Mention rate vs citation rate: what is the difference, and which should you track?
A mention is your brand name inside the answer text. A citation is your page listed as a source. The two come apart constantly: Semrush, in a study of 3,981 domain appearances published 9 June 2026, found 61.7% of appearances were citations where the brand name never appeared in the answer. Track both, per prompt, and write the denominator down, because a ratio quoted without its denominator is why two studies of the same thing can report opposite results. Then add a third column nobody reports: whether you were among the options the answer actually recommended.
What is the difference between mention rate and citation rate?
Mention rate counts how often an engine names your brand in the text of an answer. Citation rate counts how often one of your pages appears as a source attached to an answer. A buyer reads the first and ignores the second; your analytics sees the second and cannot see the first.
The two are measured from different places, which is why they behave independently. A mention comes out of the generated text, so it depends on what the model decides to say. A citation comes out of the retrieval and attribution layer, so it depends on which pages were fetched and judged usable. One answer can contain your page as a source and never say your name, or say your name three times and link to a competitor's blog.
Semrush's study of 3,981 domain appearances across 115 prompts and 14 countries, run with Growth Memo and published 9 June 2026, gave the first case a name: a ghost citation, defined as an appearance where the platforms used the page as a source link but the brand name never appeared in the answer. In that sample 74.9% of appearances included a citation while 38.3% included a brand mention, and 61.7% were ghost citations.
Both numbers are worth having for different reasons. Mentions are what shortlists are made of, because the buyer sees a name. Citations are the part you can move with a page, because a page is the thing being retrieved. Reporting only one of them hides whichever problem you have, and our page on what counts as a good AI citation rate covers how to read the citation half against a benchmark.
Why do two studies give opposite answers about mentions and citations?
Two studies give opposite answers because mention rate and citation rate are fractions, and the published versions divide by different things. Nothing is being faked. The denominators are simply not the same, and almost nobody states them.
Compare the two largest public samples. Semrush, above, reports citations as more common than mentions: 74.9% against 38.3% of appearances. BuzzStream, in a study of 12,000 AI responses last updated 14 July 2026, reports the overlap in both directions and gets 23.1% for mentioned brands that were also cited and 69.9% for cited brands that were also mentioned. Read carelessly, the first says citations dominate and the second says mentions dominate.
Read properly, they agree about the world and differ about the question. Semrush is asking what share of appearances carried each kind of evidence. BuzzStream is asking, given one, how often you also get the other. A brand can be in both datasets and produce a different headline percentage in each, which is exactly what happens to the vendors selling you dashboards built on either convention.
The practical rule is short. Any mention or citation rate you are shown is meaningless until you know what sits under the line: every answer generated, every answer where your brand appeared at all, or every prompt in the tracked set. Adding prompts or engines moves the first, and leaves the second roughly where it was. Our page on why two AI visibility tools show different numbers works through the rest of the discrepancy.
Which engines mention you without citing you, and which do the reverse?
Engines split on this sharply, and the split is big enough to change which metric is worth watching per platform. Gemini tends to name brands without linking them. ChatGPT tends to link without naming.
In the Semrush sample, Gemini mentioned brands in 83.7% of appearances but produced a citation link 21.4% of the time, while ChatGPT cited brands 87% of the time and mentioned them in only 20.7% of answers. Google AI Mode mentioned brands at nearly twice ChatGPT's rate. Those are close to mirror images of each other on the same prompts.
BuzzStream's per-platform figures, from 3,000 responses each on Google AI Mode, AI Overviews, Gemini and GPT-5.4-mini, land in the same place from the other direction: 92.7% of cited brands were also mentioned on GPT-5.4-mini against 63.9% on Gemini. Both studies are vendor research with stated methods rather than peer-reviewed work, and they used different engine versions in different months, so treat the direction as the finding and the decimals as approximate.
What follows for your tracking is a per-engine reading rather than one blended number. If your category is dominated by Gemini answers, a brand mention is the thing to chase and your citation count will look thin no matter what you do. If your buyers live in ChatGPT, citations are the earlier signal and a flat mention rate is not evidence the work has failed. Our guide to which engine to optimise for first covers how to decide which one you are actually in.
Does the type of prompt change your mention rate?
Prompt type changes mention rate more than almost anything you do to a page, and the effect runs in a useful direction: the prompts closest to a buying decision produce the most mentions. Comparison questions name brands. Definition questions do not.
In Semrush's breakdown by query type, informational prompts had an 89.3% citation rate against an 18% mention rate, while commercial prompts ran at 84.4% citation and 35.6% mention, and comparative prompts reached 43.3% mention, which the study puts at 2.4 times the informational rate. How-to prompts sat at 42.8% mention.
