Sources & Citations is Koalr's record of the pages each AI engine drew on to write its answer to one of your prompts — kept per answer and per engine, split into the pages you own and the pages you do not, and typed by what kind of site each is. Being named and being read are separate outcomes, recorded separately. Across 267,963 citations in Koalr's off-page research, 84% pointed at pages the mentioned brand did not own.
Your AI Visibility Score in under 5 minutes. No card required.
Sources recorded per answer, per engine, on every plan.
They come out of the same answer and are recorded in two separate steps from different parts of it, which is why the two counts never reconcile and are not supposed to.
One step reads the prose and records which brands were named. The other reads the list of pages the engine returned, resolves each to a domain, decides whether it is yours, and stores it against that answer. Because they draw on different parts of the answer, the record can show an answer that named you without reading you, and an answer that read you without naming you.
Named and cited is the outcome you want. Neither is the honest starting position. Named but not cited means the engine knows of you from somewhere else — worth finding out where. Cited but not named is the one people miss: your page did the work and the answer described the thing without saying who provides it. That last state usually points at a page that explains a topic well and never establishes whose page it is.
When an engine cites you without writing your name, there is no sentence about you to read, so the answer contributes nothing to your sentiment rather than a neutral middle. An answer that never described you should not be able to pull the measure of how you are described.
A list of the pages beating you is not useful until it is sorted by what you can do about them, because the two answers go to different people. Koalr makes that split on every cited page, so the same URL never gets one recommendation on one screen and the opposite on another.
Directories, marketplaces, review platforms, comparison sites, forums and social threads, industry bodies, government and academic sites, reference works like Wikipedia.
Somebody else controls the entry, which means the move is to be listed, corrected or updated rather than to publish anything. These are usually the fastest wins on the list, because the page is already being cited — it simply has the wrong information about you, or none.
Editorial articles, a competitor's own pages, anything published by someone with no reason to add you.
There is no listing to join. The only way to take the slot is to publish something an engine prefers for that question. Slower, and the compounding kind: a page that earns citations keeps earning them.
Alongside the split, each cited page is typed by what kind of site it is — a company's own site, editorial, a community or forum, a review platform, a reference or directory, an institution, a social platform. The type is what turns a list of domains into a pattern: a category whose answers come mostly out of forums is a different problem from one whose answers come out of trade press, and they are not fixed the same way.
An engine answers "best CRM for a small B2B team" and returns four pages behind its reply: an independent round-up post, a software directory's category page, a competitor's comparison page, and a forum thread from two years ago.
Fragments are stripped, the domain is worked out from the URL itself rather than from whatever label the engine attached, and each one is checked against your own domains. None of these four are yours.
The same URL cited by two engines is two records, and cited again in next week's run is another. That is the right shape for spotting a page the engines keep returning to, and the wrong number to read as “how many pages cite me”. For that, look at distinct domains.
Some engines return dozens of URLs behind a single reply. Koalr keeps a bounded set per answer, preferring one from each distinct domain before a second from any, so the record is biased towards the breadth of sources rather than the depth of any one. Treat it as a floor on volume and a fair read on variety.
Engines differ in what they expose. Some return only the pages that shaped the reply; others return everything they looked at. Where an engine does not separate the two, neither can Koalr, so a source list can include a page the answer glanced at and did not use.
Your page can be the source for an answer whose recommendation is somebody else. Citations tell you the engine trusts the page; the mention record — your AI Visibility Score, shown in-app as Prompt Visibility Rate — and the leaderboard tell you who it named.
Nothing here is a backlink metric and it should not be read next to one. A page can be cited constantly by AI engines with barely any links pointing at it, and the reverse is just as common.
Sources and citations are on every plan, including Starter, and recorded on all seven engines. What changes is how many questions you are collecting them from: 25 tracked prompts per site on Starter, 40 per site on Growth and 50 per site on Business. Full detail is on the pricing page.
A mention means an AI engine named your brand in its answer. A citation means the engine drew on one of your pages to write that answer. They are separate outcomes recorded separately, and you can have either without the other. Across 267,963 citations in Koalr's off-page research, 84% pointed at pages the brand did not own — so being talked about and being read are routinely two different things.
Every time Koalr runs one of your prompts, it stores the answer and the list of pages the engine returned alongside it. Each URL is resolved to a domain, matched against your own domains to mark it as yours or somebody else's, typed by what kind of site it is, and stored against that answer. Because the record is per answer and per engine, you can see that one engine built its reply from a directory listing while another built it from your product page.
Because they answer the question with whatever page answers it best, and for most commercial questions that is a round-up, a directory, a review platform or a forum thread rather than a vendor's own site. In Koalr's off-page pull of 267,963 citations, 84% pointed at pages the mentioned brand did not own. Those pages are not out of reach — most of them take listings or corrections — but you will only work on them if you know which ones they are.
Yes, and it is worth finding when it happens. The engine used your page to build its answer and then wrote around you — describing the thing without attributing it. The page is doing the work and the brand is getting none of the credit, which usually points at a page that explains a topic well but never establishes who is explaining it.
Koalr sorts every cited page into two routes. Pages where someone else controls the entry — directories, marketplaces, review platforms, comparison sites, forums and social threads, industry bodies, government and academic sites, reference works — are ones you can get listed on or corrected, and they are usually the fastest wins because the page is already being cited. Everything else has to be out-written: there is no listing to join, so the only way to take the slot is to publish something the engine prefers.
Not on its own. A citation count records citation events, not distinct pages: the same URL cited by two engines counts twice, and cited again next week counts again. It also says nothing about whether you were named in the answer. Read citations next to your AI Visibility Score — pages being read while your brand goes unnamed is a different problem from not being read at all.
Yes. Every engine exposes its sources differently and each is read on its own terms, so the record is per engine and per answer. This matters more than it sounds: the same question often produces very different source lists across engines, and a page that one engine leans on can be entirely absent from another's answer to the same question.
Scan your site free and get your AI Visibility Score, plus your first ranked fix list, in minutes. No card, no demo call.
One scan, 25 real prompts, 7 AI engines, under five minutes, zero setup.
Chase