About
We measure what AI says about companies, and where it got the idea.
Capibera is a measurement practice, not a platform. There is no dashboard to log into and no seat to buy. You get a report, written by people, with every number traceable to the answer that produced it.
Every model has an opinion about your company. None of them asked you.
Four engines. Twenty questions. Live web search on.
Engines
ChatGPT, Claude, Gemini and Perplexity. The four your buyers actually open, tested in the same session window.
Questions
Written for your category, weighted to buying intent: who leads, what to pick, what goes wrong, what it costs.
Answers logged
Each answer stored verbatim with its citations, timestamped, so the finding can be re-checked later.
Signals scored
Mention rate, position, sentiment, accuracy, competitive share, citation breadth, source authority, entity strength.
Why live search matters
With search disabled, a model answers from training data that may be a year old. With it enabled, it answers from what it can fetch right now, and it tells you which pages it used. That citation trail is the whole point: it turns an opinion into something you can act on.
Why the same 20 questions
Model outputs vary between runs. A fixed question set, fielded the same way each time, is what makes the 45-day comparison meaningful rather than anecdotal. We publish the question set inside the report so you can re-run it yourself.
What this measurement can and cannot tell you.
It is a sample, not a census
Twenty questions across four engines is 80 answers. That is enough to establish pattern and direction, and not enough to claim a precise market share of AI attention. We report ranges where the data is thin and say so plainly.
Engines are non-deterministic
Ask the same question twice and the wording changes. Mention and citation behaviour is far more stable than phrasing, which is why we score those and not language.
Benchmarks are category-relative
A median Signal Score in a crowded software category is not comparable to one in professional services. Every score you receive is benchmarked against companies fielded in the same category.
Nobody controls the models
Any vendor promising a guaranteed placement in an AI answer is selling something they cannot deliver. What can be changed is the material the models draw on, and whether the answer moves is then a matter of record.
Estimates are labelled as estimates
Expected-lift figures in the action plan are modelled from comparable engagements. They appear in the report marked "est." and never in the score.
What we are not.
A monitoring dashboard
Continuous tracking is useful once you know what to fix. First you need the diagnosis.
An SEO retainer
Most of the work this audit points to happens off your own site, and much of it is free to do.
A prompt-optimisation service
You cannot write a prompt on your buyer's behalf. You can change what the model finds when they write theirs.