BuzzStream's prompt-type split shows the same thing from the overlap angle and adds a warning. Single-brand prompts had 39.0% of mentioned brands also cited; list and category prompts had 7.2%. So on exactly the prompts a buyer uses to build a shortlist, your name and your page are least likely to arrive together.
The reporting consequence is worth stating plainly, because it is where most monthly reports go wrong. A tracked set weighted towards definitional questions will show a healthy citation rate and a weak mention rate, and will look like a content problem when it is a prompt-selection problem. Fix the prompt list before diagnosing the pages, and keep it fixed afterwards, which our page on how many prompts to track treats as the first measurement decision.
Is a mention worth more than a citation?
A mention is worth more to the buyer and a citation is worth more to you as a diagnostic, so the question is not which to prefer but which to act on. The mention is the outcome. The citation is the lever you can reach.
There is a third state that neither metric captures, and it is the one that costs deals. Lily Ray, analysing 100 B2B best-of queries in Google AI Overviews at three dates between April and June 2026, found that where a brand's own self-promotional listicle was cited, the brand was left out of the recommendation in 224 of 323 cases, 69%. Cited, and not recommended. If your report counts that as a win, your report is wrong.
The mechanism behind the gap is now documented from a second direction. Chen and Koudas, in a study of 158,099 answers published 10 October 2026, found that the engines name brands they already know, sometimes with no supporting page at all, and that changing only a brand name reordered product rankings when the underlying facts were identical. Who gets named is partly a brand outcome. Who gets cited is mostly a page outcome.
Our own category study points the same way about where the citations land. Across six categories in our study of how four engines answer process skid sourcing questions, between 73% and 89% of citations went to suppliers' own capability pages rather than to directories or listicles, and no supplier was named by all four engines in any contestable category. Citations were winnable on the page. Being the one named was not settled at all.
Which numbers belong on your monthly report?
Put three numbers on the monthly report, one per prompt, and keep the prompt list still. The three answer different questions and only together do they tell you what to do next.
| Metric | What it counts | What it tells you to do | How it misleads |
|---|---|---|---|
| Mention rate | Share of answers naming your brand in the text | Work on presence off your own site, where brand memory is built | Moves when you add prompts or engines, even if nothing about you changed |
| Citation rate | Share of answers linking one of your pages as a source | Rewrite the page that should have been quoted and was not | Counts ghost citations as wins, where your name never appeared |
| Recommendation rate | Share of answers placing you among the options suggested | Nothing on its own; read it against the other two to see which half is broken | Needs a human or a careful rule to judge, so most tools leave it out |
Record two more things beside each figure: how many runs it came from, and what the denominator was. A single run of a prompt is noise, and a percentage whose base moved between months is not a trend. Our guide to tracking AI visibility across engines sets out the run counts we use, and reporting AI visibility to executives covers which of the three survives contact with a board slide.
If you would rather see the three numbers for your own category before building the spreadsheet, our free visibility assessment runs the prompts, records who is named and who is cited, and reports where you sit on both.
What else do people ask about mentions and citations in AI search?
The five questions below about mentions and citations come up when a marketing team is handed two dashboards that disagree and has to decide which number to put in front of a leadership meeting.
What is the difference between a mention and a citation in AI search?
A mention is your brand name appearing in the text of an AI answer. A citation is your page appearing as a linked source under or beside that answer. The two happen independently: Semrush's study found that 61.7% of appearances in its sample were citations with no brand name in the answer at all, a case it calls a ghost citation.
Which matters more for B2B, mentions or citations?
Mentions matter more for whether you get shortlisted, and citations matter more as a signal you can act on. A buyer reading an answer sees names, not footnotes. A citation tells you which of your pages the engine judged usable, which is the part a rewrite can change.
Why does my tool show a different mention rate than another tool?
Because mention rate is a fraction and the vendors choose different denominators. Dividing by every answer generated, by answers where your brand appeared at all, or by prompts in the tracked set gives three different numbers from the same data. Ask any vendor what sits under the line before comparing their figure with anybody else's.
Does being cited mean the AI recommends me?
No, and the gap is large. Lily Ray checked 100 B2B best-of queries in Google AI Overviews and found that where a brand's own listicle was cited, the brand itself was left out of the recommendation in 224 of 323 cases, which is 69%. Being used as a source and being named as an option are separate outcomes.
What should we put on the monthly report?
Report three things per prompt rather than one blended percentage: whether your brand was named, whether one of your pages was cited, and whether you were among the options recommended. Keep the prompt list fixed so the denominator does not move, and record the number of runs behind each figure.