AI tools that shut down: keeping a catalog honest
Products in this market close on schedules, not silently. OpenAI set Sora's web and app shutdown for 26 April 2026 and its API for 24 September 2026, and PlayHT closed in December 2025 after Meta acquired PlayAI in July 2025. A catalog that never removes anything slowly becomes a museum.
By the The AI Tools Index team
The unglamorous half of running a tools index is deletion. Every catalog is judged on what it adds and quietly ruined by what it fails to remove, because a reader who clicks two dead entries stops trusting the other two hundred.
What closures look like in practice
Rarely sudden, usually scheduled, and almost never announced where a catalog would notice. OpenAI's discontinuation notice for Sora set an end date of 26 April 2026 for the web and app experience and 24 September 2026 for the API, giving a long window in which a stale listing looks perfectly healthy while pointing at something with a published expiry.
Acquisition closures follow a different shape. Meta acquired PlayAI in July 2025 and the PlayHT platform shut down in December 2025, which is the standard pattern: the team continues, the product does not, and a catalog matching on company name rather than product URL misses it entirely.
The third kind: the product survives, the capability does not
Harder to detect and more damaging to a reader. Udio remains available, and downloads were disabled following settlements with Universal Music Group in October 2025 and with Warner. Nothing in a link check would flag that. The URL resolves, the homepage loads, and the single fact a reader came for, whether they can keep the output, has changed.
This is the category of change that argues against ever describing a tool's capabilities from a prior session's notes.
What this index does about it
- URL re-checks: a listing whose URL stops resolving leaves the live total rather than sitting in it.
- Removal, not archival, from the headline count. A closed product does not keep inflating a number.
- Renames handled as edits to one row, so a relaunch never manufactures a second entry.
- No capability claims copied forward. If a detail is not verified on the day, it is not printed.
Why removals should be visible rather than quiet
Because the removal is information. A reader evaluating a category benefits from knowing that two products in it closed this year, and a category with a high closure rate deserves a more cautious purchase than one without. Deleting silently keeps the catalog tidy and throws away the most useful signal it generates.
What a reader should do with all this
Two habits. Check the date on any tool recommendation before acting on it, including ours. And confirm the one capability your decision rests on, on the vendor's own site, on the day you decide. Both take under a minute and they cover the failure mode that no catalog can fully solve.