Brand Mentions vs. Citations in AI Answers: What to Measure, Ghost Citations, and Four Metrics

If you read an AI visibility dashboard without separating brand mentions from citations, you will read the numbers backward. A common pattern: your domain shows up again and again as a source link in ChatGPT answers, yet the answer text only names your competitors. Semrush and Growth Memo (Kevin Indig) analyzed 3,981 domain appearances in AI answers and found that 74.9% carried a citation, but the brand name appeared in the answer text in just 38.3%.^[1] The short version: mentions and citations are different metrics, and you need to measure them separately. This page covers the definitions, the measured data, and which metrics to track in what order.

Brand mentions vs. citations: definitions in 30 seconds

• A brand mention is your brand name appearing in the text of an AI answer.^[1]
• A citation is your brand's domain shown as a source link in an AI answer.^[1]
• A linked mention is your brand name appearing in an answer that also links to your brand's domain.^[2]
• A ghost citation is your page linked as a source while the answer text never names your brand.^[1]
• An unlinked mention is your brand name appearing with no link to your brand's domain.^[2]

The IAB's AI visibility measurement standard draws the same line. Under the Presence dimension of its Four Ps, Mention Rate and Citation Rate are separate metrics: mentions describe "how often you are talked about," citations "how often you are used as a source."^[5][6]

Mentions vs. citations at a glance

Figures in the table are Semrush (2026) shares by appearance.^[1] The IAB metric split follows an explainer on the IAB Four Ps.^[6]

Evidence: what the measured data shows

1. Citations and mentions mostly happen apart

In the Semrush study, a citation and a mention appeared in the same answer only 13.2% of the time. A name with no citation made up 25.1%, and ghost citations (a link with no name) made up 61.7%.^[1] The sample covered 115 prompts, 14 countries, and four engines: ChatGPT, Google AI Overviews, Gemini, and Google AI Mode.^[1] A tool that only counts citations cannot tell you whether your brand was actually named in more than half of your appearances.

2. Engines lean in opposite directions

Gemini named brands and linked less; ChatGPT did the reverse.^[1] Blend the two into a single "AI visibility score" and that difference disappears. The multi-engine measurement guide shows how to break results out by engine.

3. The share of mentions that carry a link varies roughly fivefold by engine

Ahrefs pulled about 31,000 mentions of its own brand from the 150 million prompt database behind Brand Radar and counted, engine by engine, how often a mention came with a link.^[2] The average was around 28%, with AI Overviews lowest and Perplexity highest.^[2] This is data for a single brand, Ahrefs, so treat it as a directional signal.

Weighted by impressions (a proxy for search volume), the same analysis put Perplexity at 78.4% and Gemini at 71.3%.^[2] Linked mentions cluster in high-volume queries. Any time you report a link rate, state whether it is per mention or impression-weighted.

4. Clicks on citation links are rare

The idea that a citation equals traffic needs checking too. Pew Research Center analyzed 68,879 real Google searches by 900 U.S. adults and found that a link inside an AI summary was clicked in about 1% of visits.^[4] When an AI summary appeared, users clicked a standard search result 8% of the time, roughly half the 15% rate on pages without a summary.^[4] With clicks that scarce, whether the answer text names you (a mention) carries more of the weight for awareness.

5. Mentions are driven by signals off your site

Ahrefs studied 75,000 brands and found that the factor most strongly correlated with brand visibility in Google AI Overviews was the volume of brand mentions across the web (Spearman 0.664).^[3] Branded anchor text came in at 0.527, branded search volume at 0.392, and backlink count at 0.218.^[3] These are correlations, not causes. They still support a working hypothesis: citations depend on page quality, mentions on how much the brand shows up across the web. For how AI picks its sources, see How AI chooses its citation sources.

What to measure: four metrics

Fold mentions and citations into one "visibility score" and you lose the ability to diagnose anything. At minimum, track these four separately. Each formula is just the definition written out.

The ghost citation rate above uses cited answers as its denominator. Semrush's 61.7% uses all appearances, so the two numbers are not directly comparable.^[1]

Add the remaining IAB Four Ps dimensions to round out measurement: Prominence (position in the answer), Portrayal (accuracy and sentiment of the description), and Persuasion (strength of recommendation, clicks after citation).^[6] For details, see the IAB Four Ps explainer and the AI search share of voice (AI SOV) guide.

How to measure mentions and citations separately: five steps

1. Split your prompt set by type. Separate informational prompts (what is, how to) from comparison prompts (recommendations, comparisons, alternatives). Informational prompts had an 89.3% citation rate and an 18% mention rate; comparison prompts had a 43.3% mention rate.^[1] Mixed together, the average means nothing.
2. Collect results per engine. Log ChatGPT, Gemini, Perplexity, Google AI Overviews, and AI Mode individually. Engines lean in opposite directions, so combine them only at the very end.^[1]
3. Code every answer into one of four cells. Linked mention, unlinked mention, ghost citation, or not visible. Code competitors the same way so the comparison holds.
4. State your counting basis. Note whether figures are per mention or impression-weighted. On the same data, Gemini's link rate is 16.8% one way and 71.3% the other.^[2]
5. Match each metric to its own fix. A high ghost citation rate calls for more on-page definition sentences with your brand as the subject, plus proprietary data. A low mention rate calls first for more brand mentions on the outside web: reviews, comparison articles, communities.^[3] To confirm referral traffic, connect this to GA4 AI assistant channel measurement.

When choosing a tool, check two things first: does it export mentions and citations as separate columns, and does it store the raw answer text per engine? Those two questions work for global tools such as Semrush, Ahrefs, and Profound, and for Korean solutions such as BOIDA, which highlights support for Korean-language engines. Why tools report different numbers is covered in AI visibility tool measurement discrepancies.

Common misreadings

• Reading more citations as more brand awareness. Ghost citations may be what grew. They made up 61.7% of appearances.^[1]
• Reporting a single cross-engine average. The mention rate gap between Gemini and ChatGPT is more than 60 percentage points.^[1]
• Measuring mention rate on informational queries alone. The informational mention rate was 18%, less than half the comparison rate.^[1] Leave out comparison queries, which sit closer to purchase decisions, and your mention rate comes out low.
• Converting citations into traffic. Links inside AI summaries were clicked in about 1% of visits.^[4]

Wrap-up

A brand mention means your name gets said; a citation means your link gets used as evidence. In the measured data, the two signals mostly moved independently (13.2% co-occurrence), and they split in opposite directions depending on engine and query type.^[1][2] The starting point is four separate metrics (mention rate, citation rate, linked mention rate, ghost citation rate), recorded by engine and by query type. For GEO fundamentals, read What is GEO; for the full market picture, see the Global GEO/AEO landscape 2026.