8 · Documentation
Sources and website
What the AI reads from, what separates named from cited, and what the domain check tests about your own website.
The search channels answer from pages they read at the moment of the question. Whoever writes those pages shapes the answer. The sources view shows which domains those are; the website view shows whether your own website can be among them.
Named and cited
A source can play two roles in an answer:
- Named (retrieved): the search considered the page relevant and presented it to the model. It appears in the answer's source list.
- Cited: the model used the page in the text, linked from a sentence or with a footnote. The stronger form.
Whoever adds both together measures findability and calls it citation. augenmerk counts both separately and names the stronger form first. This yields three situations per page and question:
| Finding | Meaning |
|---|---|
| not retrieved | not findable for this question |
| retrieved, not cited | findable, but not convincing |
| cited | used in the text |
Not every system discloses the distinction. Where citations and retrievals are exactly equal, the system simply does not provide it; the sources view marks such channels with an asterisk rather than claiming 100 %.
How we count
Every mention of an address in an answer is counted. If the same domain appears three times in one answer, it counts three times; the ranking of domains measures presence in the text, not the number of answers. Next to it stands the number of answers in which the domain appears.
Google's redirect address appears in Gemini answers instead of the real address. It would be the most frequent domain in the list without saying anything about the brand, and is therefore excluded from the ranking.
Sentence and source
A sentence is assigned to a source only where the answer itself makes the link, for instance with numbered footnotes. Where that is missing, the sources view shows the mention without a sentence rather than guessing one. The sentence for the footnote is highlighted in the answer excerpt, so you can see what the AI draws on a source for.
Sources only behind competitors
The overview and the sources view highlight domains that are cited when the AI names competitors, but never for your own brand. These are the places where a recommendation forms without the brand being involved: comparison portals, tests, press. If the brand is missing there, that site is the first to contact.
Care with interpretation: this is a link, not proof. A brand can be in first place and almost never be cited itself, because test magazines write about it. There are two routes: being a source yourself or being the subject of a source. The second has more effect; the first is the one you control yourself.
Your own website
For your own domain augenmerk counts how often it is named and cited in search answers, and separates home page from subpages. The difference is the diagnosis: whoever appears only with the home page is being looked up, not read. Pages that answer a question get cited.
The domain check
With every run augenmerk checks your own website:
- Reachability: if the site rejects automated requests, no AI can read it. The only finding that shows a cause rather than an effect.
- robots.txt against 35 AI bots: GPTBot, OAI-SearchBot, ClaudeBot, PerplexityBot, Google-Extended and others. Reported per bot: may read, partially, blocked. Explicitly named bots are a decision, not a default, and therefore notable in both directions. Administrative paths such as login or checkout do not count as content blocks.
- llms.txt: present or not — for information. By its own account, Google Search does not evaluate the file; whether other agents read it is open. We derive no action from it.
- Home page: title, description, heading, number of internal links. With more than 60 links before the content, an AI reading from the top is left with little content.
- Lighthouse mobile: scores for performance, accessibility, best practices, SEO and agentic browsing.
- Google standing: keywords, visits, referring domains and their monthly history, from third-party monthly search data — estimated. Plus the domains that share your keywords on Google; whether those are portals or the tracked competitors, the view says per project.
We read robots.txt and llms.txt ourselves and identify ourselves by our own name when doing so.
Search Console
Whoever connects their Google Search Console replaces the estimate with Google's own count: impressions and clicks over time, the strongest pages and the most-searched queries. How to set up the connection is explained under The application. Search Console knows only your own domain; competitors still come from third-party search data.
The gap view
A good Google ranking does not mean the AI uses the page as a source. The gap view puts both side by side: pages that rank on Google for a search query — with Search Console connected, your strongest pages by real impressions — and whether any AI cites them. Pages with demand and without a citation are the most concrete task list there is; start with the page with the most searches. The search queries the AI itself runs complete the picture: whoever ranks up front for them on Google gets read.
What we do not check
We do not check structured data or the sitemap today and therefore claim nothing about them. A classification of sources by type, such as testing body, press, comparison portal, community or retail, so far covers only part of the domains and is shown only once it is reliable.
Updated: 2026-09-16