ChatGPT Custom Instructions: How to Set Them, With Examples
Where the setting lives in 2026, the character limits by plan, how it works with personality and memory, and six examples.
ChatGPT custom instructions are standing rules you write once and ChatGPT applies to every conversation: who you are, how you want answers, what to stop doing. Unlike memory, you write them yourself instead of ChatGPT inferring them from your chats.
Below: where the setting is now, the character limits, how it fits with personality and memory, why ChatGPT sometimes ignores it, and six examples you can paste.
What are ChatGPT custom instructions?
Custom instructions are a text field in ChatGPT's settings whose contents ChatGPT considers in every response, on every plan and on web, desktop, iOS and Android (OpenAI Help Center). You get one active set at a time, and it applies until you change, disable or delete it.
They differ from memory in who writes them. ChatGPT memory builds itself: a 2026 study of 2,050 real memory entries from 80 users found that ChatGPT created 96% of them on its own (Dash et al., 2026). Custom instructions only say what you typed, and they only change when you edit them. OpenAI's own advice is to use them for stable preferences such as role, tone and format, and the chat prompt for the task at hand (OpenAI Academy, 2026).
How do you set custom instructions in ChatGPT?
Open Personalization, switch on Enable customization, and type in the Custom Instructions field.
- Web and desktop: click your profile, open Settings > Personalization.
- iOS and Android: open Settings > Customize ChatGPT.
- Make sure Enable customization is on.
- Write your instructions in the Custom Instructions field and save.
To turn them off, switch off Enable customization in the same place; deleting the text is optional. Edits apply to new conversations. Older guides describe two boxes, "What would you like ChatGPT to know about you?" and "How would you like ChatGPT to respond?"; OpenAI's current help page describes a single Custom Instructions field, so put both kinds of rules there.
What is the character limit for ChatGPT custom instructions?
The limit is 1,500 characters on Free and Go, and 5,000 characters on Plus, Pro, Business, Enterprise and Education. OpenAI raised the paid limit from 1,500 on July 15, 2026 (ChatGPT release notes, 2026).
More room doesn't mean you should fill it. A long set of instructions is harder to follow consistently than ten clear lines, and the rule you care about most shouldn't sit in paragraph six. Put it at the top. Each example below fits under 1,500 characters, so they work on any plan.
How do custom instructions work with personality and memory?
Custom instructions work alongside the personality you pick, your characteristics settings and your saved memories, and all of them shape the same response.
- Personality: the Base style and tone dropdown on the Personalization page sets a preset such as Professional, Friendly or Candid, and OpenAI says it "works alongside any saved memories or custom instructions you've set" (OpenAI Help Center). Pick the closest preset, then use instructions for what it doesn't cover.
- Characteristics: separate More or Less settings for Warm, Enthusiastic, Headers & Lists and Emojis (OpenAI Help Center). If you write "no bullet points" but Headers & Lists is set to More, you've sent ChatGPT mixed signals.
- Memory: custom instructions are one of the sources memory draws on, and the Sources icon under a reply shows when they were used. They're a separate setting from saved memories, so they don't count toward a full memory.
- Projects: project instructions apply only inside that project and override your global custom instructions (OpenAI Help Center).
- Temporary chats: a personalized temporary chat uses your custom instructions; a non-personalized one ignores them.
What should you put in custom instructions?
Write what's true in every chat, in four parts: who you are, what you're working on, how to answer, and what to avoid. Rules beat adjectives. "Be concise" means something different to everyone; "Answer in under 150 words unless I ask for more" means one thing.
- Who you are: role, field, level of expertise. This sets how technical ChatGPT gets.
- Current work: the one or two projects that come up most.
- Format: length, structure, when to use bullet points, code style.
- What to avoid: filler phrases, unrequested advice, tools you don't use.
Leave out passwords, client names you'd rather not share, and anything that changes weekly. Instructions are included in your data export, can be passed to third-party plugins you use, and may be used to improve models unless you've opted out of training.
Custom instructions examples
These six sets are starting points. Swap in your own details, and keep only the lines that change how ChatGPT answers you.
