How citations move your AI visibility score
Citations and mentions are different things, but one often leads to the other. How citations feed a visibility score, which kinds matter most, and how to tell whether a new citation actually moved anything.
Our visibility score counts names, not links. An answer that cites your page but recommends a competitor does not count. An answer that names you without linking anywhere does.
So why do we spend so much time on citations? Because for engines that search the web before answering, the names in the answer usually come from the pages the engine read. Citations are the inputs. The score is the output. If you want to move the output, you have to understand the inputs.
Four combinations, four meanings
Every collected answer falls into one of four groups for your brand.
Named and cited. The engine read your page and recommended you. This is the outcome you are working toward.
Named, not cited. The engine knows you, but its picture of you came from somewhere else: third-party pages, or what the model learned in training. Worth checking that the description is accurate, because you do not control the source.
Cited, not named. The engine used your page, perhaps for a definition or a statistic, and then recommended someone else. This is the most fixable group. The engine already trusts the page. It just is not finding the case for your product in it.
Neither. You are absent from both the answer and its sources. The work here is either a page that deserves citing or a presence on the outside pages that are cited.
To illustrate with invented numbers: across 200 answers, a brand might be named and cited in 15, named only in 25, cited only in 20 and absent from the remaining 140. Its visibility score is 40 out of 200, or 20%. But the 20 cited-only answers are a short list of pages where small edits could move the score directly.
Your pages versus everyone else's
Citations of your own domain matter, but on category prompts the most valuable citations are often third-party pages that name you. When an engine cites a roundup that includes you, the roundup does the recommending. When it cites a roundup that leaves you out, the same mechanism works against you, every time that page gets used.
That is why we look at the third-party domains cited on the prompts where a client is never named, and count how many prompts each one appears on. A single roundup cited across a dozen of your missing prompts is a better target than a dozen sites cited once each.
Not all citation types behave the same
It helps to group cited sources by type: editorial roundups, review and comparison sites, community threads such as Reddit, video such as YouTube, news coverage, and vendor pages, including your competitors'. Each type has a different route in.
You can request an accurate listing on a review site. You can pitch a publication. You can take part in a community thread openly, as a named representative of the company, within that community's rules. You can make a video that answers the question better than the one being cited. You cannot do much about a competitor's own page except publish a better answer. Knowing the type tells you which kind of work a prompt needs.
The chain behind every citation
A citation sits in the middle of a chain. A page exists. A crawler fetches it. An engine cites it. The answer names you. A buyer clicks or searches for you. Some of those buyers convert.
Each link can break, and each one is measurable. If a new page is never fetched by an AI crawler, it cannot be cited, and no amount of content polish will help until the access problem is fixed. If it is fetched but never cited, the page is probably not the best answer to the prompt. If it is cited but you are not named, look at what the cited passage actually says. If you are named but no visitors arrive, the prompt may be less commercially important than it looked.
Walking a prompt along this chain is the fastest way we know to find out what to do next.
How to tell if a citation moved the score
Say a roundup adds you, and over the following weeks it gets cited on six of your tracked prompts. This is an illustration, not a result. The question is whether being named rose on those six prompts specifically.
Look at those prompts before and after the citation first appeared, over at least two reporting periods. Compare them with prompts where the roundup was not cited. If naming rose on the six and stayed flat elsewhere, the citation is a plausible cause. If naming rose everywhere, something else moved, probably the engine. Even a clean pattern is association, not proof, and we report it that way.
Expect lag. A new citation does not change answers the next day, and the effect can be uneven across engines.
Citations that do not help
A few kinds of citation look good on a chart and do nothing for the score. Your page cited for a statistic in an answer that recommends competitors. An outdated page cited with old pricing or a retired product, which can actively hurt by giving buyers wrong information. A citation on a prompt no buyer asks. Citation counts on their own reward all three, which is why the score, not the citation count, is our headline.
Which citations to chase first
Prioritise sources that are cited across many of your prompts and on more than one engine, where your inclusion is realistic. An accurate listing on a review site you are missing from is often a quicker win than a hard editorial pitch. Fix outdated pages of your own that are being cited, since the engine already trusts them.
How we track citations in Promptwatch
Citation work for every client runs on Promptwatch. Citation analytics shows page-level and domain-level citations for each prompt, including Reddit and YouTube citations and offsite mentions, with a breakdown by citation type. Citation trends show how cited sources and volume shift over time, which is how we date when a roundup started being cited and line it up against the prompts it affected.
Agent Analytics covers the start of the chain. It reads real-time AI crawler logs, flags crawl errors, and shows the crawl-to-citation path for a page. Prompt tracking covers the naming step, and visitor analytics covers the end, with AI-referred visits and conversions. Seeing all of it in one place is what lets us say which link in the chain broke, rather than guessing.