Why AI tools disappear in the months after launch
A large share of launched AI tools stop being reachable within months, and almost nobody measures it. This post explains why it happens, why broken URLs cost more than they used to, and how a launch leaderboard is set up to count survival from its own rows rather than guess at it.
By the Launchelion team
Every launch archive turns into a graveyard eventually. Domains lapse, side projects get abandoned, acquisitions close products, and free tiers that were never costed get switched off. What is strange is how little anyone measures it, given that a launch feed's whole premise is a record of what shipped.
The three ways a launched tool goes away
The third category is the reason a survival measurement has to be done by checking, not by reading announcements. Abandonment does not produce a press release.
- It is switched off deliberately. OpenAI's help center notice on the Sora discontinuation gives 2026-04-26 for the end of the web and app experience and 2026-09-24 for API access.
- It is acquired and folded in. PlayHT's platform shut down on 2025-12-31, after Meta acquired PlayAI on 2025-07-11.
- It is quietly abandoned: the domain still resolves for a while, the login stops working, and nothing is ever announced. This is by far the most common case and the only one nobody reports.
Why a dead URL costs more than it used to
Kevin Indig's audit of AI citation data, published on Growth Memo, found that 64 of 365 cited URLs were already broken, costing an estimated 203 citations. When answer engines assemble responses from pages they retrieved earlier, a URL that stops resolving does not just lose its own traffic; it removes a source that was being quoted on your behalf.
For a maker, that is an argument for URL stability that has nothing to do with rankings. Moving your product page, your changelog or your pricing page breaks the references other people built to you.
What keeps a tool alive past month three
Cost control and a reason to keep going. AI products have a running cost per request that most side projects never modeled, and the abandonment cliff often arrives the month the free tier bill exceeds what the maker is willing to donate.
The other half is human. Tools with one paying customer survive at a noticeably different rate from tools with zero, not because the revenue is meaningful but because somebody is waiting for it to work. Getting to first revenue quickly is a survival tactic before it is a business one.
Both halves argue for the same launch-week decision: charge something, however small, and put a real limit on the free tier. A maker who does that has a bill they can cover and a reason to keep answering support mail long after the launch traffic stopped.
How this site intends to count it
The honest way to answer how many launched tools survive is to check the ones you recorded yourself: take the listings with a known launch date, re-check reachability at fixed intervals, and publish the share still resolving alongside the method, the sample and the limitations.
No number appears in this post because that check has to produce it, not the other way around. When there is a figure, it will arrive with the date it was measured and the definition of reachable used, at a URL that does not move.