Skip to main content
When an AI engine responds to a prompt, it produces three distinct types of brand signals. Understanding the difference is key to reading your visibility metrics correctly.

​
Mentions

A mention is when the AI names a brand in its response text, the prose it writes to the user.
In this example, “ING” and “DKB” are mentions. They count toward Share of Voice and Appearance Rate. What counts:
  • Brand names the AI recommends, compares, or discusses in its own words
  • A brand named several times in one answer is one brand being discussed, not several. Share of Voice and Appearance Rate count it once per answer, and how often it recurs is reported separately as evidence of how central it was to that answer
  • Variations like “C24” and “C24 Bank” are normalized to avoid double-counting the same brand
What does NOT count:
  • Brand names that only appear inside citation links or source references (e.g., a media outlet cited as a source)
  • Brand names inside URLs
  • Generic terms like “the bank” or “this provider”

​
Citations

A citation is when the AI explicitly links to a URL as a source in its response.
This is a citation of sparkasse.de. It counts toward Citation Share. Citations can only exist when the engine actually searched the live web before answering. See Grounded responses below. What counts:
  • URLs the AI includes as inline links or footnote references in its response
  • Matched by domain. If the AI links to any page on sparkasse.de, that counts for Sparkasse regardless of what the link text says
  • The anchor text (the clickable text) is tracked for context but doesn’t affect the metric
What does NOT count:
  • Brand names in the response text without a link. Those are mentions, not citations
  • URLs that appear only in the AI’s search results but not in the final response. Those are sources

​
Sources (Search Results)

A source is a URL that appeared in the AI engine’s retrieval pool, the pages it considered while generating its response. Not every source makes it into the final response. The AI retrieves many pages, then selects which ones to cite. What counts:
  • URLs the AI engine retrieved during its search, whether or not it cited them
  • Matched by domain
Why it matters:
  • Sources are the denominator in Selection Rate
  • If your domain appears in sources but not in citations, the AI is finding your content but choosing not to use it. That’s usually a signal that your content alignment needs work

​
Grounded responses

Not every AI answer involves a web search. Engines sometimes answer from their built-in knowledge and sometimes search the live web first. A response where the engine searched is called grounded. zued records this for every sampled response. Grounding matters because it determines your leverage:
  • Grounded answers are built from live web content. Your pages can be retrieved, cited, and quoted, so content changes can influence what the engine says.
  • Ungrounded answers come from the engine’s existing knowledge. No page is fetched, so content changes can’t move them in the short term.
Grounding is also why citations and grounding belong together: citations and sources only exist in grounded responses. The grounding rate is shown as context under Citation Share, and is available per page, per prompt, and for each individual response. A low grounding rate on a topic means the visibility ceiling for your content there is lower, regardless of how well it’s written.

​
How they connect

A brand can appear as all three simultaneously: mentioned by name in the text, cited with a link, and present in the search results. Each is tracked independently. Whether a response is grounded is a property of the response itself and provides the context for reading citations and sources.
A high number of source appearances with few citations means AI engines are finding your content but not selecting it. Focus on improving your Alignment Score to close that gap.