Custom Instructions

What is Custom Instructions?

Custom instructions are a personalization setting in AI chatbots such as ChatGPT that let you record preferences once and have them applied to every new conversation. They sit in the background, so the assistant reads them before answering and shapes its reply to match, without you retyping the same context each time.

Most tools split custom instructions into two fields. The first asks what the assistant should know about you: your job, your goals, the projects you work on, or the audience you write for. The second asks how you want it to respond: the tone, the length, the format, and any rules such as avoiding jargon or always giving a short summary first.

• Set once in settings, then applied to every new chat automatically.
• Field one: context about you, such as role, goals, and background.
• Field two: response style, such as tone, length, format, and rules.
• You can toggle them off for a single conversation when they do not fit.

Custom instructions vs memory

People often confuse custom instructions with memory, but they work in opposite directions. Custom instructions are rules you set on purpose. Memory is facts the assistant saves on its own as you chat, then reuses later. One is a configuration you control directly, the other is learned from your conversations over time.

A simple way to tell them apart: custom instructions suit permanent preferences that rarely change, such as always answering in plain English. Memory suits details that shift, such as the client you are working with this month or a project deadline you mentioned in passing.

• Custom instructions: rules you write, stable, and always applied.
• Memory: facts the AI saves during chats and recalls when relevant.
• You edit instructions in settings; you review or delete memories in a separate memory panel.
• Use instructions for how the AI behaves, memory for what it knows about your situation.

Custom instructions vs a system prompt

A system prompt is the developer-level instruction that sits underneath a product and defines the assistant's core behavior and guardrails. It is written by the company that builds the tool, not the everyday user, and it usually stays hidden. Custom instructions are the user-level version of the same idea: your personal layer on top of whatever the developer already set.

In practice the system prompt has priority. Your custom instructions personalize responses within the limits the developer defined, so they cannot override safety rules or the product's built-in behavior. Think of the system prompt as the house rules and custom instructions as your own preferences inside the house.

• System prompt: set by developers, defines core behavior, usually hidden.
• Custom instructions: set by you, personalize responses on top.
• System-level rules take priority over your personal preferences.

Custom instructions examples

Good custom instructions are specific and short. Vague requests like be helpful add little, while concrete rules change the output noticeably. The examples below show the kind of phrasing that works in each of the two fields.

• About you: I am a small business owner writing marketing emails for a non-technical audience.
• About you: I am a beginner learning Python and prefer explanations without heavy jargon.
• Response style: Keep answers under 200 words and start with a one-line summary.
• Response style: Use plain English, avoid corporate buzzwords, and give a numbered list when steps are involved.
• Response style: When I ask for code, add short comments and explain any tradeoffs.

How to set custom instructions

Custom instructions live in the settings or personalization area of most AI chat apps. You open settings, find the customization or custom instructions option, fill in the two fields, and save. The change applies going forward to new conversations rather than ones already open.

Each field has a character limit, so the wording has to stay tight. Historically the cap was around 1,500 characters per field, and some products later raised it for paid plans. Because limits and menu names change between apps and updates, check the current settings screen in your own tool rather than relying on a fixed number.

• Open settings, then the customize or personalization section.
• Fill in the two fields and save.
• Changes apply to new chats, not ones already open.
• Character limits vary by product and plan, so verify in the app.

The privacy angle

Whatever you type into custom instructions is stored by the chatbot provider and read on every request. Depending on your account settings and the provider, that content can also be used to help improve or train future models. Business and enterprise tiers commonly exclude customer content from training, and many consumer tiers offer a setting to opt out, but the default varies. For that reason, avoid putting passwords, financial account numbers, or sensitive personal details into these fields.

Custom instructions steer how a chatbot talks. They are not built to be a secure vault for the facts and files you want an assistant to reliably remember and retrieve. A dedicated memory that is private by architecture keeps that separate: per-user isolation, encryption at rest, and no training on user data. MemX takes this approach, giving your saved information its own private-by-architecture store rather than folding it into a chatbot's personalization box. This is privacy by design, not end-to-end encryption.

• Custom instructions are stored and read on every request.
• They may inform model training depending on the provider and your settings.
• Do not include passwords, financial data, or sensitive identifiers.
• For facts and files you want remembered, a separate private-by-architecture memory keeps data isolated and out of training.