The half-life of a tool listing: rot in AI catalogs
Catalog entries decay on three clocks: the URL, the price, and the capability. The URL is the only one a script can check. Kevin Indig's Growth Memo analysis found 64 of 365 AI-cited URLs were broken, costing 203 citations, which is the same decay measured from the citing side.
By the The AI Tools Index team
Every fact in a tools catalog has an expiry date, and the three main ones expire at very different speeds. Understanding which is which tells you how much of any directory, including this one, you should trust without re-checking.
Three clocks, three speeds
The third clock is the one that hurts readers, and it is the reason this index does not publish capability claims it has not verified on the day.
- The URL. Slowest to change, easiest to verify, and the only one a script can settle without judgement.
- The price. Changes several times a year, with no notification anybody outside the company receives.
- The capability. Fastest and least visible: a feature is removed, a licence narrows, an export disappears, and the homepage looks identical.
How much decay is normal?
Kevin Indig's Growth Memo analysis of AI-cited pages found 64 of 365 cited URLs were broken, and estimated the cost at 203 citations. That is roughly one in six links dead in a corpus that had been deliberately selected as worth citing, which is a fair proxy for what happens to any list nobody re-checks.
The same analysis found primary research made up only 2.7% of AI-cited pages while earning about 3.3 times the citations per page, which is the flip side of the same coin: pages that are the source of something get cited, and dead pages get nothing.
What re-checking actually involves
- Resolve the URL and follow redirects. A redirect to a parent company homepage is a closure, not a move.
- Confirm the product still has its own sign-up path rather than becoming a feature of something else.
- Confirm the name, because a rename looks exactly like a healthy listing until somebody reads it.
- Confirm the category still fits, because products grow surfaces.
- Remove rather than archive anything that fails, so the live total stays true.
Why we do not publish a scraped pricing table
Because it would be wrong within weeks and confidently wrong the whole time. A price shown on a catalog page carries the catalog's authority while carrying none of the vendor's accountability, and the reader has no way to see how old it is. Linking to the vendor's own pricing page transfers both the accuracy and the accountability to the party that controls the number.
That is a real cost. A pricing table would be genuinely useful and it would be genuinely unreliable, and between those two we pick the one we can stand behind.
What a permanent URL is for
Pages here keep their addresses. A category page that has been at the same URL for years is worth more to a reader and to any system citing it than a fresher page at a new address, and moving a URL throws away everything accumulated against it. The counts on those pages change; the addresses do not.