Most ChatGPT Users Don't Know Custom Instructions Can Become a Personal Writing Agent
Everyone enables Custom Instructions for basic preferences—but a richly written profile can teach ChatGPT your exact voice, jargon, and style so every new chat drafts the way you write.
01. What It Is
Custom Instructions is a built-in ChatGPT setting with two text boxes: one describing who you are, and one describing how you want ChatGPT to respond. Most people drop in a line or two ('be concise', 'I'm a marketer') and move on. The overlooked power is treating those boxes as a living style guide for a personalized writing assistant.
Instead of generic preferences, you fill the fields with specifics: your profession and audience, the tone you use, words and phrases you favor or avoid, sentence length, formatting habits, and industry vocabulary. Because these instructions apply automatically to every new conversation, ChatGPT starts each chat already aligned to your voice—no need to re-explain yourself.
It's important to understand what this is and isn't. Custom Instructions shape tone, persona, and style; they are guidance, not true memory or learning. ChatGPT won't 'remember' edits across chats on its own. But by iteratively refining the instruction text yourself—copying patterns you keep correcting into the boxes—you effectively teach it over time. For voice and style personalization, this is enough. For source-grounded knowledge or complex business rules, you'd need other tools.
Why It Matters
A well-tuned profile turns ChatGPT from a generic writer into one that drafts in your voice from the first sentence, cutting the back-and-forth editing that eats the most time. Writers, founders, and busy professionals can shave minutes off every draft and produce copy that sounds like them—not like a bot—without re-pasting style notes into each new chat.
Who Can Benefit
- Writers and content creators who need a consistent personal voice
- Marketers and founders producing on-brand copy daily
- Professionals with heavy industry jargon (legal, medical, finance, tech)
- Non-native English speakers who want consistent tone and clarity
- Anyone tired of re-explaining their preferences in every new chat
02. Step-by-Step Guide
- 1
Open Custom Instructions
In ChatGPT go to Settings → Personalization → Custom Instructions (on some versions it's labeled 'Customize ChatGPT'). This works on Web, Desktop, iOS, and Android, on Free and paid plans alike.
- 2
Turn on customization
Make sure the enable customization toggle is ON—otherwise nothing you write in the fields will be applied to your conversations.
- 3
Describe yourself with writing-relevant detail
In 'What would you like ChatGPT to know about you?' add your role, audience, industry vocabulary, and the publications or formats you write for. Each field allows up to 1,500 characters, so be specific rather than vague.
- 4
Define your voice in the response field
In 'How would you like ChatGPT to respond?' specify tone (e.g. warm but direct), sentence length, words to avoid, formatting (bullets vs. prose), and reading level. Give examples of phrases you do and don't use.
- 5
Save and test in a new chat
Save, then open a brand-new conversation and ask for a draft. Remember: changes only apply to new chats, not existing threads, so always test fresh.
- 6
Refine iteratively
Notice what you still have to correct manually. Each session, distill those recurring fixes into a short rule and paste it back into the instruction fields. Over weeks this builds a sharp, personalized style profile.
What would you like ChatGPT to know about you? I'm a [your role] writing for [your audience] in the [your industry] space. My writing appears in [formats: emails, LinkedIn posts, blog articles, reports]. Key terms I use often: [list jargon]. I want my writing to feel [adjective] and [adjective]. How would you like ChatGPT to respond? Write in my voice: [tone, e.g. direct, warm, lightly witty]. Use short paragraphs and active voice. Avoid these words/phrases: [list, e.g. 'delve', 'leverage', 'in today's fast-paced world']. Prefer [bullets / flowing prose]. Default reading level: [e.g. clear, conversational]. When I share a draft, return an improved version plus a 1-line note on what you changed and why.
Pro Tips
- Add a 'change log' line to your response field asking ChatGPT to briefly note what it edited—this helps you learn your own patterns.
- Keep a separate note with your full style guide; paste a trimmed version into the 1,500-character fields and update it over time.
- If you use the 2025 Personality dropdown, align it with your Custom Instructions—conflicting settings can produce inconsistent tone.
- For one-off projects with different voices, override the instructions inside the chat itself; later turns take precedence over the saved profile.
Warnings & Limitations
- Custom Instructions are guidance, not guarantees—they can be overridden by safety policies, later messages, or long conversation context.
- They do not provide true learning or memory; you must manually refine the text to 'teach' your style.
- Editing instructions never updates ongoing chats; only new conversations reflect changes.
- For source-grounded knowledge or multi-step business rules, Custom Instructions alone are not enough—you'd need RAG or other tools.
- Data from Custom Instructions may be used to improve the model unless you opt out, and enabled third-party plugins may access them.
Related Tricks
Most ChatGPT Users Don't Know It Can Record and Summarize Any Meeting on Your Mac
ChatGPT's built-in Record Mode captures calls, brainstorms, and voice notes—then auto-generates a structured summary with action items, no separate note-taking app required.
This ChatGPT Trick Can Turn Any Photo Into a Reusable Template
Upload one image, extract its exact style as a written 'style recipe', and reuse it forever to generate consistent visuals on demand.
Most People Don't Know Gemini Can Now Work While They Sleep
Gemini Spark is a cloud-based agent that keeps running custom workflows even after you lock your phone. It turns your AI assistant from a chatbot into an always-on employee.