These fields steer the model that rewrites your transcript. The glossary protection, "never translate / never summarise" and the machine-readable output format are always enforced for you and can't be edited here — so a change can adjust style but won't break term protection or the app.

• Instruction — the core behaviour for this editing task. Describe what the model may and may not change. Phrase as direct rules ("Fix…", "Do not paraphrase…").
• Extra rules — your own additions on top of the default (e.g. "Always write numbers as digits", "Keep British spelling"). Leave the instruction default and just add here when you only want a tweak.
• It's a transcript editor, not a chatbot — don't ask it questions or request new content; it should only transform the words you spoke.
• Keep it concise and unambiguous. Conflicting or vague rules make output less predictable.
• Use Reset on any field to restore the shipped default if results get worse.