Brand mentions vs citations

When an AI engine answers a buying question, two separate things can happen to your brand. The answer can name it, and the answer can cite your website as a source. Teams often treat these as one metric called "visibility", but they can come apart, and they tell you different things.

The two definitions

A brand mention is your brand's name appearing in the text of the answer. "For small teams, consider Acme, Beta and Gamma" mentions three brands. The reader sees the name, and in a recommendation-style answer the engine is effectively putting you on a shortlist.

A citation is a source the engine links to or lists alongside the answer. It is a URL, usually shown as a footnote, a card or a source panel. Our glossary entry on the AI citation covers the term in more detail.

Why they differ

The engine writes the answer text and picks the sources through related but distinct steps, so the two lists do not have to match.

An engine can name you without citing you. It may know your brand from training data or from several third-party pages, then list you in the answer while the links point at review sites and comparison articles. Your own domain never appears.

An engine can also cite you without recommending you. Your pricing page or documentation may be used to support a statement about a category, or about a competitor, and the answer never names your brand as an option. You taught the engine something and got no shortlist place for it.

Our benchmark shows how far apart the two behaviours can sit. We ran 200 non-branded buyer prompts across 7 engines for 14 days, from 30 September to 13 October 2026, in US English, and collected 18,792 answers. Perplexity cited at least one source in 100% of its answers and averaged 10.1 citations per answer, the most of any engine. Yet it named the fewest brands, 3.9 per answer. Copilot runs the other way: 4.8 citations per answer, but 7.6 brands named, against an average of 6.2 brands per answer. ChatGPT falls between, with 9.3 citations and 5.5 brands per answer.

Citation coverage is high everywhere but not uniform. The share of answers with at least one citation was 100% for Perplexity, 99.5% for ChatGPT, 98% for Copilot, 96.3% for AI Overviews, 95.3% and 96.4% for the two Claude models, 94.8% for AI Mode and 90.7% for Gemini. Because nearly every answer cites something, the question is not whether sources appear but whether yours are among them, and whether your name appears in the text.

You can read the full method and results in the AI search visibility benchmark.

What each one is good for

Mentions are the outcome. They approximate recommendation and share of voice: of all the brands named across a set of prompts, how many are yours, and where in the answer do you appear. Position and sentiment belong here too, since being named last with a caveat is different from being named first with praise.

Citations are the lever. They show which pages the engine relied on when it wrote the answer. You cannot edit a model, but you can see which pages it read and change what those pages say, or earn a place on them.

Reading them together gives four cases:

• Mentioned and cited: the position is working. Protect the pages that earn it.
• Mentioned but not cited: the engine knows you from elsewhere. Look at which third-party pages are cited in your place.
• Cited but not mentioned: your content is useful, but nothing ties it to a recommendation. Check whether the page answers the buyer's actual question or only a background one.
• Neither: start from the prompts where competitors appear and see what is cited for them.

How to measure each

Engines vary between runs and between days, so measure over repeated runs of a fixed prompt set, separately for each engine.

For mentions, run each prompt several times per engine and record whether your brand is named, in what position, with what sentiment, and which other brands appear. The mention rate is the share of runs that name you. Share of voice is your mentions divided by all brand mentions in the set.

For citations, record every source URL in every answer. Then calculate how often your own domain is cited, which of your pages are cited, and which other domains appear most often for the prompts you care about. Group the results by engine, because the citation habits differ: an average of 10.1 sources per answer on Perplexity and 4.8 on Copilot are not comparable counts.

Keep the prompts non-branded. A prompt that includes your name will mention you almost by definition and tells you nothing about discovery.

What to do with the citation list

Sort the cited URLs by how often they appear across your prompt set, then work through them in order.

1. Your own pages. Find which of them are cited and for which prompts. Keep them accurate and current, and make sure the claims buyers ask about are stated plainly on the page.
2. Gaps on your own site. Where competitors are cited from their own domains and you have no equivalent page, that is a content brief.
3. Third-party pages. Comparison articles, review listings and directories that appear repeatedly are places where being included or described correctly changes the answer. Check that your entry is present and accurate, and contact the publisher where it is not.
4. Everything else. Forums, documentation and news pages show what kind of source each engine prefers for a given prompt type.

Then re-run the same prompts and compare. Changes to a source rarely show up in answers immediately, so give it a few weeks of daily data before judging.

Tracking both in Pineprompt

Pineprompt tracks mentions, position, sentiment, cited sources and share of voice every day across six live AI platforms, so both signals sit in one view per prompt and per engine. The citation tracking feature lists the sources behind each answer, which is the starting point for the steps above. There is no free trial; see pricing for plans.