What Mirvel measures and how each metric is calculated

Everything starts with the prompts of each project, which are the real questions a potential customer would type into an AI assistant to discover, compare or choose a company like yours, and which are defined with a specific country and language so that the answer resembles the one a user in that market would receive. Each prompt has its own frequency, which can be daily, every two days, weekly, every two weeks or monthly, and an automatic scheduler checks every hour which ones are due.

When a prompt is due, Mirvel sends it to each of the models you have enabled in the project and stores the full answer exactly as the model returns it, together with the sources it cites when there are any. ChatGPT, Perplexity, Gemini and Google AI Overview answer with the help of a web search, whereas Claude is measured with its general knowledge and without real-time search, which is worth keeping in mind when you compare results across models.

Next, a language model reads each answer and extracts every brand that is named in it, not only the ones you track, recording the position of each one, whether it is your brand, a registered competitor or another company, what sentiment the text conveys about it and which sources are cited. The metrics in the dashboard are calculated from that structured data, and they can also be exported to CSV and PowerPoint so that you can take them to a meeting or into a client report.