Methodology
Our methodology
Out in the open: how our analysis works, where its limits are and what we deliberately do not do.
How we measure
- AVC (AI Visible Crawler): our crawler checks the technical AI readability of your website: robots file, structured data (JSON-LD), LLM file and whether the content is readable for AI crawlers.
- Querying multiple AI models: we ask several AI models fixed questions, in multiple runs and languages. In the Insight package that means 12 AI models, 20 customer questions with two runs each, a third when the answers differ, in one of ten languages.
- Evaluation: tonality (how positively, neutrally or negatively the models speak about you), a fact check of their statements and a competitor comparison showing who gets named instead of you.
Limits
AI answers fluctuate: the same question can be answered differently depending on timing, model version and wording. That is why we query in multiple runs. Our results are still snapshots, not permanent states.
What we do not do
- No guarantee of being mentioned or ranked in AI answers
- No paid mentions
- No fake reviews
- No manipulation of AI models
- No spam
- No hidden text
Privacy
We only analyse publicly accessible content. More in our privacy policy.