Every tool in this directory carries a score, and the lists are ordered by it. This is what goes into it, where each part comes from, and how much of the catalogue it actually reaches.
Most AI directories rank tools by a traffic estimate bought from a third party. We did too, and it was wrong often enough to be worth replacing. The figure is measured per domain, so any tool whose listed address belonged to someone else inherited their numbers — a Chrome extension was credited with the entire Chrome Web Store, and seven Media.io sub-products were each credited with the whole of media.io. Because the mistake always borrows a larger site's figure, it landed at the top: a quarter of our own first hundred results was another site's traffic.
So the score is built from sources that can be checked instead, and each tool's components are stored separately so a position can be explained after the fact rather than asserted.
How much traffic the tool's own site gets, taken from the Tranco list — a research ranking of the top million domains that averages several commercial toplists over thirty days specifically to damp the manipulation each of them carries alone. A rank is a measurement, so it wins. Where a tool has no rank we fall back to the catalogue's own estimate, but capped at a quarter of the scale: an unverifiable number cannot outrank a measured one, however large it claims to be.
How many Wikipedia articles cite the tool's domain. Wikipedia's sourcing rules mean each citation is an uninvolved person deciding the site was worth referencing in an encyclopaedia, which is hard to manufacture and impossible to buy. We tried domain age first and discarded it: deepseek.com was registered in 2000 and kagi.com in 1993, years before either product existed, so age measures who could afford a good domain rather than who has been around.
Points across every Hacker News story linking the domain. It is a narrow, developer-heavy sample and it is weighted accordingly — a video dubbing tool can be large and absent here. Posts below 10 points do not count at all, because the median result is a single submission with three points, which is what it looks like when someone posts their own link and nothing happens.
Whether readers here click through to the tool. This is the only first-party signal we have and currently the weakest: it needs more than 25 clicks before it counts, and almost nothing clears that yet, so it is close to dormant. It will carry real weight as the log fills. One click is not evidence, and a scale that treats it as evidence is worse than no scale.
We publish a band — “top 10,000 worldwide” — rather than a visit count, because the inputs do not support a visit count and printing one would be inventing precision. Where a tool's listed address is a Chrome Web Store page or a tracking link, no traffic figure is shown against it at all; the number recorded there is the host's. Where several listings share one domain, the figure goes to the listing that is that site — Media.io keeps it, its six sub-products do not — and where no listing is the site itself, none of them gets it. And a tool we have no outside signal for is left unranked rather than placed at the bottom, because those are not the same statement.
The gaps are the honest part of this page. Just under half the catalogue has no external signal of any kind — those tools are too small or too new to appear in a toplist, an encyclopaedia or a developer forum, and no amount of arithmetic changes that. They are listed, searchable and unranked.
The signals are collected and the scores rebuilt on the first of each month. Every reading is kept rather than overwritten, so a ranking can be recomputed against a past month and a weighting can be revised without re-collecting anything. The normalisers are fixed numbers, not whatever this month's largest value happened to be — otherwise every tool's score would shift whenever one new tool was added, and a ranking that silently rebases is not a ranking.
Found something ranked wrongly? That is usually a tool whose listed website points somewhere that is not its own — tell us which and we will correct the record. Meanwhile: tools by measured traffic, browse by category or search the index.