AI mentions vs citations vs recommendations: the difference
Short answer: AI visibility reports count three different things under similar words. A mention is your brand named in an answer. A recommendation is being put forward as a choice on a question that did not name you. A citation is the answer relying on you as a source, by linking to you or naming you as its source. They happen at very different rates, so before you compare tools, or your own numbers month to month, check which one you are looking at.
| Brand mention | Recommendation | Source citation | |
|---|---|---|---|
| What happened | Your brand is named in the answer text | You are put forward as an option to choose | The answer links to your site, or names you as its source |
| Example | “Firms in the area include Acme and others.” | “For small charities, Acme is a strong choice.” | “…according to Acme’s guide [acme.example]” |
| Needs a question without your name? | No | Yes | No |
| Counts if… | Your name, an alias or your bare domain appears in the visible text | The question was open and you were endorsed, not just listed | A link to your own domain, or an attribution such as “according to Acme” |
| Does not count if… | Your domain only appears inside a link address | You were named only because the question named you, or negatively | The content only looks as if it came from you, with no link or attribution |
The examples use a fictional business, Acme, and are constructed for illustration.
What counts as a mention?
A mention is your brand appearing in the text of the answer. In the SearchScore Tracker the detector matches your display name, any aliases you have given us, and your bare domain written in the text, such as “acme.example is worth a look”. It then removes some cases that look like mentions and are not:
- Name only inside a link. If your domain appears only in a link’s address and never in the words the reader sees, the answer used you as a source without naming you. We do not count that as a mention.
- A “no information” answer. “I don’t have information about Acme” contains the name but is an absence, not a mention.
- A different business with the same name. Where the answer clearly attributes the name to another company or product, it is not counted.
What counts as a citation?
We follow the IAB’s definition in Measuring Visibility in the AI Era (August 2026): “A citation is a linked reference (containing a hyperlink to the source) or a named reference (identifying the publication by name without a hyperlink). Implied citations, where content appears drawn from a source without explicit attribution, are excluded because they cannot be measured consistently across platforms or providers.”
In our Tracker that becomes two checks:
- Linked: a hyperlink to your own domain, either written into the answer or in the list of sources the AI platform returned with it. A link to a directory listing about you, or to a competitor, is not your link.
- Named as a source: a deliberately narrow set of attribution phrases, such as “according to Acme”, “Acme says”, “Acme reports”, “source: Acme” or “Acme’s data”. A loose pattern would turn the citation rate back into a mention rate, which defeats the point.
One difference from the IAB wording matters. IAB’s definition would accept a link to your site even if your name never appears in the answer. Our detector only checks for a citation once a mention has been found. So in SearchScore figures every citation is also a mention, and a source link to your site with no visible mention of your name counts as neither. Other tools may count that case differently, which is one reason two tools can disagree.
Where does “recommended” fit?
“Recommended” is the word buyers care about, and it is the hardest to measure. A mention can be:
- a neutral listing: “Firms in Bristol include Acme, Beta and Gamma.”
- an endorsement: “Acme is the best fit if you need weekend appointments.”
- a negative mention: “Some reviews say Acme is slow to respond.”
- a quoted source: “According to Acme’s pricing guide, most jobs take two days.”
Only the second is a recommendation in the plain sense. And it only means much on a question that left the engine free to name someone else. On “is Acme any good?”, being named shows the engine found you, not that it chose you. Our guide to citation sentiment covers how to judge whether the mention is favourable.
The SearchScore Tracker does not report a recommendation rate. Two things in it get closer to the idea. Its headline mention rate counts only decision-stage questions, and leaves out questions that name your brand, so a brand cannot inflate it with questions it nearly always wins. And its AI Visibility Score gives each mention more or less credit depending on its sentiment and its position in the answer. Neither is the same as a measured “recommendation rate”, and we do not present them as one. Our free check and product pages use “recommended” in its everyday sense; the numbers behind them count mentions, as defined here.
Three illustrative answers
These are constructed examples, not real AI answers. The question in each is “Who should I use for boiler servicing in Leeds?” and the fictional brand is Acme Heating, acme.example.
Recommended, not cited. “For servicing, Acme Heating is well reviewed and offers fixed prices. Beta Gas is another option.” Acme is named and endorsed. There is no link to acme.example and no attribution, so it is a mention, arguably a recommendation, and not a citation.
Cited, not recommended. “Boilers should be serviced every year, according to Acme Heating’s maintenance guide [acme.example/guide]. For Leeds, popular choices include Beta Gas and Gamma Plumbing.” Acme is named and linked as a source, so it is a mention and a citation. It is not recommended: the answer sends the buyer elsewhere.
Both. “Acme Heating is a good choice: its published prices start from a fixed fee [acme.example/prices], and it covers all of Leeds.” Named, endorsed and linked.
How are the rates calculated?
