AI mention

A mention is the brand's name appearing in the answer text a user reads. A citation is a source link the engine attaches to part of its answer, crediting the page the information came from. The two are independent: a brand can be mentioned in an answer whose citations all point to third-party review sites, and a brand's blog post can be cited as a source for an answer that never names the brand at all. Conflating them produces confused strategy, because they are earned differently and they pay differently.

Mentions are what buyers actually see, which makes them the metric closest to revenue. When an assistant answers a question about the best tool in a category, the names in that paragraph are the shortlist; the citation links, where they exist at all, are footnotes most users never open. Citations matter as the mechanism, since engines compose answers from the sources they retrieve and trust, but the mention is the outcome the mechanism produces. Programs optimize citations on the way to earning mentions.

The distinction also splits the work. Earning citations is largely a content and authority problem: publishing pages engines retrieve and lean on. Earning mentions additionally requires that the retrieved sources talk about the brand by name, which is why third-party presence matters so much. 5W Research found Wikipedia and Reddit together account for more than a quarter of ChatGPT citations, and a brand absent from the pages an engine cites will struggle to be named in the answers those pages feed.

The counting rule that keeps mention data trustworthy is literal matching: a mention counts only when the brand name actually appears in the stored answer text. The alternative, letting a parser or a judging model credit implied or fuzzy references, quietly injects interpretation into the denominator and produces scores nobody can audit. Under literal matching every counted mention links back to specific words in a specific stored answer, and a skeptical stakeholder can check any data point by reading it. Reachroller scores this way by design, counting a mention only when the name appears verbatim in the response returned by the official API.

The second requirement is repetition, because a mention is a probabilistic event rather than a fact about the engine. SparkToro measured under a 1 percent chance that two identical ChatGPT runs return the same list of recommended brands. One answer that names you proves you can be mentioned; it says almost nothing about how often. Mention rate over repeated runs of a fixed question set is the meaningful statistic, and its trend over weeks is the signal a program manages against.

Context is the layer on top of the count. A mention can be a recommendation, a neutral listing, or a warning, and it can lead the answer or trail it in an also-ran clause. Mature tracking stores the full answer so position and framing can be reviewed, and it separates unbranded questions, where a mention means discovery, from branded questions, where mentions are guaranteed and the interesting signal is what the engine says about the brand instead.

The commercial weight of a mention comes from where buying research now happens. G2 found in 2026 that 51 percent of software buyers start research with an AI chatbot more often than with Google, and Forrester's survey of 18,000 buyers found 94 percent used AI during their most recent purchase. The names an assistant produces in response to a category question function as the initial shortlist, formed before any website is visited and before any ad is seen.

Two G2 numbers show the mechanism directly. 69 percent of B2B software buyers chose a different vendor than they originally expected because of what an AI chatbot told them, and 33 percent bought from a brand they had never heard of before the AI named it. That second number is a pure mention effect: the brand entered the deal because its name appeared in generated text. The inverse holds with equal force, since a brand the assistant never names cannot be the unfamiliar option a third of buyers end up choosing.

This is also why mention tracking belongs upstream of attribution debates. Analytics tools see the visit that eventually arrives, if one arrives at all, and label it direct or organic long after the recommendation happened. The mention is the earliest observable trace of the influence, which makes mention rate over time the practical leading indicator for a channel whose clicks are sparse and whose influence is real. Teams that wait for referral traffic to prove the channel exists are measuring the shadow of an event their competitors are already optimizing.