How we rank

We count people, not posts

A skill moves up when more different people say they want to get better at it. One person posting ten times counts once, which is most of what keeps bots and pile-ons from steering the board.

We look for what people want to learn or improve (“how do I get better at…”, “I want to learn…”, “stuck on…”), not for everything that mentions a topic. News and adverts about a skill don’t count.

Where it comes from

Search: what people type into Google, YouTube and TikTok search when they set out to learn something (“how to get better at…”), and Google Trends where available. Attention: how often the skill’s Wikipedia article is read on phones (desktop counts are left out because crawlers inflate them). Discussion: new questions about each skill on Stack Exchange, how busy each skill’s own subreddit is, public posts on Reddit, Bluesky and Hacker News, and how many posts on X say “learn…”, “get better at…” or “improve my…” about each skill. Video: the views on new “how to” videos on YouTube and on fresh TikTok posts about each skill.

Each source is scored on its own scale first, because a busy platform shouldn’t look like a popular skill.

Rising means rising for that skill

Growth is measured against the skill’s own usual level over the previous weeks, not against other skills. Small numbers are smoothed and treated with caution, so going from 2 posts to 6 doesn’t count as a confident +200%. A single day far above a skill’s recent level is capped, so one crawler or one viral post can’t carry it. The same post copied or cross-posted counts once, one person counts for at most three skills a day, and if most of a skill’s posts come from a single community, that source is left out for the skill.

Some skills rise every year at the same time: fitness in January, study skills in September. Once we have a year of history, that seasonal rise is taken out, and those skills are labelled Seasonal rather than Exploding.

One ranking from many sources

The sources are combined by rank (reciprocal rank fusion), so no single source can carry a skill to the top on its own. Where a category has two or more sources with history, a skill needs signal from at least two of them, and a minimum number of people, before it can be ranked. While a category’s other sources are still building that history (the first weeks, or areas a source doesn’t cover), one is enough, and the board is marked “Early data”.

We rank on a cautious estimate of growth. The board doesn’t reshuffle on small changes: a skill joins at rank 20 or better and stays until it drops below 30, and a move smaller than the uncertainty in its rank is shown as “–”, not as a move.

Privacy

Only totals are published. We don’t keep the text of posts or anyone’s name or handle (a scrambled code that changes every month is enough to count each person once), never quote individuals, and only show head-counts in broad buckets (“1k+ people”). Skills the system finds on its own appear only after a person has reviewed them, and names that touch health conditions, religion, sexuality or addiction are never put forward automatically.

Sample data

Until the first editions are scored, the board shows a sample edition, marked “Sample data”. Its numbers are illustrative and are not real rankings.