ChatGPT Custom Instructions: Full Setup Guide

ChatGPT custom instructions are standing rules that apply to every new chat. This guide covers where to find them, the current character limits, how to write them, five ready-to-paste templates and why ChatGPT sometimes ignores them. It also shows how to customize ChatGPT beyond instructions, with Projects, memory and plugins.

Custom instructions are text you write once in your settings, and ChatGPT reads them at the start of every new conversation. They typically hold two kinds of information. The first is background about you, such as your job and expertise level. The second is rules for how to answer, such as length, tone and format.

This is different from ChatGPT memory. Custom instructions are what you tell ChatGPT, and memory is what ChatGPT picks up on its own from your chats.

Custom instructions apply to new conversations. You can edit or delete them at any time, but the change only affects chats you start afterward.

If you're using the ChatGPT app or web version, follow the steps below:

1. Open settings. On web and desktop, click your profile and open Settings, then Personalization. On iOS and Android, open Settings, then Customize ChatGPT.
2. Turn customization on. Toggle Enable customization.
3. Write your instructions. Enter them in the Custom Instructions field. Depending on your account and app version, you may see the older pair of boxes instead, "What would you like ChatGPT to know about you?" and "How would you like ChatGPT to respond?". If you see two, split your text that way. If you see one, put both in it.
4. Fill in the extras if you want them. The Personalization page also has short fields for a nickname and occupation, and a Base style and tone menu with presets such as Professional, Friendly and Candid. OpenAI says the preset works alongside your saved memories and custom instructions.
5. Save and start a new chat. Test with a request that your rules should clearly change.

The character limit depends on your plan. Free and Go users can save up to 1,500 characters. Plus, Pro, Business, Enterprise and Education users can save up to 5,000, a limit OpenAI raised from 1,500 on July 15, 2026.

A reliable instruction prompt structure has four parts.

• Who you are. Your role, field and experience level, so ChatGPT pitches answers at the right depth.
• What you want. The default format, length and tone of answers.
• Rules. A short list of things to always do and things to avoid.
• What to do when unsure. For example, ask one clarifying question or say that you're not certain.

Write rules you can check. "Keep answers under 150 words" works better than "be concise", because the model can tell whether it followed the first one. Here is how a few vague rules look once they're rewritten.

Leave out one-off tasks, private information such as passwords and anything you wouldn't want stored in your settings. Also avoid contradictory rules, such as "always ask questions before answering" next to "never ask questions". OpenAI's prompting guide for GPT-6 Astra warns that conflicting standing instructions make newer models stop, ask permission or hedge, and it recommends reviewing old instructions from time to time.

More space isn't a reason to fill it. A short, specific brief usually beats a long one.

Each template fits within the 1,500-character limit on Free and Go. Replace the brackets with your details.

1. Everyday assistant

I'm a [role] in [industry]. I know [topics] well and I'm newer to [topics]. I write in [US/UK] English.

Lead with the answer, then explain. Keep replies under 200 words unless I ask for more. Use plain language and define jargon the first time. If my request is ambiguous, ask one clarifying question before answering. If you're unsure of a fact, say so instead of guessing. Use bullet points only for lists of three or more items.

2. Writing that sounds human

Write like a person talking to a smart friend. Use contractions. Mix short and long sentences, and don't start two sentences in a row with the same word. Skip openers such as "Certainly" and "Great question". Don't use em dashes. Avoid the words delve, tapestry, landscape, leverage and game-changer. Don't write "it's not X, it's Y" constructions. Don't end with a summary of what you just said or an offer of further help. Prefer concrete details and examples over general claims. If I share a sample of my writing, match its tone.

3. Editor

Act as my editor. When I paste a draft, fix grammar and clarity, keep my voice and meaning, and change as little as needed. List each substantive change in one line with the reason. Don't rewrite whole paragraphs unless I ask. Flag any claim that needs a source. If something is unclear, ask before changing it.

4. Coding helper

I code in [languages and frameworks] at a [beginner/intermediate/advanced] level.

