Custom instructions that make ChatGPT sound like you

What it is good at

• Applies automatically to every new chat, so you never re-explain your role and audience.
• Reduces the generic 'AI voice', hedging, em dashes, list-heavy answers, more than any single in-chat prompt.
• Can include hard rules such as banned words or a fixed reply length that persist without repeating them.
• Works alongside project-level instructions, so you get a general baseline plus per-project overrides.

Where it falls short

• It is still a request, not a lock, heavily complex or long tasks can drift back toward default style.
• Overly long instructions can confuse rather than help; concise and specific beats exhaustive.
• It does not retroactively fix chats you already had before setting it up.

How to use it for creative work

1. 01

   Answer the 'about you' field with facts, not adjectives

   State your role, what you make, who buys it and your market, rather than 'I am creative and driven'. Facts change what it retrieves and how it frames answers; adjectives do nothing.

2. 02

   Give the response style field concrete rules

   Specify sentence length limits, whether you want bullet points or prose, and a formatting preference such as no headers under 300 words. Concrete rules are followed; abstract tone words are interpreted loosely.

3. 03

   Paste two examples of writing you like

   Include a short sample of your own copy or a competitor's tone you admire and say 'match this register'. Examples do more work than description.

4. 04

   Add explicit no-go words

   List clichés you are tired of seeing, 'elevate', 'seamless', 'game-changing', as words to never use. This one line noticeably cleans up copy.

5. 05

   Re-test with a real brief after saving

   Run one actual piece of work through it immediately, then edit the instructions again based on what still felt off. Treat it as a first draft you refine, not a one-time setup.

Ready-to-use prompts

I am a [role] at a [type of business] selling to [audience]. My main outputs are [copy type/design type]. Assume this context in every answer unless I say otherwise.

Reply in plain sentences of no more than 20 words on average. Use bullet points only for lists of five or more items. Never use these words: [list]. Do not open with a summary restating my question.

Match the tone, sentence rhythm and vocabulary level of this sample in all future replies: [paste 100-150 words of your own writing].

For any product description request, always return exactly: a 6-word headline, a 40-word paragraph, and three bullet features of under 8 words each. No other format.

Read my current custom instructions below and tell me which parts are vague enough that you would interpret them differently on different days: [paste instructions].

Questions people ask

What should I write in ChatGPT custom instructions?

Concrete facts about your role, audience and output type in the first box, and specific formatting and tone rules, including banned words and sentence length, in the second. Avoid vague adjectives; they get interpreted inconsistently.

Do custom instructions work in every chat?

Yes, they apply globally to new chats by default. Inside a project, project-specific instructions are used instead of or alongside the general ones, with project instructions taking priority.

Why does ChatGPT still sound generic after I set instructions?

Usually because the instructions describe a tone in adjectives rather than giving concrete rules or a real writing sample to match. Replace description with an example and a hard rule set.