All three use the same denominator: the readings that produced an answer. A reading is one question on one engine in one scan, and a failed call is left out of both the top and the bottom of the fraction, never counted as “not mentioned”.
- Mention rate = readings with a mention ÷ readings that answered
- Strict citation rate = readings with a linked or named citation ÷ readings that answered
Worked example (illustrative numbers): 10 questions, asked on 6 engines, twice each, is 120 readings. Say 4 calls fail, leaving 116. If 29 answers mention you, the mention rate is 29 ÷ 116 = 25.0%. If 6 of those also link to you or name you as a source, the strict citation rate is 6 ÷ 116 = 5.2%.
Two details change real reports. The Tracker’s headline works this out per engine and then averages the engines that answered, so one engine’s outage does not drag the number down. And the headline uses decision-stage questions only. A report that pools every question, including ones that name you, will give a different, usually higher, mention rate.
How big is the gap in practice?
Across 6,457 question-and-engine readings stored in the SearchScore Tracker between 2 June 2026 and 11 August 2026, over 15 domains and 109 scans, the mention rate was 24.3% and the strict citation rate was 4.4%, a difference of about 5.5 times (SearchScore measurement standards). That figure pools every stored question, including ones that name the brand, so it is not the Tracker’s headline formula. It describes those 15 brands and those scans, not AI answers in general.
The gap depends heavily on the engine. In the same data, Perplexity returned a source list on 65.8% of its answers and Gemini on 11.4%.
Whatever the exact figures, the direction is consistent: citations are much rarer than mentions, because a citation needs two things to happen. The engine has to return sources at all, and one of them has to be you. An engine that rarely returns links cannot cite you with a link, however well you are doing, so a low citation rate on that engine says as much about the engine as about your brand. The full method, the per-engine table and the caveats are on our measurement standards page.
Which number should I track?
Both, separately, for different reasons.
- Mention rate tells you whether AI assistants put your name in front of buyers. It is the better measure of reach, and it is the headline in the SearchScore Tracker.
- Citation rate tells you whether AI treats your pages as a source worth linking to. It matters most for clicks and on engines that show sources.
They can move independently. A rising mention rate with a flat citation rate can mean engines know your brand but use other sites to describe you. That is a prompt to look at which sources AI cites in your category.
What should I ask of any AI visibility report?
- What counts as a hit? Any appearance of the name, a link, or the brand named as a source? Does a bare source link count without the name?
- Which questions? Questions that include your brand name inflate the rate. Open category questions are the harder, more honest test.
- What happens to failed calls? If they count as “not mentioned”, the rate falls whenever an engine has an outage.
- How are duplicates handled? Is one answer that names you three times one hit or three? Rates should count answers, not occurrences.
- Is the word consistent? If the dashboard says “cited” and the method page says “mentioned”, the dashboard is overstating.
Our guide to AI visibility tool accuracy covers other reasons two tools give different numbers for the same brand. A mention or citation rate is also a different thing from a website readiness score: see what a GEO score measures.
Where to start
If you have not measured anything yet, start with the free methods, including the free AI recommendation check. When you need the same questions measured every week across 6 engines, the SearchScore Tracker reports the mention rate as its headline and the strict citation rate beside it, and never labels one as the other.
Sources
- IAB, Measuring Visibility in the AI Era, August 2026 (Presence metrics: Mention Rate, Citation Rate; Publisher metrics: citation definition): https://www.iab.com/guidelines/measuring-visibility-in-the-ai-era/ (checked 12 November 2026)
- SearchScore, Measurement standards, “A mention is not a citation”: https://searchscore.io/measurement-standards/
Definitions on this page describe the SearchScore Tracker’s detectors as implemented on 12 November 2026. If the detectors change, this page changes with them.
Frequently asked questions
What is the difference between an AI mention and an AI citation?
A mention is your brand named in the text of an AI answer. A citation is the answer relying on you as a source: a link to your site, or your brand named as the source of the information. The IAB framework defines a citation as a linked reference or a named reference, and excludes content that only appears to be drawn from a source.
Is every citation also a mention?
It depends on how a tool counts. Under the IAB definition a link to your site could count as a citation even if your name never appears in the text. In the SearchScore Tracker, a citation is only checked for once your brand has been detected as mentioned, so every citation we count is also a mention, and a source link to your site with no visible mention of your name counts as neither.
Is being mentioned by ChatGPT the same as being recommended?
No. A neutral list, an explicit endorsement, a negative mention and a quoted source are all mentions. Being recommended means being put forward as a choice on a question that did not already name you. Check what kind of questions and what kind of mention a tool counted before reading it as a recommendation.
Does the SearchScore Tracker report a recommendation rate?
No. Its headline is a mention rate on decision-stage questions, which leaves out questions that name your brand. It reports the strict citation rate separately. Recommendation is a distinction we use to read the results, not a separate metric we report.