Best practice L1 · Basic prompts informational

Set Custom Instructions Once, Stop Repeating Yourself

What it is

Custom instructions are two standing fields — who you are and how you want answers — that ChatGPT applies to every new chat automatically.

set once, applies everywhereWho you are + how to answer3Custom instructions (always on)2Every new chat1
Custom instructions are a standing frame under every chat.

Why it works

Without them you re-state your preferences ('be concise', 'I'm a developer', 'British English') at the start of every conversation, or you forget and get generic output. Setting them once shifts the default toward the answers you actually want, permanently.

When to use it

Immediately, for anyone who uses ChatGPT regularly. The higher the volume of chats, the more this compounds.

When not to use it

Don't encode task-specific detail here — this is for durable, cross-topic preferences. Project-specific context belongs in a Project; one-off needs belong in the prompt.

Prompt

"What should ChatGPT know about you":
- Role and expertise level: <e.g. senior backend engineer>
- Domains you work in: <e.g. Go, Postgres, infra>

"How should ChatGPT respond":
- Be direct; skip preamble and disclaimers.
- Show reasoning only when I ask.
- Use British English and metric units.
- When code is involved, give the smallest correct example first.

Example

A designer sets 'explain technical topics without assuming I code' — from then on, answers about APIs and databases come in plain language by default, with no per-chat reminder.

Advanced version

Keep the 'how to respond' field tight and let heavier context live in Projects and custom GPTs. Custom instructions are the global default; the other features are scoped overrides. Layering them cleanly avoids contradictory guidance.

Common mistakes

  • Leaving them blank and re-typing the same preferences in every chat forever.
  • Overloading them with one project's details, which then bleeds into unrelated conversations.
  • Writing wishy-washy instructions ('be smart') that don't actually change behaviour.

Related