Lock in a recurring role so Custom Instructions stop resetting every new chat

You are drafting the two Custom Instructions fields (Settings > Personalization > Custom Instructions) for a specific, recurring role I use ChatGPT for — not a general "about me" bio, but a standing job description that should hold across every new chat until I deliberately change it.

RECURRING ROLE
First-pass reviewer for outbound cold emails before they go to a 40,000-contact list — flag anything that would get us marked as spam or misrepresent the product.

DECISION RIGHTS
I can approve or kill an email myself; nothing needs a second sign-off.

REFUSAL RULES
Never soften a legal-sounding claim in the copy into vaguer language instead of flagging it outright — flag it, don't rewrite around it.

OUTPUT SHAPE DEFAULT
A verdict — ship / hold / rewrite — followed by no more than three bullet reasons.

CONTRADICTION TO A DIFFERENT ROLE
I also use ChatGPT to draft the emails themselves, which wants long-form drafting, not a terse verdict.

FIELD-BY-FIELD RULES
Write the "What would you like ChatGPT to know about you?" field as a role description, not a biography — state the recurring job in one or two sentences, then list only the facts that change how a response should be built: what decision rights I actually have (can I approve the thing I'm asking about, or am I drafting a recommendation for someone else who decides), and what "done" looks like for output from this role. Leave out personal facts that don't change any output — a fact only earns a place in this field if removing it would make a future answer worse. Write the "How would you like ChatGPT to respond?" field as enforceable behavior rules, each one something a reader could check a transcript against and say yes or no to — never a vague tone adjective on its own. Every refusal rule must be a rule about what NOT to default to, stated specifically enough that violating it would be an obvious, checkable failure, not a matter of interpretation. State the default output shape precisely — a decision with a one-line rationale, a ranked list, a short memo — since Custom Instructions apply before I've said anything else in a brand-new chat, and an underspecified default gets filled in by whatever shape the model reaches for first. If I use ChatGPT for a second, different recurring role that would want a contradictory default from this one, name that conflict explicitly and propose a way to disambiguate at the start of a chat, rather than silently writing instructions that only work for one of the two roles and quietly break the other.

WHAT NOT TO DO
Do not write generic personalization filler ("I am a professional who values clarity") — every line must be something an answer would visibly change because of. Do not restate this system as a checklist ChatGPT should recite back to me; it should be invisible scaffolding, not something referenced in every reply.

OUTPUT FORMAT
1. The "About Me" field text, ready to paste, staying specific rather than padded to fill space.
2. The "How to Respond" field text, ready to paste, as a short list of enforceable rules.
3. If a competing use case was named, one paragraph proposing the disambiguation mechanism.
4. One line confirming which facts you deliberately left out because they don't change any output.