Baseline Measurement
A baseline measurement is the first systematic inventory of your AI visibility at a fixed starting point, before you take any action. It records whether and how often AI assistants like ChatGPT, Claude or Perplexity mention, recommend or cite your brand. These starting values serve as the reference against which you measurably compare later progress.
Why the baseline measurement matters
Without a defined starting point, you cannot judge whether your work on AI visibility is having an effect. The baseline measurement creates exactly this foundation: it captures how your brand stands in AI answers today, before you optimize. Only then do later changes become reliable. If the mention rate rises from 12 to 30 percent, that statement is only worth something if the 12 percent were cleanly documented. The baseline measurement also protects you from self-deception. Without it, people tend to overestimate successes or to interpret random fluctuations as progress. It is the honest baseline against which every claim about impact must prove itself, internally as well as toward clients.
How a baseline measurement works
First you define a fixed set of prompts, meaning test questions that your target audience would realistically ask an AI. For example: "Which tax advisors in Cologne are recommendable?" You put these questions to several AI assistants and record the answers. You capture whether your brand appears, in which position, in what tone, and whether a source is linked. It is important to keep conditions constant: the same prompts, the same models, the same period. Because AI answers fluctuate, you repeat each question several times and average the results. In the end you have a documented set of metrics such as mention rate, citation rate and position. Exactly these numbers form your reference value for all following measurements.
Common mistakes
The biggest mistake is a sample that is too small. Anyone who asks only once measures chance instead of reality, because AI models can answer differently to identical prompts. Equally risky: changing the prompts later. If you compare different questions in the second round, you are not measuring progress but two different things. The timing is also often underestimated. Models get updated, so you should note which model in which version you tested. Another classic is missing documentation: if you do not save the raw answers, you cannot later reconstruct what exactly changed. And finally, many forget to assess not just the plain mention but also the tone and the accuracy of the statement.
Relation to AI recommendations
The baseline measurement is the foundation of all work on AI recommendations. The goal of Generative Engine Optimization is for assistants to mention your brand more often, more accurately and more positively. Whether that succeeds is shown only by the before-and-after comparison, and the "before" is provided by the baseline measurement. It makes visible where you have blind spots: perhaps you are recommended well for one question but not mentioned at all for a related one. Such gaps can be closed in a targeted way. The baseline measurement is also central for communication with clients. It turns vague promises into verifiable goals: instead of "more visibility" it becomes "mention rate from 15 to 40 percent in six months".
Example
A bicycle shop in Leipzig wants to know how present it is in AI answers. Before any optimization, the owner puts ten typical questions to ChatGPT, Claude and Perplexity, such as "Where can I buy a cargo bike in Leipzig?". He asks each question five times and notes whether his shop is mentioned. Result of the baseline measurement: in 3 of 50 answers it appears, never with a link. This 6 percent mention rate is his starting value. Six months later he repeats exactly the same questions and sees in black and white whether his measures have worked.
Common questions
How often should I repeat a baseline measurement?
The baseline measurement itself is a one-off, it is your fixed starting point. After that, at regular intervals such as monthly or quarterly, you carry out follow-up measurements with exactly the same prompts and compare them against the starting value.
Is it enough to test just one AI assistant?
No. Different assistants like ChatGPT, Claude, Gemini or Perplexity draw on different sources and answer differently. For a realistic picture you should include several systems in your baseline measurement and evaluate them separately.