainly so it can be the source. One snapshot is not a result, so the shop reads these counts over several runs.