Show the code first, then a short explanation. Don't change code I didn't ask about. State any assumptions you made. If several approaches work, give the simplest and name the trade-off of the others. Add comments only where the logic isn't obvious. Mention security or performance risks briefly if you spot them.

P.S. If you're wondering what's the best AI for coding, read our detailed comparison.

5. Research and analysis

Separate facts from inferences and label each. Cite sources when you use them, and say when you can't verify something. Give the strongest counterargument to any conclusion you offer. End with what would change your answer. Use a table when comparing more than two options.

Instructions reduce errors but don't remove them, so keep checking facts that matter.

After saving, open a new chat and run three tests. Ask a simple question to check length and tone, make a request that should trigger one of your rules and send an ambiguous request to see whether it asks a clarifying question. Change one rule at a time, so you know which edit made the difference.

A custom ChatGPT can mean several things, and each tool works at a different level.

Two details matter in practice. Inside a Project, project instructions apply only there and override your global custom instructions. Custom GPTs don't use custom instructions or memory at all, according to OpenAI, and each conversation starts fresh.

Custom GPTs are also on their way out. OpenAI plans to retire them on December 11, 2026 and replace them with plugins, where a GPT's instructions become a skill and its knowledge files become reference files. If you rely on a custom GPT, migrate it before that date.

OpenAI hasn't published a priority order for every conflict between custom instructions, memory, project instructions, personalities and your message. When results look wrong, work through this checklist.

• Check the toggle. Enable customization must be on.
• Start a new chat. Edits don't change conversations that already exist.
• Check what else is enabled. Custom GPTs ignore custom instructions, and inside a Project, the project's instructions take over.
• Look for contradictions. Compare your instructions, saved memories and the message you sent. Use your message for one-off exceptions and your settings for defaults.
• Make the rules measurable. Replace "be brief" with a word limit and "sound natural" with specific habits to follow or avoid.
• Trim the block. If you're near the character limit, cut background details that don't change the answer.
• Reset the tone preset. If the style clashes with your rules, set Base style and tone back to Default.

In a very long chat, restating a key rule in your message helps, since early instructions can fade as the conversation grows.

ChatGPT adds your custom instructions to every new chat automatically. SYNTX.AI doesn’t have a settings field for them, but the idea carries over, because a model reads everything earlier in the conversation.

Start a chat with “Follow these rules for this whole conversation”, paste your instruction block and make your request. If the answers start to drift over time, repeat the key rule.

On SYNTX.AI, ChatGPT models such as GPT-6 Astra and GPT-6.1 Sol are available alongside Claude, Gemini and Grok. You can also use leading models for image, video and audio generation, including Seedance, ElevenLabs, Suno, Ideogram and Kling, all under one subscription.

Does ChatGPT still have custom instructions?

Yes. You'll find them under Settings, then Personalization on web and desktop, or Settings, then Customize ChatGPT on mobile. The limit is 1,500 characters on Free and Go and 5,000 on paid plans.

How do you set rules in ChatGPT?

Put standing rules in custom instructions, rules for one body of work in a Project's instructions and one-off rules in your message. Write each rule so that you can check whether it was followed.

How to make ChatGPT sound human with custom instructions?

Give it specific habits instead of "sound natural". Ask for contractions, varied sentence lengths, no stock openers, no summary at the end and concrete examples over general claims. Template 2 above puts these rules in one block.

Can I train ChatGPT on my own data?

Not in the sense of retraining the model. Custom instructions and memory shape how it responds, and Projects let it draw on files you upload, with 5 files on Free, 25 on Go and Plus and 40 on Pro, Business, Enterprise and Edu, each up to 512 MB. Real model training means fine-tuning through OpenAI's API, which is available for selected models.

What should I put in the custom instructions for ChatGPT?

Four things. Say who you are, how you want answers formatted, which rules it should always follow and what to do when it's unsure. Keep each rule specific and avoid contradictions.

Why does ChatGPT ignore custom instructions?

Common causes are customization being switched off, an old chat that started before your edit, a Custom GPT or Project overriding your global rules, contradictory instructions and vague wording. The checklist above covers each fix.