Why we measure at all

Star ratings alone can't tell you if a skill works. People rate things on day one, before they know. And ratings are easy to fake. So on Trilly, the main numbers come from what people actually do with a skill over weeks.

The three numbers that matter most

  • People using it this month. How many different people used the skill in the last 30 days.
  • Still use it a month later. Of the people who started a month or two ago, how many are still using it. This is the best sign a skill is worth keeping. We show it as a phrase, like "7 in 10 still use it a month later".
  • Does the job. How often a use ends well, not in a thumbs down or being given up.

We also show reviews, but only from people who added and used the skill, and each review says how long they used it.

What gets sent, and what never does

After a task, your AI sends Trilly one short note with the result. It never includes what you typed, what your AI wrote back, or your files. Feedback in your own words is only sent if you choose to send it. Our privacy policy has the details.

Keeping the numbers honest

  • Makers can't edit them. No maker can type in or change a number.
  • Changed copies don't count. Each note carries a fingerprint of the skill. If someone edited their copy, their notes are left out.
  • Odd bursts are ignored. Hundreds of notes a minute, or other impossible patterns, aren't counted.
  • Small numbers stay hidden. Until enough people use a skill, we say "not enough data yet" rather than show a number that could mislead.

The exact rules, thresholds and ranking maths are on how the numbers work.

What we can't measure

A skill you download is a set of files on your computer, and an AI may not send the note every time. Pasted copies (in a ChatGPT Project or a Gem) can't report at all. So the numbers show what we can see, and each one tells you how many copies it's based on.

See the numbers in action on any skill page, for example in skills for real estate agents.