Notice: This page requires JavaScript to function properly.
Please enable JavaScript in your browser settings or update your browser.
Learn Making AI Remember Your Preferences | Advanced Prompting and Knowing the Limits
Prompt Engineering for Work

bookMaking AI Remember Your Preferences

Swipe to show menu

System prompts apply within a specific tool configuration or session. Custom instructions are the consumer-facing equivalent — a set of persistent preferences you configure once in your AI tool of choice, and that apply automatically to every conversation you start.

If you use ChatGPT, Claude, or Gemini regularly, setting up custom instructions is one of the highest-return, lowest-effort things you can do. It takes ten minutes to configure and saves you from re-explaining yourself in every new chat.

How Custom Instructions Work

Custom instructions are a standing configuration you set in your account settings — separate from any individual conversation. Once set, the model reads them before every new chat and uses them to calibrate its behavior.

Most tools structure custom instructions around two core questions:

1. What should the AI know about you? Your role, your industry, your typical tasks, the kind of content you produce, who your audience usually is.

2. How should the AI respond? Your preferred tone, your formatting preferences, your default output length, and any standing rules about what to include or avoid.

You don't need to answer both exhaustively. Even a few well-chosen sentences in each field will meaningfully improve the consistency of your outputs.

Screenshot description: A settings panel interface — styled generically to represent any major AI tool's custom instructions screen. Two labeled text fields are visible. Field 1, labeled "What should the AI know about you?" contains filled-in example text: "I'm a product manager at a fintech startup focused on small business lending. I work primarily with engineers, designers, and senior leadership. My tasks include writing product specs, stakeholder updates, and customer-facing release notes." Field 2, labeled "How should the AI respond?" contains: "Keep responses concise and direct. Use bullet points for anything with more than three items. Default to plain language — avoid jargon unless I'm clearly writing something technical. Never use corporate buzzwords. When I ask for a draft, give me one version, not multiple options unless I ask." A small annotation below both fields reads: "Set once. Applies to every conversation."

Writing Effective Custom Instructions

The most useful custom instructions are specific rather than general. Compare these two versions:

Too general: I like clear and concise writing.

Specific and actionable: Keep all responses under 200 words unless I ask for a longer output. Use bullet points for any list with more than three items. Default to plain language — no business jargon unless the context is clearly technical.

The model can act on the second version. The first is too vague to change behavior in a meaningful way.

What your custom instructions should cover:

  • Your role and the type of work you do most often;
  • Your typical audience — who your outputs are usually for;
  • Your preferred default format — prose, bullets, tables;
  • Your preferred default length — short, medium, or a word count;
  • Standing tone preferences — formal, direct, conversational;
  • Anything you consistently have to ask AI not to do — generic openers, unnecessary caveats, repeated suggestions to "consult a professional."

Keeping Custom Instructions Current

Custom instructions should reflect how you actually work right now — not how you worked six months ago. They're worth revisiting when:

  • Your role, team, or focus area changes significantly;
  • You notice you're regularly asking for the same adjustment in every conversation — that adjustment belongs in your custom instructions, not in every prompt;
  • A new AI tool you're using has its own custom instructions field with different options than what you've configured elsewhere;
  • Your writing style preferences have evolved based on what you've learned about what works.

Treat custom instructions as a living document — not a one-time configuration you set and forget.

question mark

Which statement best describes how custom instructions work in AI tools?

Select the correct answer

Everything was clear?

How can we improve it?

Thanks for your feedback!

Section 4. Chapter 2

Ask AI

expand

Ask AI

ChatGPT

Ask anything or try one of the suggested questions to begin our chat

Section 4. Chapter 2
some-alt