Skip to content

Research brief · Worked example

Research with evidence you can check

Here is a short brief built around a specific question. The sources are real; the reader and decision are sample inputs.

Sample research task

Question: "Can we stop personal Claude and ChatGPT conversations from being used for model training?" Decision: which settings to check before using either tool for work. Audience: a small business owner. Official sources checked October 5, 2026.

Short answer

Both services offer controls over model training on personal accounts. Check the setting for the actual account you use, and read the exceptions. Training controls alone do not establish whether a tool is suitable for confidential work.

Key findings

Claude personal plans. Anthropic lists user permission, explicit testing opt-ins, and safety review as routes for using conversation data. Incognito chats are excluded from model improvement. Submitting feedback can allow training use of the related conversation. This article covers personal plans and their Claude Code sessions; commercial products have separate policies. Anthropic's model training policy

Claude's setting. In Settings → Privacy, Anthropic labels the toggle "Help Improve our AI models." Turning it off stops new chats and coding sessions from being used for future training. It also stops previously stored sessions from being used in future training runs, but does not reverse training already started. Safety review has separate rules. Anthropic's settings guide

ChatGPT personal accounts. Turn off "Improve the model for everyone" under Settings → Data controls. OpenAI says new conversations then will not be used for training. They can still remain in history, and choosing to submit feedback may permit use of the associated conversation. Personal Codex tasks follow this control, with an additional, separate "Include environments" setting for environment data. OpenAI's data-use policy, Data controls

ChatGPT Temporary Chat. A chat is excluded from model improvement while it remains temporary. Saving it converts it to a regular chat, where the account settings apply. Temporary chats can use existing memories when personalized, and a copy may be retained for up to 30 days for safety. Temporary Chat documentation

Risks and unknowns

  • A training opt-out does not settle storage, retention, access, or your obligations to clients and employers.
  • This brief covers personal accounts. It does not establish the terms of a particular business workspace or connected service.
  • The pages and settings can change. The date checked tells you when this brief was verified.
  • No account was inspected. The settings currently enabled for the sample reader are unknown.

What this means for the decision

Our recommendation is to inspect the plan and controls on the account being used, then check whether the intended material is allowed there. For an initial exercise, use public information or fictional notes.

That recommendation is our interpretation of the policy limits. It is separate from the providers' descriptions of their controls.

Sources

All five pages were checked October 5, 2026.

What to check in your own brief

A working link is the first check. Open it, find the relevant passage, and ask whether its limits change the conclusion. Use the Before you trust it checklist.

Try the research playbook