docAnalyzer

Set persistent instructions

Tell docAnalyzer how you like to work once, applied to every chat in scope.

If you find yourself starting every chat with "answer in bullet points" or "I'm a tax lawyer, frame answers accordingly", set those as custom instructions once and they'll apply to every chat in scope.

Where to set them

Account settings → Instructions. The editor takes a free-form prompt: anything you'd otherwise paste into the first turn of a chat.

Things people commonly set:

  • Tone: "Answer concisely, no preamble, no recap of the question."
  • Role: "I'm an M&A lawyer; assume legal context unless I say otherwise."
  • Format preferences: "Default to short bulleted lists. Use tables when comparing more than three items."
  • Constraints: "Never paraphrase quoted contract language; quote it verbatim."
  • Domain shorthand: "When I say 'the MSA' I mean the master services agreement template."

Personal vs. per-workspace scope

Instructions default to personal scope: they apply to every chat in your personal workspaces. If you work across very different domains (a legal workspace and a marketing one, say), narrow the scope per workspace from the same editor. Workspace-specific instructions override personal ones for that workspace.

Team-shared workspaces have their own instruction layer that team admins can set.

How they work

Your instructions are applied to every chat in scope. They act as strong guidance, not a hard rule. For critical constraints, it can help to repeat them in the chat itself.

This is not a model setting and doesn't depend on which model you've picked. The instructions apply to every model in the catalog the same way.

What instructions are good for (and what they aren't)

Good fit:

  • Persistent style preferences you'd otherwise repeat each chat.
  • Background context that frames how you want answers.
  • Light formatting and tone control.
  • Shorthand vocabulary you use across many chats.

Less good fit:

  • One-off direction for a specific question (just put it in the chat).
  • Hard guarantees about behavior: instructions shape what the chat is told, but can't enforce anything end-to-end. Treat them as strong guidance, not a contract.
  • Bypassing model safety guidelines (they won't).

What's next

Was this helpful?