What is the difference between a mention and a citation?

A mention is the name of a brand within the text of the answer, whereas a citation is a link to a page that the model points to as a source, and the two are independent, because a model can name your brand without linking to your website and can link to your website without naming your brand.

Why does Claude show no citations?

Because Mirvel measures Claude without access to real-time web search, which is the way to see how it talks about your brand using only what it knows, and without search there are no consulted pages to cite, so the number of citations is zero by design and not because of an error.

What should I do if many citations come from directories?

It means that the models use those directories as a regular source for your industry, so keeping your profile complete, up to date and well described in them is a direct way of influencing what is said about your brand, and you can check in the weekly evolution whether the change shows up in your metrics.

Citations and sources in AI: where the models get their answers

Updated on October 4, 2026

Citations are the web pages that artificial intelligence models link to or mention as a source when they put together an answer, and Mirvel collects every one of them, groups them by domain and shows you which sites feed the answers in your industry, including how often your own domain appears and which of your sections are cited, so that you know where it pays to have a presence for the models to talk about your brand.

How many sources each model cites in a real project

In this example project the models cited 10,376 sources from 1,326 different domains, with very little concentration, and while Perplexity and Gemini cite sources in all of their answers, Claude cites none because it is measured without web search.

sources per answeranswers with sources

Perplexity16.5100%

Gemini13.9100%

AI Overview6.368%

ChatGPT1.861%

Claude00%

Real data from a production project: 1,503 responses analyzed between August 6 and October 4. Brand names have been replaced with generic labels.

Which models cite sources and which don't?

ChatGPT, Perplexity, Gemini and Google AI Overview answer with the help of a web search and can therefore cite the pages they consulted, although they do so with very different intensity: in the example project Perplexity includes about 16.5 sources per answer and Gemini about 13.9, both in 100% of their answers, Google AI Overview about 6.3 in 68% of them and ChatGPT about 1.8 in 61%.

Claude is measured with its general knowledge and without web search, so it returns no citations and it is entirely normal to see zero sources in its answers, a fact that Mirvel states explicitly in the detail of each run so that it is not mistaken for a failure.

What does the dashboard show?

The dashboard shows the most cited domains with their percentage of the total, the proportion of citations that come from industry directories and aggregators, which are an obvious opportunity to optimize your profile, and a breakdown of the citations of your own domain by folder, which lets you see which sections of your website the models actually take as a reference.

In the case of Gemini, the links it returns are Google redirects that expire after a few days, so Mirvel recovers the real domain of the source from the title that accompanies each link, and that way the data from all models can be compared on equal terms.

How should you read source concentration?

If a few sources accumulate a good part of the citations, you have a short list of sites where it is worth getting a presence, whereas if the citations are widely spread, as in the example project, where the most cited domain barely reaches 4% and it takes 84 domains to reach half of the citations, the strategy has to rely more on the overall quality of your presence than on a single site.

It is also worth watching the share of your own domain, which in the example is 3.7%, because it indicates how far the models turn to your website as a source compared with third parties, and because a rarely cited own domain often signals that the content answering your customers' questions is not proving useful or accessible to the models.

Frequently asked questions

What is the difference between a mention and a citation?: A mention is the name of a brand within the text of the answer, whereas a citation is a link to a page that the model points to as a source, and the two are independent, because a model can name your brand without linking to your website and can link to your website without naming your brand.
Why does Claude show no citations?: Because Mirvel measures Claude without access to real-time web search, which is the way to see how it talks about your brand using only what it knows, and without search there are no consulted pages to cite, so the number of citations is zero by design and not because of an error.
What should I do if many citations come from directories?: It means that the models use those directories as a regular source for your industry, so keeping your profile complete, up to date and well described in them is a direct way of influencing what is said about your brand, and you can check in the weekly evolution whether the change shows up in your metrics.