The baseline measurement is the first systematic survey of the AI visibility of a brand, before measures begin. It records how often the brand is mentioned and cited in the answers to a fixed prompt set, how it is rated, which competitors and sources appear instead, and whether the model knowledge is correct. Everything measured later is measured against this reference.
How does a baseline measurement work?
It starts with the prompt set along the customer journey and a decision on the relevant systems: in the German market at least ChatGPT and Google AI Overviews, plus Gemini, Perplexity and Copilot depending on the target audience. Each prompt is asked several times in each system, because every answer is unique. From the answers, Mention Rate, Sentiment, Citation Rate and Share of AI Search relative to competitors are derived; the source analysis shows where the evidence comes from. This is supplemented by a query of the model knowledge without web search and a technical check of whether your own pages are retrievable for the systems.
The Digital Visibility Audit by CodaAI is such a baseline measurement: for company-specific buyer questions, it evaluates whether ChatGPT and Google AI Overviews mention a company, whom they recommend instead, and why.
Why does the baseline measurement matter for AI visibility?
Because realistic goal-setting is only possible with it. The meaningful goal in AI search reads: starting from a baseline measurement, increase the Share of AI Search relative to competitors for a set of proxy prompts. Without a baseline measurement there is no “starting from”, and no way to distinguish effect from chance.
What does this mean for your website?
Measure before you change. Keep the prompt set stable after the baseline measurement, document systems, repetitions and date, and repeat the measurement at fixed intervals. Compare trends, not single values. And derive the measures from the measurement: a low Mention Rate on brand-neutral questions calls for off-page work, a low Citation Rate on brand-specific questions calls for work on the website.