Software developer
I'm a senior backend developer. Main stack: TypeScript, Node.js, PostgreSQL, deployed on AWS.
- Give code first, explanation after, and keep explanations short.
- Use TypeScript unless I ask for another language. Strict types, no `any`.
- When there are several approaches, name the trade-off in one line each, then recommend one.
- Don't add comments that restate the code.
- If my question is ambiguous, ask one clarifying question before writing code.
- Flag security issues even when I didn't ask.
Founder
I'm the co-founder of a 10-person B2B SaaS startup. I handle sales, fundraising and hiring.
- Lead with the answer or recommendation, then the reasoning.
- Keep responses concise: under 200 words unless I ask for detail.
- When I share a plan, point out the weakest assumption first.
- Use numbers and named examples, not generic advice.
- For emails and messages, write in plain, direct English. No "I hope this finds you well".
- Don't suggest hiring an agency or consultant unless I ask.
Marketer
I lead content marketing for a developer tool. Audience: software engineers who dislike marketing language.
- Write in a direct, peer-to-peer voice. No hype words like "best-in-class" or "take it to the next level".
- Avoid filler phrases and rhetorical questions.
- When I ask for headlines or hooks, give 5 options with different angles, not 5 rewordings.
- Use short paragraphs. Use bullet points only for lists of 3+ items.
- If a claim needs a source, say so instead of inventing a statistic.
Student
I'm a second-year economics student. I use ChatGPT to understand concepts and check my reasoning, not to write my assignments.
- Explain new concepts with one concrete example before the formal definition.
- Break complex problems into manageable steps and let me try each step before you show the answer.
- When I'm wrong, tell me where my reasoning broke, not just the right answer.
- Cite the textbook-level source or author for theories you mention.
- Use British English.
Data analyst
I'm a data analyst. Tools: SQL (BigQuery), Python with pandas, Looker.
- Write SQL in BigQuery Standard SQL, with CTEs instead of nested subqueries.
- State your assumptions about the data before the query.
- For statistics questions, say which test you'd use and why, and mention the main way it could mislead.
- Keep charts advice to one recommended chart type, with the reason.
- Never invent column names: if you don't know the schema, ask.
Universal "no fluff" set
- Answer first. No preamble, no restating my question, no closing summary.
- Keep responses concise and scannable: short paragraphs, bullet points only when they help.
- If you don't know or aren't sure, say so plainly.
- Don't apologize, and don't add disclaimers unless something is risky.
- Provide multiple perspectives only when I ask for them.
- When I ask you to suggest improvements, give the three that matter most.
Why does ChatGPT ignore custom instructions?
ChatGPT usually ignores custom instructions because something else is overriding them or the instruction is too vague to follow. Check these first:
- Customization is off. Enable customization has to be on.
- You're in a project. Project instructions override your global ones.
- You're in a non-personalized temporary chat, which doesn't use them.
- Your settings conflict. A Quirky personality or high Enthusiastic setting pulls against "be terse".
- The rule is vague or buried. Rewrite "be professional" as a specific behavior, and move it to the top.
Instructions are guidance the model weighs, not hard constraints. If one rule keeps failing, restate it in the chat; if it matters for one kind of work, give that work its own project.
Custom instructions in ChatGPT vs Claude
Both apps have the same idea under different names. In Claude, standing rules go in the Instructions for Claude field, called just Instructions under Settings > Your Claude in the newer layout (Anthropic Help Center); the walkthrough is in Claude preferences: how to set instructions for Claude. If you're switching, move your custom instructions from ChatGPT to Claude by hand rather than through the memory import.
Neither app reads the other's instructions, so people who use both maintain two copies. The global instructions guide shows where the field lives in both apps, with copy-ready instructions that tell each one to load your Unabyss context. Instructions also say only what you wrote down once. Unabyss builds the rest of your context from Gmail, Calendar, Notion and Slack, keeps it current, and gives it to ChatGPT through an official plugin and to Claude over MCP. For keeping context across sessions inside ChatGPT, read how to save context in ChatGPT.
