What happens to your text in AI writing tools: training and retention settings compared

Training and retention are different questions

Two things are easy to blur. Training is whether your text is used to improve the vendor's models. Retention is how long the vendor stores it, and who can read it while it is stored (for example, to investigate abuse). A tool can promise not to train on your text and still keep it for 30 days. Both answers sit in the vendor's help pages and can change, so the dated sources below are what I read on 23 October 2026. Check the live page before you rely on any line.

Personal accounts

Service (personal plan)	Used to train by default?	Where to change it	Retention and review notes
ChatGPT	Yes, unless you turn off "Improve the model for everyone"	Settings, then Data Controls	Temporary Chats are deleted within 30 days and reviewed only when needed to monitor for abuse
Claude (Free, Pro)	Not unless you report the material, for example through feedback, or opt in	Nothing to switch off by default on the page I read; check your account privacy settings for any opt-in	Conversations flagged under the usage policy may be analysed to improve safety systems; feedback conversations are stored separately
Gemini Apps	Used to improve services while Keep Activity is on	Gemini Apps Activity, Keep Activity	Auto-delete defaults to 18 months; chats seen by human reviewers are kept for up to 3 years, disconnected from your account

The sources, in the same order. OpenAI's Data Controls FAQ says the toggle applies to the account, not a device, and that turning it off stops new conversations being used for training. Anthropic's consumer training page states it will not use inputs or outputs to train Claude unless you have reported the materials or explicitly opted in, and explains the feedback and safety exceptions. Google's Gemini Apps privacy page says that with Keep Activity on, activity is used to provide, develop and improve services, including training generative AI models, and that with it off or in a temporary chat, chats are held with your account for 72 hours and not used to train models unless you send feedback.

Two details matter for writers. First, turning a setting off is not retroactive: OpenAI's FAQ speaks about new conversations. Second, deleting a chat does not always delete every copy: Google says chats already reviewed by human reviewers are retained for up to three years.

Business, API and workplace plans

Service (work plan)	Training default	Retention note
ChatGPT Business, Enterprise, Edu and the API	No, not by default	Workspace admins control retention; deleted conversations are removed within 30 days unless law requires otherwise; the API may keep inputs and outputs up to 30 days for abuse checks
Claude for Work and the Anthropic API	No, not by default	Feedback you submit is the stated exception
Microsoft Copilot (Microsoft 365)	Prompts, responses and Graph data are not used to train foundation models	Interaction history is stored; users can delete it, admins can set retention policy

OpenAI's enterprise privacy page gives the 30 day windows, says a Data Processing Addendum is available for GDPR support, and notes that its staff and bound contractors may access stored conversations for engineering support, abuse investigation and legal compliance. Anthropic's commercial training page says commercial inputs and outputs are not used for training by default. Microsoft's Copilot privacy documentation states the training position, the stored activity history and the admin retention controls.

A short checklist before you paste anything

1. Find out which plan you are actually on. The same logo can sit on a personal account and a workplace account with opposite defaults.
2. Look up the training setting and switch it off if you do not want your text used, then check it again after major app updates.
3. Look up how long chats are kept and who can read them, not only whether they train models.
4. Strip names, client details and anything covered by an NDA before pasting, whatever the plan says. Our thread on using AI writing tools with client data under GDPR covers the legal side, and the thread on whether ChatGPT trains on what you paste walks through the ChatGPT toggle step by step.
5. At work, ask for the written policy first. Members have shared what theirs covered in this thread on company AI policies.

For help picking a tool with these constraints in mind, start with the guide on what to check before you sign up.