AI mentions versus citations: what each tool counts, and how to report both

If ChatGPT recommends your brand but links to a review site, you have the mention and someone else has the citation. Both matter, different tools measure them, and many AI visibility reports blend the two into one number. This guide sets out what each data source actually counts, so you can see which of the two you are winning and what to fix.

The answer in brief

• A mention is your brand named in an AI answer. A citation is a link the answer shows as a source. Either can happen without the other.
• Google Search Console and Bing Webmaster Tools count only citations of your own site. Google reports them as impressions in AI Overviews and AI Mode, and Bing as citations in Copilot and Bing.
• Mentions only show up in a prompt log you keep yourself or in a vendor tool such as Semrush or Ahrefs, and each vendor defines a mention in its own way.
• GA4 counts neither. It counts visits, so it sees a citation only when someone clicks it.
• Report mentions and citations on separate lines, and split citations into your own pages and third-party pages that talk about you.

What is the difference between an AI mention and a citation?

A mention is your brand name in the text of the answer. “Tesco and Asda both offer click and collect” mentions two brands. Nothing in that sentence sends anyone to either website.

A citation is a link the answer shows as its source: an inline link, a numbered marker or a card in a sources panel. OpenAI’s help page on ChatGPT search says “Responses that use web search may include citations”, and that the Sources button shows “cited sources and other relevant links”. Google describes AI Overviews and AI Mode in similar terms, as answers grounded in its search index that show “prominent, clickable links to relevant web pages that support the information in the response”.

There is a third state most teams miss. Ahrefs’ Brand Radar documentation separates pages that are cited from pages that were “found but not cited”: pages the AI retrieved while writing the answer but did not show as sources. In ChatGPT these sit under “More” in the Sources panel. That is a vendor’s definition, not OpenAI’s, but it matches what you see on screen, and it is a useful signal. In our view, a page that is regularly found but not cited is close to winning the citation.

Why can a brand be mentioned without being cited?

There are two common reasons.

Not every answer searches the web. OpenAI says ChatGPT “may search the web automatically when your question would benefit from current information”. When it does not search, the answer comes from what the model already knows and usually shows no sources, so a well-known brand can be named again and again without a single link.

When the answer does search, the page it cites about you is often someone else’s: a review site, a comparison article, a forum thread or a news story. Semrush defines a citation as an AI response that cites your domain, so in its toolkit an answer that names you but links to a review site counts as a mention only. Ahrefs says of its own data that “most AI citations for your brand come from third-party websites, not your own”. That is vendor opinion, but you can check it against your own prompts in an afternoon.

The reverse happens too. Your page can be cited as evidence for a general point, a definition or a statistic while the answer recommends a competitor. You get the citation, and possibly a visit, but not the recommendation.

Semrush’s UK AI Visibility Index, published on 13 October 2026, put numbers on the gap. On average across its categories and platforms, fewer than half of the top 100 most-mentioned brands were among the top 100 most-cited domains, in every month and in both markets it studied. We covered the UK findings, including the Tesco and Asda example, in our Semrush Spotlight 2026 report. The point here is narrower: a single AI visibility number hides one of the two.

What does each tool actually count?

This is where reports often go wrong. “Visibility” means something different in each tool.

Sources: Google Search Console Help, Microsoft Bing blog (June 2026), Semrush and Ahrefs help centres, checked 22 October 2026.

Google’s Search Console help defines the report’s impressions as “how many times links to your site were shown to a user in a generative AI feature on Google Search”. The report rolled out to all sites worldwide on 31 August 2026, and the help page describes impressions only, grouped by page, country, date and device. A brand named in an AI Overview without a link to your site does not appear in it, and nor does a review site cited in your place. Where the link sits matters too, as our post on AI Overviews links and AI Mode clicks shows.

Microsoft defines Citation Share as “the percentage of citations attributed to your site out of all citations shown across all sites for that same grounding query”, and calls it “an observational metric” that is “not a ranking system or a competitive scoreboard”. Use it to see whether your share for a grounding query is moving. It cannot tell you which rival took the rest, because it does not show competitor domains.

