Start here
The free tool below runs entirely in this browser — no account, nothing stored or transmitted. Redact anything you would not want kept before you type it.
Open the companion tool →Why a written request works better than a settings tour
Memory features are spread across settings: a memory list, a personalization profile, a history toggle, and a training consent that may live elsewhere entirely. In-product deletion often removes one of those, which is why the request names all of them.
The request also produces a record. If a company later changes how it handles this data, your request and their reply are the documentation of what was true when you asked.
The protocol
- Look for the assistant's memory or personalization settings first and export whatever you can see. This tells you what exists before you ask.
- Send the request in writing through the app's support or privacy channel, naming each artifact: memory entries, personalization profile, chat history, and any model-training use of your content.
- Ask for the retention period and for confirmation once deletion is complete.
- Set a follow-up date and keep the confirmation. If nothing arrives, escalate to the company's privacy contact and note the dates.
- Decide separately whether to keep using memory features. Deleting the current profile does not stop it from being rebuilt unless you turn the feature off.
What the tool does
Build a written request covering everything an assistant stores about you, with the deletion steps and a follow-up date. It runs in this browser; nothing is sent or stored.
Limit first: This is a request-drafting aid, not legal advice. Response obligations and deadlines depend on where you live and on the company's own policy.
Open the companion tool →One human next step
If the company does not respond to a deletion request within a reasonable period, a consumer privacy agency where you live can take the complaint — several US states have dedicated privacy enforcement. Keep your request and their reply together.
Research log and safety checks
AI-assist path: not used for a verdict. This is a request builder: the artifact that matters is the letter the reader sends and the dated record of the reply, which no model can produce for them. §6.10: not a frontier-pace topic.
§0.5 protection result: category is data-privacy-exposure; crisis-adjacent: no; fear/urgency/scarcity toolkit used: no. The protective function is free and the paid feature is convenience only.