How it works
A consistent AI check, explained in plain language.
We ask AI models defined questions, collect their written answers, and compare the answers using the same reference points each time. There is no human scoring panel. The result is evidence for a decision, not a claim that an AI model is always right.
-
1. Ask a defined question
Each check uses a consistent instruction. We give the model the brand, copy, buyer role, or business idea you submit and ask for a written response. The inputs and questions are saved with the audit.
-
2. Compare the answer with clear reference points
Semantic Similarity Rating (SSR) compares a response with reference points such as “not mentioned,” “mentioned beside alternatives,” or “clearly recommended.” This turns open-ended writing into a consistent comparison; it does not make the result a survey percentage.
-
3. Show the pattern, not just one number
The results show where responses cluster and where models disagree. An overall summary is shown only where it is useful. A new-offer check shows possible gaps rather than pretending customer-problem text is a rating scale.
-
4. Report failures honestly
If a model is unavailable, times out, or returns an unusable answer, the audit reports that problem. It does not invent a result to make the chart look complete.
How to use a result
The bars are similarity measurements, not statistically calibrated probabilities. “No human scoring panel” means people do not assign the result; it does not mean AI models are neutral, complete, or correct. Treat the output as a transparent starting point for investigation.
If you already use an AI-visibility monitoring tool, keep it. Perceptra OS answers a related earlier question: given your current brand and wording, how do several models interpret the business?