Mentions, Citations and Recommendations Are Three Different Things

What is a mention, a citation and a recommendation?

A mention is the simplest: your brand's name appears somewhere in the answer text. Nothing more is required — it could be a passing comparison, a list entry, or a single clause.

A citation is a source attributed in the answer — a link, a footnote-style reference, or a named publisher credited for a claim. Critically, a citation doesn't have to point to your own site. An engine can cite a review site, a comparison article, or a news piece that mentions you, without ever crawling or linking your own pages.

A recommendation is the answer actively positioning your brand as a good fit for the question asked — not just present in a list, but framed as the pick, or one of a short set of picks, for what the person described. A recommendation can appear with or without a citation: a model answering from what it learned during training can recommend a brand by name with no source attached at all, because there's nothing to cite — it isn't retrieving anything live.

These three can combine in any order. Google's own guide to its generative AI features describes the underlying mechanism — pages retrieved and ranked, then an answer generated and grounded in them — without collapsing "retrieved," "cited" and "recommended" into one concept, which is the same distinction this post makes at the content level.

How do they appear in a real answer?

In the abstract: an answer can name three brands, link to a source for only one of them, and recommend a brand it never named directly cited at all (for example, describing a type of product and then naming your competitor as "a strong option" with no linked source). Four combinations are worth knowing by name:

Illustrative pattern table, not a transcript of any real answer.

Three real examples, run live for this post, one per pattern:

Named, not cited. Asked "what are some well-known email marketing tools?" the answer named Mailchimp first — "one of the most well-known email tools out there" — alongside HubSpot, Constant Contact and others. None of the nine linked sources pointed to mailchimp.com, or to any of the other named tools' own sites; every citation was a third-party roundup or review page.

Cited, not recommended. Asked "is Salesforce too complex for a small sales team?" the answer named and described Salesforce at length — steep learning curve, enterprise-first, "47 pipeline stages" — and cited several sources specifically about it (a Capterra reviews page, an AWS Marketplace reviews page, two SMB-advice blogs). The bottom line the answer gave: "Most small sales teams don't need Salesforce." Named, extensively cited, and actively not recommended, in the same response.

All three together. Asked "what are the best project management tools for small teams?" (the same check behind how ChatGPT chooses its sources), the answer recommended Plaky as a budget pick — "unlimited users and files for free with no user caps" — and cited plaky.com directly, the top-ranked source in the underlying search. Worth noting: that citation was to Plaky's own blog post, not an independent third party — still a real citation, but a different kind of evidence than a neutral source vouching for the brand.

Evidence, all three: AI engine (Claude, web search enabled), 10 October 2026, logged-in research session. Prompts: "What are some well-known email marketing tools?"; "Is Salesforce too complex for a small sales team?"; "What are the best project management tools for small teams?"

Which matters most for which goal?

It depends on what you're trying to prove. If the goal is basic awareness — does the engine even know we exist in this category — a mention is the floor and the easiest to get. If the goal is traceable discovery, where someone could click through to verify a claim or visit your site, citation is what matters; a mention with no citation gives a reader nothing to act on. If the goal is actually winning the moment someone asks "which one should I pick," recommendation is the one that counts — you can be named and cited in an answer that ultimately recommends a competitor.

Most teams default to tracking mentions alone because it's the easiest to detect. That's a reasonable starting metric, but reporting it as "AI visibility" without separating out citation and recommendation rates overstates what you actually know. A brand mentioned in 80% of answers but recommended in none of them has a real gap that a single blended number hides.

How should each be counted?

Count all three per answer, per engine, per run — not as one combined pass/fail:

• Mention: does the brand name appear anywhere in the response text? Yes/no, per answer.
• Citation: is there a linked or named source attributed to a claim about the brand? Separately note whether that source is the brand's own domain or a third party.
• Recommendation: does the answer explicitly position the brand as a fit for the question — "a good option for," "recommended for," appearing first in a short list framed as picks — versus simply appearing in an unranked list of names?

Log these as three separate rates across your prompt set (for example: "named in 9 of 10 runs, cited in 4 of 10, recommended in 2 of 10"), the same way share of voice should be calculated against a fixed, stated prompt set rather than one lucky run.

What are common counting mistakes?

• Treating any citation as "your site was crawled." A citation to a third-party article that discusses you isn't evidence your own pages are indexed or readable by the engine — check the citation's actual URL, not just that a citation exists.
• Counting a recommendation that has no citation as unreliable and discarding it. A citation-free recommendation from training memory is still a real recommendation — it's informative about what the model has learned about you, even though it isn't traceable to a live source the way a cited answer is.
• Blending all three into one "visibility" percentage. This is the single most common mistake, and the one our piece on single-number AI visibility scores covers in more depth — a blended score can rise because mentions went up while recommendations stayed flat, and nobody reading the number would know.
• Counting a competitor mention as "we were visible too." Two brands named in the same answer isn't the same outcome for both — check whether you were the one cited and recommended, not just present.
• Not noting signed-in state or personalization. A logged-in session, prior chat history, or location settings can change which sources an engine pulls and which brand it recommends; an unlabeled screenshot is hard to trust or reproduce.

FAQ

Can a brand be cited without being named? Not in the usual sense — a citation is attached to a specific claim about a specific brand, so if you're cited, you're also named. The reverse is far more common: named with no citation at all.

Does a citation always mean the engine crawled my site? No. The citation can point to a third-party page that discusses your brand — a review site, a news article, a comparison piece — without the engine ever visiting your own domain.

Is a recommendation with no citation less trustworthy than one with a citation? It's less traceable, not necessarily less accurate. A citation-free recommendation usually means the model is answering from training memory rather than a live search; it can still reflect what the model has learned about you, correctly or not.

How many mentions, citations and recommendations should I expect to see? There's no universal benchmark — it depends on your category, how many runs you check, and which engine. Track your own numbers over time and against named competitors rather than comparing to an external "good" number nobody has published honestly.

What to do next

1. Ask five to ten buyer-style questions on one engine, in fresh chats, and log mention, citation and recommendation separately for each answer — not a single yes/no.
2. Check where each citation actually points: your own domain, or a third party discussing you.
3. If recommendations are rare even when mentions are common, that's your real gap — not "low visibility" but "named, not chosen."

The AI Citation Snapshot shows one live answer to one question and marks whether you were named and which sources were cited — a fast way to see the distinction on a real answer rather than a hypothetical one.