How does a routine choose what to write about?

The pool it may choose from

A routine only ever looks at tracked prompts, and only those its filters allow. You can narrow the pool by window, tag, or persona, or point the routine at a saved report and it borrows that report's filters live. Inside the pool, a prompt qualifies only when it has at least five AI answers in the window, so the pick rests on a pattern across answers rather than on one response. Prompts this routine already covered inside the window are left alone until the window rolls over; rejecting an article frees its prompt sooner.

The three strategies

• Lowest visibility takes the prompt with the lowest mention rate. It always has a candidate.
• Competitor gap looks for the prompt where competitors are mentioned more often than you are, and skips the run when you are ahead everywhere.
• Mentioned late is for the prompts you do appear in, but near the end of the answer, and takes the lowest position score.

The scores are the same numbers your reports use, so the prompt a routine picks is the one a report would show as weakest under the same cut. When two prompts tie, the one with more competitor mentions wins, and after that the one with more search volume.

When nothing qualifies

A skipped run is normal, and the history says why, for example "No prompt had competitors ahead of your brand." A new workspace produces nothing until tracking has run long enough to give prompts five answers. When a run does pick a prompt, the researcher starts from the last three AI answers for it, the ten pages and domains they cite most, and the competitors they name.