Infinite willingness is a bug
Type a topic into a writing tool and you get an article. Type a terrible topic and you get an article of the same length and confidence. The tool has no mechanism for telling you that nobody wants this, because producing output is the only thing it knows how to do.
That is fine when the cost of a bad draft is zero. It is not zero. It is a slot on the calendar, an editor's afternoon, and a page on your site that will underperform quietly for years.
Putting a number in the middle
Our runs make six moves, and the third one produces a number. After the reading and the clustering, before any prose exists, the run measures how much of what it found circles the same problem, and publishes that figure along with the threads behind it.
Publishing it is uncomfortable in its own way. It means the product sometimes hands you a figure that argues against the thing you just asked for, in writing, before you have read a word of the draft.
What the number is actually worth
A low figure is not a failure. It is a finding. It says the audience is not asking this, at least not anywhere public, at least not yet. That is genuinely useful, and it arrives before you have committed anything rather than three weeks after publishing.
It also changes what you are willing to try. When a weak idea is cheap to check, you check more ideas, including the odd ones that turn out to have a crowd behind them.
A number you can argue with
A system that sounds equally confident about every topic gives you no information about any of them. The only reason to trust an answer is that a number sits underneath it, and that the number was free to come back low.
That is the whole argument for showing the working. Not transparency as a virtue, but transparency as the thing that makes a confident answer mean anything at all.
A tool that sounds equally sure about everything tells you nothing about anything.
