# ChatGPT Custom Instructions for Writing: 4 Sets to Copy (2026) > Four ChatGPT custom instruction sets to copy, for every chat, work emails, blog posts and academic work, each under the 1,500-character limit. Source: https://hidengpt.com/guides/chatgpt-custom-instructions-for-writing # What custom instructions should I give ChatGPT for writing? Updated October 10, 2026 · 8 min read · by the HidenGPT team Good custom instructions for writing describe your style in terms ChatGPT can check (sentence length, words to use and avoid, how to open and close), say what to write instead of just listing what to avoid, and fit the limit: OpenAI's help page gives 1,500 characters on the Free and Go plans and 5,000 on Plus, Pro, Business, Enterprise and Education. Custom Instructions apply to every chat, so keep one short general set there and put task sets into project instructions, which override it inside each project. The four sets below are each under 1,500 characters. Swap in your own numbers before you use them. | Set | Length | Fits the 1,500 limit on Free and Go | Paste it into | |---|---|---|---| | 1. General writing rules | 737 characters | Yes | Custom Instructions | | 2. Work emails | 725 characters | Yes | The instructions of a project called Email | | 3. Blog posts and newsletters | 810 characters | Yes | The instructions of a project for your blog or newsletter | | 4. Academic work at university | 984 characters | Yes | The instructions of a project for one course | ## How to set them up On the web or the desktop app, open Settings, select Personalization, check that Enable customization is on and paste Set 1 into the Custom Instructions field. On iOS and Android the menu is called Customize ChatGPT. OpenAI offers Custom Instructions on all plans, with room for 1,500 characters on Free and Go and 5,000 on the paid plans. For the task sets, make one project each: select New project in the sidebar, open the project's ••• menu, choose Project settings and paste the set as the project's instructions. OpenAI's projects page says project instructions apply only within that project and override your global Custom Instructions. Because of that override, don't count on Set 1 reaching inside a project; each task set repeats the rules that matter for its job. Then test each one. Ask for the same short piece in a plain chat and in the project, and read the two drafts side by side before you trust the set. ## Set 1, for every chat, and Set 2, for work emails Set 1 is short on purpose, so it stays sensible for anything you ask. Copy this: When you write anything I will send or publish, write plain English for an adult reader. Mix short and long sentences, averaging about 15 words. Use contractions. Start with the point and skip warm-up lines. Name the thing instead of describing how important it is, and write “is” rather than “serves as”, “use” rather than “utilize”. Leave out “delve”, “tapestry”, “in today's fast-paced world”, “it's important to note”, “plays a crucial role” and “not only... but also”. End on the last useful point, not a summary. Use commas, periods and parentheses instead of em dashes. Never invent facts, names, numbers, quotes or sources: if something is missing, write [CHECK] and keep going. If the task is unclear, ask me one question first. Set 2 goes into a project called Email. Copy this: You draft my work emails. My readers are busy colleagues and clients, so the request or the news goes in the first two sentences. Keep most emails under 120 words, in paragraphs of two or three sentences, with no headings and no bullet points unless I ask. Match the formality of the email I'm answering: if they wrote “Hi” and signed with a first name, do the same. One clear ask per email, with a date when there is a deadline. Close with the next step, then “Thanks,” and my first name, unless the thread is formal. Never open with “I hope this email finds you well” or “I wanted to reach out”, and apologize at most once. Suggest a subject line of six words or fewer. Use only facts I give you and mark gaps with [CHECK]. Change the numbers to match how you write. If your own emails usually run to 200 words, a 120-word cap will make every draft sound clipped, and people who know you will hear it. ## Set 3, for blog posts and newsletters Paste this into a project for your blog or newsletter, and upload two or three past posts as files so ChatGPT can see the voice the rules describe: You help me draft blog posts and newsletter issues for readers who already know the basics of my topic. Open with a specific scene, number or claim from my notes, not a question or a definition. One idea per paragraph, four sentences at most. Use subheadings only in posts over 800 words, and write them as plain statements. Write in the first person and use my examples; where a point needs an example I haven't given, write [EXAMPLE] instead of making one up. Use a list only for real steps or parallel items, and no more than one list per post. Vary sentence length, with a short sentence after a long one now and then. End with one concrete thing the reader can do or check, not a recap. Avoid “in this post we will explore”, “let's dive in”, “game-changer”, “unlock your potential” and “the key takeaway”. The [EXAMPLE] placeholder does the most work here. Without it, ChatGPT fills the gap with an anecdote that sounds like yours but never happened, and regular readers are the first to notice. ## Set 4, for academic work at university Use this only in courses that allow AI help, and only in the way they allow it; some allow planning and feedback but not drafting. Paste it into a project for one course: I'm a university student. Help me plan, draft and revise my own academic writing, within my course's rules on AI, which I will paste when they matter. Write clear, formal English: third person unless I say otherwise, no contractions, and precise verbs such as “argues”, “shows” or “suggests” instead of “talks about”. Keep my thesis and my argument; if you think the argument is weak, say so in a note instead of changing it. Never invent sources, quotations, page numbers or data. Cite only sources I give you, in the style I name (APA 7, MLA 9, Chicago or Harvard), and where a claim needs a source I haven't given, write [SOURCE NEEDED]. Each paragraph opens with a claim, then gives evidence, then says what the evidence shows. Hedge only where the evidence is uncertain. Avoid “delve”, “this essay will explore”, “in