GA4 sees the click, not the answer. ChatGPT links usually carry utm_source=chatgpt.com, but from 8 to 16 October 2026 many arrived without it and landed in Direct. Our GA4 audit guide shows how to check your own figures for that period.

The vendor tools count differently from each other. Semrush counts a mention for each prompt in which your brand appears, and a citation for each AI response that cites your domain. Ahrefs counts a mention once per response however many times the brand appears, and gives a domain one citation per response however many of its pages are cited. Google is blunt about all of them: “No third-party tool has access to our internal ranking or AI systems.” Two tools can disagree about the same brand in the same month and both be right by their own rules, so compare each tool with itself over time, not with another tool.

Which of the four positions is your brand in?

Run the same set of prompts and sort every answer into one of four positions. The fix depends on the position.

Mentioned and cited. The answer names you and links to your page. Protect it. Keep that page current, and check the answer still describes you correctly.

Mentioned, not cited. The answer names you, but the link goes somewhere else or nowhere. Look at what is cited. If it is a third-party review or comparison, check what that page says about you, and ask whether your own site has a page that answers the question as directly. If nothing is cited, the answer may have come from the model without a search, and the levers are slower: accurate, consistent facts about your business on the sites that write about your market.

Cited, not mentioned. Your page supports the answer, but the answer recommends someone else. Usually the page answers a general question well and says little about what you sell. Add the specific, checkable details that would let an answer name you: who you serve, what it costs, where you deliver and how you differ.

Neither. Start with the basics before any AI-specific work. Google says a page must be indexed and eligible to show with a snippet to appear as a supporting link, and the site must be included in Search generative AI features in Search Console, and OpenAI says a site must allow OAI-SearchBot to be eligible for ChatGPT search. Our guide to GEO for business covers that groundwork.

One thing not to do is chase mentions for their own sake. Google’s guidance says “seeking inauthentic ‘mentions’ across the web isn’t as helpful as it might seem”, because its AI features rely on the same ranking and spam systems as the rest of Search.

How should you report mentions and citations together?

Keep them on separate lines. For each priority prompt, a monthly sheet needs five columns:

1. Mentioned: yes or no, and how the answer describes you.
2. Own citation: which of your pages was cited, if any.
3. Third-party citation: which other domain was cited about you.
4. Found but not cited: where your tool shows it.
5. Platform, date and whether you were signed in.

Then put the platform data beside it: Search Console AI impressions by page, Bing citations by page with Citation Share for your main grounding queries, and GA4 sessions from AI referrers to the same pages.

Read the two lines together. When mentions rise and citations stay flat, start by checking your pages. When citations rise and mentions stay flat, check what your pages say about you. Model updates and changes to your prompt set can move either line too.

Semrush’s own release said marketers “cannot treat brand visibility in LLMs as a single score”. Its toolkit still offers a 0 to 100 AI Visibility score. Treat that, or any blended score, as a summary line rather than the report. For choosing the prompts, our guides to prompt research and measuring visibility in ChatGPT go into more detail.

What should a UK marketing team do this month?

1. Open the Generative AI performance report in Search Console and note which pages get AI impressions. That is your Google citation baseline.
2. Check AI Performance in Bing Webmaster Tools for the same pages, and note Citation Share for your main grounding queries.
3. Pick 20 questions your customers actually ask, run them in ChatGPT, Gemini and Google AI Mode, and sort each answer into the four positions above.
4. For every answer where you are mentioned but not cited, list the third-party pages that were cited and read what they say about you.
5. For every answer where you are cited but not mentioned, add the specific facts about your business to the cited page.
6. Next month, report the two numbers separately, with a note of any tracking problems such as the October ChatGPT tagging bug.

If you want help setting up the prompt log or reading the platform data, talk to AIWIZ about answer engine optimisation.