Named versus cited: the two ways an AI answer can include you

Two different events that look the same on screen

When someone asks ChatGPT, Perplexity or Gemini for a recommendation, the answer can include you in two ways. It can name you: 'Brand.ch is a Swiss retailer known for…'. Or it can cite you: a numbered footnote or inline link to a page on your domain that the answer was built from. The first comes from what the model already knows, from training or from other sites that talk about you. The second comes from a live retrieval of your page. Both count as visibility, but only the second produces a click, and only the second is something you can move by changing your own site this week.

What has to happen for a citation

Answer engines that cite (ChatGPT search via OAI-SearchBot, Perplexity, Google AI Overviews and AI Mode, Claude with web search) run a retrieval step: they search, fetch a handful of candidate pages, extract the main content and pick passages. Your page has to clear every stage. If any fails, the answer may still name you from memory, but the link goes to whoever cleared them all.

1. Fetchable. robots.txt allows the search agent (not only the training crawler) and the WAF does not challenge it. See robots.txt for AI crawlers and bot protection.
2. Readable without a browser. The content is in the server response, not assembled by JavaScript. See client-side rendering.
3. Extractable. The main content sits in <main>, under one clear heading hierarchy, so the passage the engine quotes is yours and not the cookie banner.
4. Attributable. Title, description and canonical give the engine a clean label and a stable URL for the footnote.
5. Fresh. A dateModified or sitemap lastmod that reflects reality; for time-sensitive questions engines prefer dated sources.

These five are the access and legibility dimensions of the Qomvia score. That is not a coincidence: the score was built as the checklist for getting cited.

What has to happen for a mention

Being named is slower to move because it depends on the model's picture of your entity. That picture is assembled from many sources: your own site, Wikipedia and Wikidata if you are there, review platforms, news, directories, social profiles. Two things on your own domain still help. Organization JSON-LD with sameAs links to your profiles tells the model that all those mentions are one entity, and the model's fetched copy of your homepage (when it does fetch) is the most authoritative statement of what you are. Off-domain, consistent naming everywhere matters more than volume: a brand spelled three ways is three weak entities.

Mentions without citations are also a warning sign. Qomvia's AI monitor flags the pattern 'named but never linked': the models know you and send the click elsewhere, which almost always means one of the five citation stages above is failing on the page that should be the source.

Measure both, fix the one you control

The AI monitor asks the models the questions your customers ask and records, per answer, whether you were mentioned, whether you were linked, your position in any list and which rival was named instead. Read the two rates together. High mention, low citation: the site is the problem, run the readiness score. Low mention, low citation: the model does not associate you with the topic; that is a content and entity job, and citations of good topic pages are the fastest way to build it. High citation, low mention is rare and usually means the answer quotes you as a source without recommending you, which is fine.

Questions

Does a citation always mean a visit?

It means a link a reader can follow. Click-through varies by engine and by how prominent the footnote is, but without the link the click-through is zero.

Can I be cited without being named?

Yes. An answer can quote your page as a source for a fact without recommending your brand. It still sends readers and still tells the model your page is reliable on that topic.

How quickly can citations change?

As soon as the engine re-fetches the page, which for actively crawled sites is days, not months. Mentions from model memory shift slowly and mostly through third-party coverage.