When an engine answers a prompt it often runs several web searches first. Predict searches gives an estimate of how many it will run for a prompt you have not tracked yet.
How to use it
- Open Sources and choose the Engine searches tab.
- In the Predict searches box, type a prompt.
- Choose Predict.
What you see
- About N searches per answer, with the figure for each engine beside it when more than one reported searches.
- A list of the searches the engines most often ran for similar prompts, each with the share of those answers that ran it.
- Hover the first line to see how many answers it comes from and which of your prompts it is based on.
How it is worked out
It looks through the prompts you already track for ones that share enough of their words with yours, ignoring common words such as "the" and "best". It keeps the five closest. Only prompts whose answers reported their searches count, and a prompt with exactly the same wording as yours is left out. The estimate is then the average and the most common searches in those answers.
No engine is asked and nothing is charged. The estimate comes from your own history, so it is only as good as the prompts it can find.
When there is nothing to go on
If no tracked prompt is close enough, the box says No similar prompt with reported searches yet. Where the pattern list is switched on, it then shows a rough guess instead: see Likely searches by pattern.
To start tracking a search an engine really ran, see Find new prompts to track.