# Tell it who you are, once A one-page text that stops you re-explaining yourself in every chat. Start here · v1.1 · draft · https://hayzlabs.com/start/tell-it-who-you-are/ In short: - Write a short text on your role, how you want answers and what to avoid, and keep it at or under 1,500 characters. - Paste it into your assistant's custom instructions, rerun an earlier task and keep only the lines that changed the answer. - Keep the master copy in a file in a folder you own, and when you change one, change the other. This is step 4 of [Start here](https://hayzlabs.com/start/start-here/). ## Steps 1. **Write the five lines (10 minutes).** Copy the template below and fill it in. Anything you write here goes into every chat, so leave out what [What not to paste](https://hayzlabs.com/start/what-not-to-paste/) told you to leave out: passwords and access keys, other people's details, anything under an NDA. 2. **Count the characters (2 minutes).** Paste the text into a notes app that shows a character count. Stay at or under 1,500. If you are over, delete the line that would change the least. 3. **Paste it into your assistant (5 minutes).** Find the custom instructions setting (a box of text the assistant reads before every chat), or the instructions field of a project (a workspace for related chats, in apps that offer one), and paste the text there. The box below says where to look. 4. **Test it (5 minutes).** Run one task from [Your first useful hour](https://hayzlabs.com/start/your-first-useful-hour/) again in a new chat. Compare with your first result. Keep the lines that changed the answer and delete the lines that did not. 5. **Save your copy as `ABOUT_ME.md` (2 minutes).** Put this plain text file in a folder you own. The setting in your assistant is a paste of this file. When you change one, change the other. Paste this template into a notes app and fill it in: ```text title="Tell it who you are · personal instructions · v1.0" About me: in . I know well. Explain from the basics. How to answer: lead with the answer, then the reasoning. Use plain words. If you are unsure, say "I don't know" and say what would settle it. Format: short paragraphs. Use a table when comparing three or more things. No emojis. Do not: flatter me, pad the answer, or add warnings I did not ask for. When a task is ambiguous: ask one clarifying question, then carry on. ``` Replace the following: - ``: your job or main activity, such as `project manager`. - ``: the field you work in, such as `logistics`. - ``: where it can skip the basics, such as `spreadsheets and budgeting`. - ``: where you want the basics, such as `contract law`. How to tell it worked: shorter openings, answers that start with the answer, and fewer follow-up corrections from you. The template is our suggestion and has not been tested. > **Where to look, as of 2026-10-11.** ChatGPT: Settings > Personalization, with Enable customization on, then Custom instructions (phone apps: Settings > Customize ChatGPT). Claude: your initials (lower left) > Settings > Instructions for Claude. Gemini: Settings & help > Personal Intelligence > Instructions for Gemini > Add. Menus move; the vendor's help page wins. ## Why a file you own ChatGPT custom instructions hold up to 1,500 characters on Free and Go. Since 15 July 2026, Plus, Pro, Business, Enterprise and Education users can save up to 5,000 ([OpenAI help](https://help.openai.com/en/articles/8096356-custom-instructions-for-chatgpt)). We keep the template at 1,500 so it fits every plan. Built-in memory (where the assistant saves things it picks up about you from your chats) is a separate feature, and OpenAI says of it: "Unlike an explicit saved memory, information derived from chat history can change as ChatGPT updates what is most useful to remember." ([OpenAI memory FAQ](https://help.openai.com/en/articles/8590148-memory-faq)). So your own copy is the master, and the setting is a paste of it. > [!LIMIT] > Shorter is safer. In a 2026 ETH Zurich benchmark (a standard test) of coding agents (AI tools that write and run code on their own), files written by an AI did not improve success (about -2%; not significant) and raised costs by about 20%, and human-written files helped slightly at best (+2.4%; not significant) ([Gloaguen et al., ICLR 2026 workshop](https://arxiv.org/abs/2602.11988), benchmark). That test used coding agents. We know of no test for an everyday-work text like this one, so step 4 tests it on your tasks. ## What you keep A text of 1,500 characters or fewer, pasted into your assistant, and a copy in `ABOUT_ME.md`. ## Next - [Prompting in one page](https://hayzlabs.com/start/prompting-in-one-page/) covers the prompt itself, which your about-me file does not replace. - [Step 5: Check before you trust](https://hayzlabs.com/start/check-before-you-trust/) (15 to 20 minutes). ## Sources 1. OpenAI Help. "Custom Instructions for ChatGPT" and ChatGPT release notes (15 July 2026 entry). Vendor page. [Link](https://help.openai.com/en/articles/8096356-custom-instructions-for-chatgpt) · [Link](https://help.openai.com/en/articles/6825453-chatgpt-release-notes) 2. OpenAI Help. "Memory FAQ." Vendor page. [Link](https://help.openai.com/en/articles/8590148-memory-faq) 3. Gloaguen et al. "Evaluating AGENTS.md: Are Repository-Level Context Files Helpful for Coding Agents?" ICLR 2026 Workshop on Memory for LLM-Based Agentic Systems. Benchmark. [Link](https://arxiv.org/abs/2602.11988)