How we measure our own visibility
Forecast Record measures its own visibility with the same rule as everything else it publishes: once a week, 50 fixed questions — 25 that name the site and 25 general ones — go to seven search engines and AI assistants, twice each. Version 1.0 of this method has been in force since ; source: our own methodology record.
Figures as of
What we ask
Once a week, in a fixed window, the same set of 50 questions is put to every surface twice. Half of the questions name the site; the other half are questions a bettor would ask without knowing the site.
The general questions are checked for any hint of the site's name or domain before they are asked.
Where we ask
A surface is the tool asked, not the company behind it. There are seven:
- Google Search: the organic results and the AI Overview block
- Bing Search: the organic results
- two AI assistants with web search
- the Perplexity, Gemini and Qwen model interfaces
Each record names the country it applies to.
What we record for each answer
Four separate things:
- whether we were named
- whether a link to us was given
- whether the figure quoted about us matched ours on that date
- whether that figure is still current
Every raw answer and every address found is kept, so any figure we publish about our own visibility can be traced back to the answer it came from.
When a request fails
A failed request, such as a timeout, a rate limit or a refused search, is recorded as a failure with its reason, apart from answers that leave the site out. A surface without a successful request is reported as unmeasured.
Limits of the method
An answer from a surface that did not search is the model's own knowledge: an address of the site in such an answer is recorded as invented, never as a citation. Every raw answer and every address found is kept with its date.
The same data, machine-readable
The same text as this page, from the same snapshot.
Check us