AI Brand Mentions vs Citations: Why Being Named Is Not Being Cited

Almost every AI visibility report collapses these two into a single score. That is the mistake. You can be quoted by an assistant a hundred times and remain invisible to the reader, or be named constantly with nothing to click. This page explains both failure modes, what the published data says about how common each one is, and what to do differently about each.

What are AI brand mentions?

An AI brand mention is any occasion where an assistant writes your brand name into the text of its answer. It is a visibility outcome, not a traffic outcome. The reader sees your name, forms an impression, and may search for you later, but nothing on the page sends them to your site.

Mentions are driven by what the model associates with your category. If your name appears often enough, in enough independent places, alongside the right topics, assistants will produce it when the topic comes up. That is why mentions correlate with entity strength rather than with page optimisation.

What is an AI citation, and how is it different?

An AI citation is when the assistant links your page as a source for what it just said. It is a traffic and trust outcome. Citations are driven by page-level factors: whether your page answers the specific question directly, how retrievable it is, and how clearly the claim sits on the page.

The difference matters because the two are earned in different ways. A mention is earned across the web. A citation is earned on your page. Work that improves one does not automatically improve the other, which is the whole reason this distinction is worth a page of its own. Set out side by side, the AI citations vs mentions comparison comes down to six things.

The two ways your brand goes missing

There are two distinct failure modes, and they are opposites. Most teams only ever look for one of them, which is why the other keeps happening unnoticed.

The ghost citation. The assistant uses your page, links it, and never says your name. Kevin Indig coined the term, and Semrush defines it exactly that way in its Ghost Citations Study published 9 June 2026: your site gets a source link, but the AI never says your name in the answer. Your content did the work. The reader has no idea it was yours.

The unsourced mention. The assistant names you and links nothing. The reader hears about you and has nowhere to go. This is the direction almost nobody measures, and it is more common than the ghost citation.

How often does this actually happen?

Often enough that a single-number visibility score is close to useless. Three independent studies published within eight weeks of each other all found large gaps, in both directions.

Read the BuzzStream pair together, because they are the clearest statement of the problem. Roughly seven in ten citations do name the brand, but fewer than one in four mentions come with a link. Getting named is the easier half. Getting named and linked in the same answer is rare.

Which AI platforms hide your brand most?

Perplexity and Google AI Mode are the worst offenders, and Microsoft Copilot is the best behaved. The Search Engine Land analysis by Nikki Lam and Samanyou Garg, published 29 July 2026, examined roughly 16 million brand appearances over a 30 day window and broke the ghost citation rate down by engine.

The spread is the point. A brand tracked only on Copilot and a brand tracked only on Perplexity are not measuring the same thing, and neither number transfers to the other engine.

Gemini and ChatGPT behave in opposite ways

Semrush's study found the two largest assistants running almost mirror-image patterns. When a brand appears in a Gemini answer it is named in the text 83.7% of the time but cited as a source only 21.4% of the time. ChatGPT does the reverse, with an 87% citation rate against a 20.7% mention rate.

Gemini talks about you. ChatGPT quotes you. If your reporting blends the two engines into one score, those two opposite behaviours cancel each other out and you learn nothing.

There is a structural reason underneath this. The Semrush 2026 AI Visibility Index, published 26 June 2026 from 126 million US prompts, found ChatGPT cites an average of 15 sources per response while Gemini cites an average of 3. More citation slots means more chances to be linked, and less pressure to name anyone in the prose.

Why the studies disagree, and what to do about it

Semrush says 62%, Writesonic says 40%, and both are right about their own dataset. Different prompt sets, different engines, different time windows and different definitions of a brand appearance produce different numbers. That is not a flaw in the research, it is the nature of measuring a probabilistic system.

The practical rule is the one set out in our guide to measuring AI visibility: never compare a number from one tool against a number from another. Compare your own measurement against your own earlier measurement, taken the same way, on the same prompts, on the same engines.

It also explains why so many teams feel lost here. The Semrush index found that 45% of marketing leaders cannot accurately measure their brand visibility within AI generated answers. That is not incompetence. It is the absence of a shared standard until very recently.

Should you optimise for mentions or citations?

Both, but not with the same work, and not in the same place. Mentions are won off your site. Citations are won on it. Neither, it is worth saying, requires rebuilding your website, which is a question we take apart in SEO vs GEO.

To earn more AI brand mentions

If the goal is to increase visibility with AI brand mentions rather than clicks, the work happens away from your website.

• Build entity consistency first. If your name, description and category are inconsistent across the web, assistants cannot form a stable association. Fix that before chasing volume.
• Get into third party round-ups. Kumar's analysis of 100,000+ prompt responses found 21% of citations come from best-of listicles, and those pages are also where your name sits beside your category.
• Publish something worth naming. Search Engine Land's recommendation from the ghost citation data is direct: develop proprietary named research. An assistant summarising a generic claim has no reason to say who said it. An assistant summarising your named study has to.

To earn more AI citations

• Answer the question in the first 60 words of the section. Retrievable answers get lifted. Buried ones do not. Our guide on getting cited by AI chatbots works through this page by page.
• Put your brand attribution next to the claim. This is the sharpest fix in the Search Engine Land piece. If the sentence containing the finding does not contain your name, the assistant has nothing to carry across when it summarises.
• Keep your own site in order. Kumar's data found 78% of citations go to corporate websites. Your site is not a secondary channel here, it is the primary one, and it is where AI SEO work actually happens.

A practical process you can run this month

Six steps. The first four are free and need no tool at all.

• Write 20 to 30 real buyer questions. Natural language, not keyword strings. Cover informational, comparison and cost intent. Freeze the list so results stay comparable.
• Run each prompt on each engine, and log two columns, not one. Column A, were you named. Column B, were you linked. This is the whole method: to monitor brand mentions in AI search properly you have to record the citation alongside the mention, because it is the gap between the two columns that tells you what to fix.
• Repeat the run. Answers vary between runs, so a single pass is a sample rather than a measurement.
• Classify each result into one of four buckets. Named and linked, named only, linked only, absent. The two middle buckets are your work list.
• Diagnose by bucket. Heavy "linked only" means your brand name is too far from your claims. Heavy "named only" means your pages are not answering the question directly enough to be worth linking.
• Re-measure on the same prompts. Same questions, same engines, same number of runs. Anything else is a new experiment, not a comparison.

Tools, and what a usable one has to do

You do not need a paid tool to start. You do need one to sustain it at any scale.

What you already have, free. Your server logs and analytics will show AI crawler activity and referral traffic from assistant platforms, which tells you about citations. They will tell you nothing about AI brand mentions, because a mention generates no request. That blind spot is exactly why mentions need deliberate tracking.

The manual prompt panel. A spreadsheet, a frozen prompt list and a scheduled hour. It is slow, it does not scale, and it is more honest than most dashboards because you can see every answer you counted.

What to demand from any paid tracker. The Interactive Advertising Bureau published Measuring Visibility in the AI Era on 3 August 2026 precisely because more than 20 vendors sell this measurement using incompatible methods. It sets a bar worth borrowing. Before you buy, insist the tool discloses query volume, sample size, prompt type coverage, testing cadence, reproducibility and platform coverage. On top of those six, add one more for this topic specifically: it must report mentions and citations as separate figures. A tool that gives you one blended visibility score cannot tell you which of the two problems you have.