conclusion”, “it is important to note” and vague credit such as “experts say”. When I ask for feedback, list problems paragraph by paragraph instead of rewriting. Name your citation style in each request, and check every reference against the source yourself. Telling ChatGPT not to invent sources lowers the risk; it doesn't remove it. ## Why the sets are written this way Each choice follows advice the two largest vendors publish about their own tools. Say what to do, not only what to avoid. OpenAI's page on building GPTs recommends positive, concrete instructions over long lists of prohibitions, so every avoid list above sits next to a plain alternative: “is” instead of “serves as”, the request instead of a warm-up line. Give the reason. Anthropic's prompting guide says that explaining why an instruction matters helps Claude apply it to cases you didn't list. “My readers are busy colleagues” does more than “be concise”, and it costs a few characters. Write in the shape you want back. The same guide says, about Claude, that the formatting of a prompt can carry over into the reply. It's a fair habit to assume in ChatGPT too, which is why these sets are paragraphs rather than bullet lists. Give the model a way out. [CHECK], [EXAMPLE] and [SOURCE NEEDED] let it leave a gap you can see instead of filling it with something invented. The phrases in the avoid lists come from the English list that HidenGPT's red pen flags, more than 60 patterns from “delve” to “in conclusion”. If a phrase that bothers you isn't in a set, add it, and take out any you actually use. ## Common mistakes - Writing a wish list of adjectives such as “engaging, professional, concise”: give numbers and habits ChatGPT can check its draft against. - Putting every set into Custom Instructions: keep one short general set there and each task set in its own project. - Only banning words: say what to write instead, such as “is” rather than “serves as”. - Writing past the limit: Free and Go take 1,500 characters, so keep each set under it and put the rule you care about most first. - Skipping your course's rules: for academic work, check what your syllabus allows before you set up the project. - Never testing a set: run the same request with and without it and compare the two drafts. ## Questions ### What is the character limit for ChatGPT custom instructions? 1,500 characters on the Free and Go plans and 5,000 on Plus, Pro, Business, Enterprise and Education, according to OpenAI's help page read on 10 October 2026. Every set on this page fits the smaller limit. ### Where do I find custom instructions in ChatGPT? On the web or desktop app, go to Settings, then Personalization, turn on Enable customization and use the Custom Instructions field. On iOS and Android, open Settings and choose Customize ChatGPT. ### Can I use different custom instructions for different kinds of writing? Yes, through projects. Custom Instructions are one global set, but each project has its own instructions, which apply only inside it and override the global set. Make one project per kind of writing. ### What are the best ChatGPT custom instructions for writing? The ones that describe you. A set copied unchanged from any page, this one included, makes ChatGPT sound like the person who wrote it. Start from a set above and replace the sentence length, the words and the openings with your own. ### What custom instructions work for creative writing? Describe the voice of the story rather than your own: tense, point of view, how much the narrator explains and how long scenes run. For example: “Past tense, close third person on Mara. Scenes of 300 to 600 words. Show feelings through what she does and never name them. No recaps at chapter ends.” Keep it in a project for that book. ### Should I put my writing samples in custom instructions? Usually not. At 1,500 characters there is barely room for one paragraph. Upload samples as files to a project instead (OpenAI lists 5 files per project on Free, 25 on Go and Plus, and 40 on Pro, Business, Enterprise and Edu) and keep the instructions for rules. ### Do custom instructions override the personality setting? They work together. OpenAI says your personality works alongside your custom instructions and memories, and that drafts such as emails or résumés follow your instructions and the context rather than the personality. For writing, the instructions matter more. ## Sources - [OpenAI Help Center: ChatGPT Custom Instructions](https://help.openai.com/en/articles/8096356-chatgpt-custom-instructions) (checked October 10, 2026) - [OpenAI Help Center: Projects in ChatGPT](https://help.openai.com/en/articles/10169521-projects-in-chatgpt) (checked October 10, 2026) - [OpenAI Help Center: Customizing your ChatGPT personality](https://help.openai.com/en/articles/11899719-customizing-your-chatgpt-personality) (checked October 10, 2026) - [OpenAI Help Center: Creating and editing GPTs (writing effective instructions)](https://help.openai.com/en/articles/8554397-creating-and-editing-gpts) (checked October 10, 2026) - [Claude Platform Docs: Prompting best practices](https://platform.claude.com/docs/en/build-with-claude/prompt-engineering/claude-prompting-best-practices) (checked October 10, 2026) ## Related - [My draft reads like ChatGPT wrote it. How do I make it sound like a person?](https://hidengpt.com/write/make-ai-writing-sound-human) - [I want LinkedIn posts that sound like me, not like a LinkedIn post generator.](https://hidengpt.com/write/linkedin-post-generator-in-your-voice) - [I need an AI essay writer that only uses the sources I give it and doesn't make up citations.](https://hidengpt.com/write/essay-with-real-citations) - [I need a thank-you email after my interview that doesn't sound copied.](https://hidengpt.com/write/thank-you-email-after-interview) - [How do I make ChatGPT write like me?](https://hidengpt.com/guides/how-to-make-chatgpt-write-like-you) - [Why does ChatGPT use so many em dashes?](https://hidengpt.com/guides/why-does-chatgpt-use-em-dashes) - [What words does ChatGPT use a lot?](https://hidengpt.com/guides/words-chatgpt-uses-a-lot) - [Is Claude or ChatGPT better at writing in my style?](https://hidengpt.com/guides/claude-vs-chatgpt-for-writing) - [Does my writing sound like AI?](https://hidengpt.com/tools/ai-cliche-checker) [Add HidenGPT as a preferred source on Google](https://www.google.com/preferences/source?q=hidengpt.com) to see these guides more often in search and AI Overviews.