Click delete on a chatbot conversation and it disappears from view at once, which makes it easy to assume the data is simply gone. It usually is not, at least not right away. For most major assistants, deleting a chat starts a process rather than finishing one, and the gap between what the button seems to promise and what actually happens is wide enough to be worth understanding.
What deletion usually means
Take OpenAI's standard practice as the common case. When you delete a conversation, it is removed from your history and then purged from the company's systems within about 30 days, unless it has been retained for legal reasons or security. That 30-day window is typical across the industry. Copies can also persist longer in backups and in the abuse-monitoring pipelines that providers run to catch misuse. So deletion is real, but it is a countdown, not a vanishing act.
This sits next to a related confusion we have written about before, the belief that a model remembers everything you type. The two myths point in opposite directions and are both wrong. The model does not quietly memorize your chats, and deleting them does not instantly scrub them from the company's servers.
When a court freezes the delete button
The clearest illustration came out of the copyright fight between The New York Times and OpenAI. In May 2025 a federal magistrate ordered OpenAI to preserve all user logs, including conversations people had explicitly deleted, so they could not be destroyed while the case proceeded. OpenAI objected on privacy grounds, and the broad preservation order was lifted in September 2025, letting the company return to its normal deletion schedule for most users.
The episode is the proof, not the exception. Your deletion preference is a setting the provider honors by policy, and a policy can be overridden by a subpoena or a judge. For the months that order stood, deleting a ChatGPT conversation did almost nothing, because a court had told the company to keep it.
How to think about it
None of this means the assistants are lying when they offer a delete option. It means the option does what data deletion has always done in cloud software: it removes your access quickly and clears the data on a schedule, subject to backups and the law. The honest mental model is simple. Treat anything you type into a chatbot as recoverable for a while after you delete it, and as potentially preservable for longer if lawyers get involved. If a conversation would be a problem in a courtroom or a breach notice, the safest move is not to delete it later. It is not to type it in the first place.
Sources
- i. openai.com
- ii. thehill.com
- iii. www.yahoo.com
Commentarii · 0