GEO Tracker / Docs
Dashboard

Mention vs. citation — what counts as which

Two different signals AI engines emit, and why we treat them separately.

4 min read

When AI answers a question, two distinct things can happen:

  1. The model can mention your brand in the prose ("Vercel offers instant rollback…").
  2. The model can cite a source URL that links to your domain, even if your brand name never appears in the visible text.

We track both. They're not the same signal — being mentioned in the sentence the user reads is a stronger position than being a footnote URL the user might click. Below is the precise distinction we use.

Mention

A mention is your brand name or your tracked domain hostname appearing inside the AI's answer text. We detect it in two passes:

  1. Layer 1 — deterministic. A regex pass over the answer looking for your hostname (netlify.com, with or without www. and HTTPS) and your brand stem (netlify).
  2. Layer 2 — LLM classification. For each mention found, a fast classifier assigns a quality bucket: mentioned (40), recommended (70), or top_recommended (90). Long-tail or ambiguous cases route to a small refine call; obvious cases stay deterministic.

A mention sets is_mentioned: true on that scan_result. The quality bucket feeds the GEO Score formula.

Citation

A citation is a URL pointing to a domain that appears in the engine's structured citation array (Perplexity references, Google AI Mode references) or extracted by regex from the prose (the fallback list). Each scan_result stores two citation fields:

  • cited_domains[] — engine-native references (preferred, higher signal). Example: Perplexity returns [{"url": "https://example.com/x", ...}, …] and we collect the hostnames.
  • competitor_mentions[] — regex fallback hits over the prose for domains not surfaced in cited_domains but mentioned inline.

Citations don't directly affect the GEO Score for the cited brand — they affect Competitor tracking. For you, the relevant signal is whether you were mentioned (above); for competitors, the relevant signal is whether they were cited (here).

Why split the two

Different products map to different patterns:

PatternWhat you typically see
Brand-strong product (Vercel, Stripe)High mentions, high citations
Cited-but-anonymous tool (a dev library page)Low mentions, high citations
Generic page in a roundup (an "alternatives" listicle)Zero mentions of your brand, citation goes to listicle, not you

Treating them as one number would hide which kind of visibility you have — and the actions you take to improve are different. Mentions improve when AI knows about you (PR, bylines, brand pages). Citations improve when AI's preferred sources link to you (good docs, schema-marked content, guest posts on cited domains).

A practical example

Imagine your tracked domain is netlify.com — a real deployment platform we used as a stand-in for the running test scenario. Here's a snippet of an actual Google AI Mode answer for the buyer question "best deployment platform":

Vercel is the standard for Next.js and modern frontends, with Git-based preview deployments and instant rollback via immutable deployments.[1] Other strong options include Netlify[2] and Render[3].

[1] https://vercel.com/kb/guide/vercel-vs-netlify [2] https://www.netlify.com/ [3] https://render.com/docs/render-vs-vercel-comparison

For Netlify (the example tracked brand):

  • Mention. "Netlify" appears once in prose → is_mentioned: true, quality mentioned (0.40) — listed in passing as one option among several, not the top recommendation.
  • Citation. cited_domains contains netlify.com → Citation search finds this scan_result for the term netlify, and Competitor tracking attributes a mention event to Vercel and Render (the other deployment platforms named in the answer).

Same answer, two parallel signals.

Read further