How the answers are sampled
Each week, each tracked question is put once to each assistant, as a buyer in your country, through the app a buyer uses where that is possible. The answer is kept whole, with the pages it used, so any reading of it can be checked against the text. Whether you were named is read from your name, the names you gave, and your website.
Why one week is noisy
Ask the same assistant the same question twice and the names can change. So one answer, or one week, moving by a name or two is mostly chance. That is also why a campaign runs once a week and not more often: a second run in the same week mostly shows chance.
- Read the trend across weeks, not one week against the last.
- Read counts across many questions (named in so many of all the answers), not a single answer.
- Compare like with like: the same questions, the same places and the same competitors. Changing the competitors restarts the comparison, and the progress page marks it.
- Google ranks move less than AI answers, but a page that just changed is tested by Google for about three weeks; see after a page goes live.
Accuracy of the pages Oftheard writes
A page draft may state only what your confirmed fact sheet and your own site support. A figure nothing supports is flagged, and a second reading checks every price, number, place and claim against the facts before you see the draft. Anything only you know is left marked for you to fill in. See the checks every draft passes.
What is not built
Oftheard does not yet check the facts an AI assistant states about you (a wrong price, say) against your fact sheet; the answers are kept whole so you can read what they said. It also does not ask the same question several times in one week to average the answers.