GEO monitoring is one of the three areas of GEO, alongside On-Page GEO and Off-Page GEO. It shows where a brand stands in AI answers and how that develops while work goes on in the other two areas. It needs a method of its own because AI answers are only generated when a question is asked: what ChatGPT, Gemini, Claude, or Perplexity say about an offering only becomes visible when you ask them. A provider of time-tracking software, for example, regularly asks questions such as “Which time-tracking software works well for contractors?” and analyzes whether, how, and next to which competitors its brand appears in the answers.

What GEO monitoring covers

Which metrics a monitoring setup collects depends on its goals and tools. Common ones are:

  • Mentions: whether answers name the brand—that is, contain a mention—expressed, for example, as a mention rate.
  • Citations: whether answers give pages of the brand’s own website as a source, expressed, for example, as a citation rate.
  • Competition: the brand’s share of all mentions or citations compared with its competitors, its AI share of voice.
  • Accuracy: whether the information about the brand is correct, for example on its offering, prices, and locations (brand accuracy).
  • Visits: how many visits reach the website through links in AI answers (AI referral traffic).

Only together do these figures give a picture of AI visibility.

How GEO monitoring works

The foundation is a prompt set: a fixed collection of questions based on what customers actually want to know, plus a list of the competitors the brand compares itself with. In prompt tracking, these questions are put to the selected AI systems automatically and repeatedly, each time with the same settings, for example for language and region. The answers are stored and analyzed for the metrics above.

Which AI systems to include depends on which ones the target audience uses. Because AI systems answer differently, it makes sense to report results separately for each system. GEOLYX analyzes ChatGPT, Gemini, Claude, and Perplexity in its GEO Monitoring.

Why only trends are reliable

A single query is a snapshot, because AI systems do not answer the same question the same way every time (answer variability). In an April 2026 preprint, a study released ahead of peer review, researchers at the University of St. Gallen conclude that assessing a brand’s visibility requires repeated measurement; the study’s first author is also affiliated with a company that sells AI visibility measurement. Visibility should therefore be described not as a single value but as a picture drawn from many answers: how often the brand comes up and how much that varies.

On top of that, AI systems and the web keep changing: providers update their models, and competitors publish new content. A single jump in the figures can therefore have causes that have nothing to do with a brand’s own work. Only a trend that holds over several months is meaningful.

That trend stays comparable only as long as the prompt set and settings stay the same and every change is documented. There is no standard method: a survey of 45 GEO studies, also released as a preprint in July 2026, finds that terminology and metrics are inconsistent. The word “visibility” alone refers to at least nine different quantities in the research literature. A transparent monitoring setup therefore discloses what it measures and how.

Reports from Google and Microsoft as a complement

Since 2026, Google and Microsoft have also provided website owners with first-party figures. These are based not on selected questions but on what was actually shown to users:

  • Google Search Console’s generative AI performance report has been available for all websites since August 31, 2026. It shows how often links to a site appeared in AI Overviews and AI Mode in Google Search, broken down by page, country, device, and date, among other dimensions.
  • The AI Performance report in Bing Webmaster Tools, which Microsoft introduced as a public preview on February 10, 2026, shows how often a website’s pages were cited as sources in Microsoft Copilot, in AI-generated summaries in Bing, and in select partner integrations. It is based on a sample, and according to Microsoft, it is designed for tracking trends over time.

Reporting progress

It can take time for measures to show up in AI answers, for example until an AI system has fetched changed pages again. It therefore helps if a report shows not only metrics but also what has already been implemented, each with a date—for example, “March: llms.txt created” or “April: structured data for product pages in progress.” That way, the state of the work is clear even while the answers remain unchanged, and later changes can be matched to what was done when.

Such a report stays clear when it shows a few metrics month by month, which also makes it easy for management to follow.