What is the difference between a mention and a recommendation?

A mention names the brand in any role, including as history or an example. A recommendation presents the brand as a good choice for what the person asked.

What is an AI citation?

A citation is a link or reference the answer gives to a source page it relied on. The page may be your site or someone else's.

Can I be recommended without being cited?

Yes. An assistant can name you from training memory with no source shown. You can also be cited without being named in the text.

Which one matters most?

Recommendations usually matter most for buying decisions, but citations and mentions are earlier signs that a brand is entering the picture. Track all three separately.

Mentioned, recommended or cited: the three ways AI names a brand: Citewake

AI assistants name a brand in three distinct ways. A mention is the brand appearing in the answer in any role. A recommendation is the answer presenting the brand as a good choice for the question. A citation is a source the answer points to, which may be the brand's own site or a third-party page that discusses it. They often travel together, but they are different events, and a measurement that merges them can mislead.

What counts as a mention?

A mention is the lowest bar. The brand's name appears in the answer. That could be as a recommended option, but it could equally be:

a historical note ("X was an early player");
a point of comparison ("unlike X, which charges per seat");
an example in a list of well-known names;
a warning ("X is not suited to small teams").

Mentions come in tones. A neutral mention names the brand without recommending it. A negative mention presents it as unsuitable or inferior. Counting every mention as good news is a common error. A brand can be mentioned often because it is the thing people are warned about.

What counts as a recommendation?

A recommendation is the assistant putting the brand forward as something to choose for the question asked. Language such as "a good option is", "I would suggest", "well suited to agencies" or a list under a heading like "best tools" usually signals it.

Judging it takes some care. Not every item in a list is a recommendation of equal strength, and some lists are only descriptive. Order does not prove preference either: a list may be alphabetical, chronological or arbitrary. A top pick is a stronger claim than inclusion, and it is only safe to call it that when the answer says so in words, such as "my top choice". More on that in why there is no real rank #1 in AI answers.

What counts as a citation?

A citation is a pointer to a source. Assistants that search the web usually show links beside their answers, or listed beneath them. A citation tells you which page the assistant relied on for some part of the response.

Two distinctions help:

Own-site citation: the cited page is on your domain.
Third-party citation: the cited page is elsewhere, such as a roundup, review or article that may list you, or may list only your competitors.

A third-party citation is a clue about where your visibility is decided. If a roundup is cited and you are not on it, that page is an obvious target. How to get onto the roundup and review pages AI relies on covers the next step.

Why keep the three separate?

Because they answer different questions and can move independently.

Signal	What it tells you	Common misreading
Mention	The assistant knows you exist	Treating any mention as endorsement
Recommendation	The assistant puts you forward	Assuming it holds across questions
Citation	A page is shaping the answer	Assuming a cited page means a recommended brand

A new brand may be cited on a few pages long before it is recommended. An established brand may be mentioned constantly and recommended rarely. If you collapse these into one score, a change in one can hide a change in another, and you lose the diagnosis.

Citewake's method keeps them apart for this reason: recommendation, own-site citation and presence on cited third-party pages are measured as separate things, each with a fixed denominator.

How do the three relate?

There is a plausible chain. Pages that list you get cited. Being listed on cited pages helps the assistant form a view of you. That view may then show up as a mention, and eventually as a recommendation. This is a reasonable working idea rather than a proven rule, so treat it as a hypothesis to watch in your own data.

What you can verify directly is simpler. Ask an assistant a buying question, look at the names, then look at the sources. Compare the two lists. Often you will see brands in the prose that match brands on the cited pages.

How do you count them in practice?

Make a simple record for each answer:

Is the brand named? If so, in what role: recommended, neutral or negative?
Is any source on the brand's own domain?
Do any cited third-party pages list the brand?

Do this across a fixed set of questions and several runs, and report each as a rate. Share of voice in AI answers shows how to turn counts into a comparison with competitors.

Where should you start?

Start with the most basic check. Ask a few buying questions, leave your name out, and sort what you see into the three bins. The Citewake checker offers a first look, and how to check if ChatGPT recommends your business walks through one assistant in detail. Knowing which of the three you are missing tells you what to do next.

A worked example

Imagine an assistant is asked for the best project management tool for a small agency. It writes a paragraph naming four tools. Three are presented as good fits. The fourth is described as "popular but better suited to large engineering teams". Below the answer, it cites two pages: a roundup article and the official site of one of the three.

Sorting this out gives a clear picture. Three brands are recommended. The fourth is mentioned, with a negative or at best neutral tone. One brand has an own-site citation. The roundup is a third-party citation, and whether a brand is "present" on it depends on whether the roundup lists that brand.

Notice how much would be lost by a single count. "Four brands named" hides that one was steered away from. "Two sources cited" hides that one was the brand's own site. Separating the three signals is what lets you see which part of the picture needs work.