ChatGPT Custom Instructions: What They Do and How to Use Them

ChatGPT custom instructions are settings that let you tell ChatGPT details and response preferences it should take into account across conversations. They are useful for setting recurring context—such as your role, preferred level of detail, or formatting—so you do not have to repeat it every time.

What “custom instructions” means

“Custom instructions” is an ordinary descriptive phrase, not a special slang term. In ChatGPT, it names a product feature for saving user-provided guidance that can shape responses. The phrase itself is made from the standard English words custom (adapted to a particular person or purpose) and instructions (directions about what to do); it did not originate as a standalone technical term.

For example, a user might write, “I teach middle school science; explain technical ideas in plain language and include a short example.” A sentence describing the feature could be: “I added my preferred writing style to ChatGPT’s custom instructions.” The feature is not a guarantee that every answer will follow every preference, and it is not a substitute for checking important information.

What the feature is for

Custom instructions are intended for information or preferences that apply to many chats, rather than details that matter only to one question. Depending on the interface and account settings, the feature may appear under personalization or a similarly named settings area. Labels and availability can change, so consult the current ChatGPT settings if a menu name differs.

Background: your occupation, skill level, or the audience you usually write for.
Response preferences: concise answers, plain language, examples, or a particular structure.
Recurring constraints: a preferred programming language or a request to state assumptions before solving a problem.

It is generally not the right place for a one-off request (“summarize this article”) or sensitive information that the system does not need. Put task-specific context in the message for that conversation instead.

How custom instructions work

When enabled, the saved guidance is made available to ChatGPT as context for conversations, so it can influence how a response is framed. It does not change the underlying model into a personally trained version, and it does not create a separate permanent rule system. The model still generates each answer from the current prompt and other available context, and may interpret instructions imperfectly or encounter conflicting directions.

A useful way to write instructions is to make them short, concrete, and easy to apply. For instance, “Use metric units by default. If a question involves code, use Python unless I specify another language. Explain unfamiliar terms briefly.” This is clearer than “Give good answers.” A practical set can cover:

Useful context: “I am learning introductory statistics.”
Desired response: “Define technical terms and show the calculation steps.”
Boundaries: “If information is uncertain, say so rather than guessing.”

Avoid overloading the instructions with a long list of demands. If two preferences conflict—for example, “always be brief” and “explain every detail”—ChatGPT may not know which one matters more. State priorities, such as “Be concise unless I ask for a detailed explanation.”

Setting, changing, and testing them

To use the feature, open ChatGPT’s settings and look for the custom-instructions or personalization controls. The exact route and wording can differ across web, mobile, account types, and product updates. Enter the guidance in the available fields, save it, and start a new prompt that gives it a fair test. If the interface offers separate questions about what ChatGPT should know and how it should respond, use the first for background and the second for style or behavior preferences.

To improve a result, revise the instruction rather than repeating it in every message. Test one change at a time: ask a question where your preferred tone, units, or level of detail should matter, then decide whether the response improved. You can edit or clear the saved text in settings. If the instruction is not being applied, check that it was saved and enabled, then make the wording more explicit. For a single conversation, you can also restate the preference in that chat.

Limitations, privacy, and common mistakes

Custom instructions guide responses; they do not guarantee accuracy, override all other system behavior, or ensure that ChatGPT will remember every detail perfectly. They also do not make a request appropriate merely because it is saved as a preference. Verify factual claims, especially when a decision depends on current or consequential information.

Do not treat the feature as private storage. Avoid entering passwords, payment details, government identification numbers, or confidential material. Review the product’s current privacy and data controls for how account information is handled.
Do not put temporary details there. A deadline or one project’s requirements will usually belong in that project’s conversation, not in guidance meant to carry across chats.
Do not use vague preferences alone. “Be helpful” gives little direction; “Use headings and keep the opening summary to two sentences” is more actionable.
Do not assume the setting changes memory. Custom instructions and memory are distinct personalization concepts. Their availability and behavior can vary, and saving an instruction does not mean ChatGPT has learned or permanently trained on it.

A related misconception is that custom instructions are a way to force ChatGPT to ignore its normal rules. They are user preferences, not a universal override. If a request conflicts with other requirements or the model’s capabilities, the saved preference may not be followed.

Custom instructions, memory, and chat prompts

These features can sound similar because each may provide context, but they serve different purposes. Custom instructions are explicit, reusable directions you choose to save. Memory, when available and enabled, can retain or use selected information from prior interactions according to its controls. A chat prompt is the message you send for a particular task and is the best place for one-time requirements or source material.

For example, a user can put “Use plain English” in custom instructions, ask “Help me prepare a presentation about photosynthesis” in a chat, and provide a particular audience or slide limit in that prompt. If you want to see or manage what has been remembered, check the relevant memory controls rather than assuming it is part of the custom-instructions text.

Frequently Asked Questions

Are ChatGPT custom instructions the same as memory?
No. Custom instructions are guidance you enter for reuse. Memory is a separate feature for retaining or using information across conversations where available. Check their controls independently.

Do custom instructions apply to every chat?
They are designed to influence responses across chats when enabled, but application can depend on the current product, settings, and conversation. They are guidance, not a guarantee.

Can I change or delete my custom instructions?
Yes. Open the relevant settings or personalization area to edit, disable, or clear them. Interface labels may change over time.

What should I write in custom instructions?
Include reusable context and clear preferences, such as your experience level, preferred units, tone, or answer format. Keep one-time task details in the chat prompt.

Can custom instructions make ChatGPT always answer accurately?
No. You can ask it to identify uncertainty or show reasoning steps, but instructions cannot guarantee correctness. Check important claims against reliable sources.

General steps for current mainstream versions; menus can differ by device and version. See the disclaimer or report an error.

More in this bed

The whole Tech How-To bed →