How measurement works

One run a day

Once a day, every active question in your project is put to every assistant the project tracks. Plans measure up to 5 assistants — ChatGPT, Gemini, Perplexity, Google AI Mode and Google AI Overviews: any 3 of them on Starter and Growth (you pick them per project), all 5 on Scale and Agency.

The number of queries a run makes is simply questions multiplied by assistants. Fifty questions on 3 assistants is 150 answers a day, which is why both numbers are capped by your plan.

Where the answers come from

Every question is put to each assistant the way a person in your market would ask it, in the country and language you choose for the project — any of 49 countries and 42 languages, on every plan. Web search is on, so each answer comes with the sources the assistant leaned on that day.

We record the answer and its sources exactly as the assistant returned them; nothing in them is written or edited by us. We do not disclose the technical details of how answers are collected, or the suppliers we collect through.

What we read out of an answer

Each answer is read for four things:

1. Which brands are named. Including yours, your competitors, and names you have not heard of — new competitors are discovered from the answers themselves, not from a list you maintain.
2. In what order. Position is where your brand sits among the brands named in that answer.
3. In what tone. Positive, neutral or negative, judged on the sentence about you and not on the mood of the whole answer.
4. Which sources it cited. Domains and pages the assistant leaned on.

What does not count

Failed queries never enter the denominator. If an assistant is down or rate-limits us, that answer is excluded, not counted as an absence. Otherwise an outage at one provider would look exactly like your brand collapsing — and it would look that way on the one chart executives read.

Why the same question moves

Assistants compose an answer each time, so results wobble day to day even when nothing about your brand changed. A single day is noise. Read the weekly direction, and use the per-question view to see whether a move came from one question or from all of them.

Beta: Bing AI

Some projects also track Bing AI - the AI-written answer Microsoft's Bing shows above its search results (often called Copilot Search). It is not the Copilot chat app. During the beta we turn it on for selected projects; it is not part of any plan and costs nothing extra.

Three things work differently:

1. Not every question gets an AI answer. For many questions Bing shows only ordinary results. That counts as "no AI answer", not as a failed query, and the Answers screen shows the share of questions that got one.
2. It stays out of "All engines". Your headline numbers and the sources list are calculated without it, so turning the beta on does not move figures you already track. Pick Bing AI in the engine filter to see its own numbers.
3. The market matters. Bing answers for the country and language set on the project.

If you would like it on your project, ask us in the chat.