In an AI answer your brand can appear in two very different places: named inside the answer itself (a mention) or linked underneath it as one of the sources the model drew on (a citation). The distinction sounds pedantic — it decides your entire strategy, because mentions win the customer while citations only win the visit, and most answers now end without any click at all.

When someone asks ChatGPT or Perplexity a buying question — "best brand monitoring platform for a mid-size team?" — the answer might read:

For mid-size teams, geopsy stands out for tracking brand visibility across AI answer engines, while established suites cover social listening at higher price points. Reviewers consistently highlight its cross-engine coverage and sentiment tracking.

Sources: g2.com · comparison-blog.io · reddit.com/r/martech

One answer, two currencies. The mention shapes the decision; the citations credit the evidence. And note who got cited: not the brand — the third parties who wrote about it.

They are independent axes

The intuition most teams carry over from SEO is that these two things travel together — that being a source and being recommended are the same achievement. They aren't. The model synthesizes its recommendation from everything it has read about you, then credits whichever pages happened to supply the sentences it wrote.

That produces four distinct positions a brand can occupy:

Mentioned + cited — you are the answer and the evidence. Mindshare plus traffic; the dominant position, rare and worth defending.
Mentioned, not cited — the model recommends you based on what the web says about you: reviews, comparisons, forums. No link, no referral traffic. You still win the customer.
Cited, not mentioned — your benchmark post sources a generic claim. Your link sits in the footnote; your brand is absent from the recommendation. You're a library, not a choice.
Neither — invisible. Your competitors are being described to your buyers in a conversation you can't see — unless you're measuring it.

Traditional GEO tooling measures whether you got linked. Buyers are won in the first two positions.

Why mentions are the senior metric

Citations drive traffic, and traffic is measurable, so citations get the attention. But look at where user behavior has actually moved. Roughly 60% of searches now end without any click at all (Bain & Company, 2025). Pew Research found that even when source links are displayed right below an AI Overview, only about 1% of views produce a click on them. The answer is consumed on the spot.

Meanwhile, the buying decision forms earlier than the click. Informational and consideration queries — "best X for Y", "is X better than Z" — are exactly where AI answers trigger most aggressively. The transactional click still happens later, but it's no longer contested: it goes to whichever brand won the mention upstream, typed directly or searched by name.

That single mechanism explains the stat everyone quotes without explaining: AI-referred visitors convert 4.4× better than traditional organic visitors (Semrush, 2025). The click isn't the moment of persuasion anymore. It's the receipt.

So the honest hierarchy is this: citations are a traffic metric. Mentions are a share-of-decision metric. One measures visits; the other measures how often the machine that sits between you and your market recommends you.

The playbook flips

Here's why the distinction changes what you actually do on Monday morning.

Citations reward your own content. Crawlable pages, clean structure, quotable claims, schema. It's SEO thinking in new packaging — publish better assets and earn the link. Necessary, familiar, and fully within your control. (Our guide to getting cited by Perplexity covers this half.)

Mentions reward the web's consensus about you. Models don't recommend brands because of what the brands say about themselves. They synthesize reputation from review platforms, comparison articles, community threads, expert roundups — the entire third-party record (consensus signal). In the example above, notice who got cited when the model recommended a brand: G2, a comparison blog, a subreddit. Not the brand's own site.

The consequence is uncomfortable for content-led teams: you cannot publish your way into a mention. You have to earn one across surfaces you don't own — and you can't manage what you can't see. That makes mention share an intelligence problem, not a content problem. The questions that matter are empirical:

Mention share — how often are you named versus competitors, across the prompts that map to your revenue? This is share of voice, counted on the answer text.
Context and sentiment — are you recommended, merely listed, or warned about? (Sentiment analysis per answer, not per quarter.)
Source graph — which third-party pages do models actually draw on when they talk about you? Those pages, not your blog, are your new ranking factors.
Engine coverage — ChatGPT, Gemini, Claude, and Perplexity each read the web differently. A mention on one is not a mention on all.

This split is baked into how geopsy measures: a brand named in the answer body earns a text rank, a brand linked as a source earns a source rank, and the two are tracked separately — because a footnote link is not a recommendation, and treating it as one inflates your numbers exactly where it matters most.

Rankings were a proxy. Mentions are the thing itself.

For twenty years, position #1 was a proxy for being chosen — the best available approximation of preference. AI answers removed the proxy. The model now states a preference directly, in prose, to a buyer who will probably never scroll past it.

Citations still matter; they carry the highest-intent traffic left in search, and being cited lifts everything below the answer. Chase them. But rank them honestly: they're the second metric. The first is whether, when the machine describes your market, your name is in the sentence.

For the market data behind this shift, see SEO to GEO: the market shift in numbers; for the wider strategy, how to rank in ChatGPT and AI search. Or see your mention share today — across every major answer engine, run over run.