Playbook · Research · Example output
Example output: a finished research brief¶
This is what the research playbook produces. A real question, a short answer, and sources you can click. Nothing here is invented — every claim links to the page it came from.
Example output
Everything below the line is the finished brief, exactly as the playbook delivers it. Sources checked July 21, 2026.
The question¶
Do AI chat tools use what I type to train their models?
Decision it supports: whether to change any settings before using AI chats for work. Written for a non-technical business owner.
Summary¶
It depends on the tool, your plan, and one setting. Both Claude and ChatGPT have a training setting you control on personal accounts, and both offer a private chat mode that is never used for training. Business plans are treated differently. The practical move: check the setting once, and use the private mode for anything sensitive.
Key findings¶
Claude (Free, Pro, Max personal plans). Anthropic says it uses your chats to improve its models if you allow it in your privacy settings, if a conversation is flagged for safety review, or if you explicitly join a testing program. Incognito chats are not used to improve Claude even when the setting is on. (Source: Anthropic Privacy Center, "Is my data used for model training?", dated March 16, 2026)
You can change the Claude setting any time. It's under Settings → Privacy → "Help Improve Claude," on desktop and mobile. Turning it off stops new chats from being used in future training. Anthropic notes that data already included in training that has started stays there, and safety-flagged conversations may still be used for its trust-and-safety work. (Source: Anthropic Privacy Center, "How do I change my model improvement privacy settings?", dated March 16, 2026)
ChatGPT (personal accounts). OpenAI says ChatGPT improves by training on people's conversations unless you opt out, which you can do through its privacy portal or the Data Controls settings. Temporary Chats don't appear in history, don't create memories, and aren't used for training. (Source: OpenAI Help Center, "How your data is used to improve model performance", accessed July 21, 2026)
Business plans differ from personal ones. OpenAI says it does not train on inputs or outputs from business products — ChatGPT Business, ChatGPT Enterprise, and the API — by default. (Same OpenAI source as above.)
Thumbs-up / thumbs-down feedback is its own channel. Both companies say that when you rate a response, the related conversation may be stored and used — with OpenAI, even if you opted out of training. (Both sources above.)
Risks and unknowns¶
- These pages describe current policy and can change. The Anthropic articles carry a March 2026 date; check the live pages before relying on them.
- This brief covers Claude and ChatGPT only. Other tools (Gemini, Copilot, and others) have their own policies — not researched here.
- "Not used for training" is not the same as "not stored." Retention periods are a separate question and weren't the focus of this brief.
What it means for the decision¶
Two minutes of setup: open the privacy settings in the tool you use, decide whether the training toggle should be on, and learn where the private chat mode is (Incognito in Claude, Temporary Chat in ChatGPT) for anything sensitive. After that, no ongoing effort.
Sources¶
- Is my data used for model training? — Anthropic Privacy Center (dated March 16, 2026)
- How do I change my model improvement privacy settings? — Anthropic Privacy Center (dated March 16, 2026)
- How your data is used to improve model performance — OpenAI Help Center (accessed July 21, 2026)
Want a brief like this?¶
The playbook interviews you, does the research, and ends with the same Before you trust it checks used here.