This translation has not been editorially reviewed yet. The German version is authoritative. Deutsche Fassung
GEO / AI
How often does an AI name my website as a source, and what does it depend on?
The citation rate is the share of AI responses to a fixed set of questions where your website is the source. It can only be estimated over many repetitions, because the same question brings a different answer almost every time. For Copilot and the AI summaries in Bing, Microsoft has been counting your website’s quotes since February 2026 itself. Why one page is cited explains studies with multiple approaches, each an observed connection. Google only mentions the basics of search as a prerequisite and admits no place in the answer.
The same question brings a different answer almost every time
Rand Fishkin of SparkToro, along with Patrick O’Donnell of Gumshoe.ai, had about 600 volunteers ask 12 questions for recommendations for brands and products to ChatGPT, Claude, and Google’s AI Overview, together 2,961 times, in November and December 2025. That ChatGPT or Google named the same list of brands in two responses was less common than once in a hundred runs, the same order about once in a thousand. On the other hand, how often a brand appears over dozens or hundreds of runs of the same question, the authors evaluate as a useful measure of which brands the AI connects with the topic. Tools that rank in the AI response consider them nonsense. The study counted brands in recommendations, non-linked sources. Who counts sources, therefore, also measures over many runs: the same questions at fixed intervals, separately per provider, and the number of answers in which your website stands as a source.
Bing counts the quotes, Google counts them to search traffic
Since February 10, 2026, Bing Webmaster Tools has been showing the report "AI Performance" as a public preview. It applies to Microsoft Copilot, the AI summaries in Bing and selected partner services. It gives the number of citations in the selected period, the number of your pages that appear as a source on average per day, the search terms with which the AI has retrieved content (grounding queries), and the citations per address. Microsoft says the number per page says how often it is quoted, and nothing about its importance or place in the answer. Google counts appearances in the AI overviews and in the AI mode in the Search Console to all search traffic, in the performance report under the search type "Web". A separate number for the AI functions is not mentioned in the documentation there. For providers without their own report, you count yourself, with the same questions at fixed intervals.
Approach: Who is at the forefront in the search, is quoted more often
For the AI overviews of Google, this is largely true: 76.1% of the pages cited there were in an evaluation of Ahrefs for the search query in the first ten hits. For the other assistants, the connection is weaker. Ahrefs sent 15,000 long searches to Google and Bing and the same questions to ChatGPT, Gemini, Copilot and Perplexity. Of the pages Perplexity cited, 28.6% were in the top ten Google hits for the question, with the other assistants about 8%; more than 80% of their citations came from pages that did not even stand for the question in Google search. The explanation of Ahrefs: The assistants search with several variations of the question and may never ask the original question. Google describes the same procedure for the AI overviews with additional searches on sub-topics. This leads to the approach of answering the partial questions of a topic on its own pages, instead of just following a single search term.
Approach: Sources, quotes and figures in the text
In the study, which coined the term Generative Engine Optimization, researchers led by Pranjal Aggarwal changed texts in nine ways. Source references, literal citations and figures increased the visibility of a source in the responses of generative search engines by up to 40%, depending on the field. was measured by a collection of search queries; For a single industry, the effect is thus open.
Approach: timeliness
Ahrefs evaluated almost 17 million quotes from ChatGPT, Perplexity, Gemini, Copilot and the AI overviews. The pages cited by AI assistants were on average 25.7% younger than the pages in the normal search results. In Google’s AI overviews, however, the first three cited pages were on average 16 days older than the search results. The author classifies this himself: On average, cited pages are 2.9 years old, topicality is one factor among many. From the mere change of date without new content advises John Mueller of Google, which cites the article on this.
Approach: Brand awareness on the web
With 75,000 brands, visibility in Ahrefs’ AI overviews was more closely related to mentions of the brand on the web than backlinks, explicitly as a context rather than a cause. This fits the interpretation of SparkToro: brands that occur over many runs of the same question over and over again, connects the AI more closely with the topic.
Approach: Questions, lists and short answers
Microsoft recommends for copilot and the AI answers in Bing a structure from which parts can be detached: questions with clear answers that an assistant can often take literally, lists, steps and comparison tables, headings that show where a thought ends, and answers in one or two sentences. A measured effect is not mentioned in the contribution; He writes himself that nothing guarantees the choice for an AI answer.
What the audit of euid.com measures
The GEO check area measures the requirements on your website: whether the robots.txt or a firewall locks out the crawlers GPTBot, OAI-SearchBot, ClaudeBot, Claude-SearchBot and PerplexityBot, whether the structured data is in the source code of the server, whether author and date are named and how the pages are structured: answer at the beginning, questions and answers, lists and tables. The SEO test area measures the basics of the search, from which the assistants get their sources. The citation rate itself is an observation over weeks and varies from answer to answer. Therefore, it does not count in any rule with points.
What is covered and what remains open
Every approach here comes from a measurement from the outside: Who measures, sees questions and answers, the selection inside the systems remains hidden. Most numbers come from Ahrefs, a provider of SEO tools, and apply to the time of the respective measurement. Google writes, AI overviews only appeared when they complement the classic search, and therefore often not at all; Indexation and display are not guaranteed. A promise about a certain citation rate can therefore not be proven.
- Sign your website into Bing Webmaster Tools and open the AI Performance report. It says how often Copilot quotes your pages and with which search terms.
- Write down the questions your customers ask before making a purchase, and ask each one several times to ChatGPT, Copilot, and Google Search. Count how many answers your website has as a source.
- Repeat the count each month with the same questions and compare the shares, not individual answers.
- In the Search Console, check whether the pages to be cited are indexed and are found on the sub-questions of your topic.
- Add numbers with source to these pages and revise them if the content changes.
- Do you distrust information such as "place 2 in ChatGPT". The order changes almost with each answer.
- Never change the date of a page. The pages cited by AI assistants are also almost three years old on average.
- Count for each provider separately. In the Ahrefs study, Perplexity cited pages from the first ten Google hits significantly more often than ChatGPT, Gemini or Copilot.




