The generator prompt
This is the whole point. Instead of staring at a blank custom-instructions box wondering what to write, you paste the prompt below into ChatGPT and it interviews you — your business, your audience, your voice, the words you'd never be caught using, how you sign off. Then it hands you back a finished, paste-ready block sized to fit the new 5,000-character limit. Copy it, drop it into a fresh chat, answer the questions like you're talking to a mate.
You are my ChatGPT Custom Instructions Architect.
Your job: interview me, then write me a polished, paste-ready set of custom instructions (up to 5,000 characters) so every future ChatGPT reply understands my world and sounds like me — without me re-explaining myself every single chat.
HOW THIS WORKS
1. You interview me a few questions at a time. Never dump all the questions at once — I'll lose the plot.
2. I answer in plain, messy language. Don't make me be articulate; that's your job.
3. When you've got enough, you write the final block and tell me exactly where to paste it.
INTERVIEW ME — go section by section, 2 to 4 questions per message, and wait for my answer before moving on:
1. ME + MY WORK: What do I do? What's my business or role? What am I known for? What do I mainly use ChatGPT for day to day?
2. MY AUDIENCE: Who am I usually writing or building for? What do they care about? Are they beginners, experts, or a mix?
3. MY VOICE: How do I want to sound — give me 3 to 5 words. Formal or casual? Warm or blunt? Funny or straight? If I can, I'll paste something I've written that already sounds like me — pull the patterns out of it.
4. MY WORDS: Words, phrases and spellings I always want used (including British/Australian vs American spelling). Then the words and phrases I never want to see — the ones that make me cringe or sound like a bot.
5. FORMAT: Short punchy paragraphs or full detail? Bullet points or prose? Headings? Emoji or none? How long should a typical answer be?
6. HOW TO TREAT ME: Should you challenge my thinking or just execute? Ask before assuming, or take a best guess and go? How much explaining on the technical stuff?
7. SIGN-OFF + DETAILS: How do I sign off emails and messages? Any name, links, title or details that should show up consistently?
8. HARD RULES: Anything you must ALWAYS do. Anything you must NEVER do.
Before you write, ask me if there's anything else you should know.
WHEN YOU WRITE THE FINAL BLOCK
- Keep it under 5,000 characters and tell me the exact character count.
- Write it as clear instructions TO ChatGPT, in the second person ("Write in...", "Never use...", "Address my audience as...").
- Organise it under two headings so it drops straight into ChatGPT's two boxes:
ABOUT ME — my work, my audience, my context (the "what would you like ChatGPT to know about you" box)
HOW TO RESPOND — my voice, formatting, rules and sign-off (the "how would you like ChatGPT to respond" box)
- Be specific and concrete. "Casual but sharp, Australian spelling, short paragraphs, zero corporate jargon" beats "professional and friendly."
- Spell out my banned-words list explicitly so it actually gets avoided.
- Give me the finished block on its own in a copyable code box — no preamble, no commentary — then one line telling me exactly where to paste it.
Ask me your first 2 to 4 questions now.
How to use it (start to finish)
If it misbehaves
If ChatGPT dumps every question at once, or writes the block before it's actually interviewed you, just reply: "Interview me one section at a time first, then write it." It'll fall back into line.
What custom instructions actually are
Think of custom instructions as the onboarding document ChatGPT reads before every single conversation. It's the north star of who you are: your business, your audience, how you want it to talk. Set it once and you stop starting from zero every time you open a chat.
Most people have never touched them. That's the opportunity — this is genuinely the highest-leverage ten minutes you'll spend on AI this month, and almost nobody bothers.
🤖 ChatGPT with nothing set
- You re-explain your business every chat
- Replies come out stiff and corporate
- It guesses who your audience is
- You rewrite most of what it gives you
- The "voice" changes every session
✨ ChatGPT that knows you
- Your context is baked in from the first word
- Replies land in your actual voice
- It writes for the people you serve
- Drafts you barely have to touch
- Consistent, every single chat
What just changed
Custom instructions used to cap out at 1,500 characters — barely room for the basics. They've now jumped to 5,000. More than triple the room means you can paste in your actual brand voice: the words you use, the words you'd never use, who you're speaking with, and how to sign off. If your AI output has been sounding generic, this is the fix.
Where to paste them (and Save)
The whole thing lives in one settings screen. Here's the exact path so you're not hunting through menus.
Open your settings
On desktop, click your name or profile picture in the bottom-left, then Settings. In the app, tap your profile, then the settings cog.
Go to Personalization
In Settings, open Personalization, then Custom instructions. This is where the magic gets stored.
Paste your block
Drop in what ChatGPT wrote you. If you see two boxes — "what to know about you" and "how to respond" — put the ABOUT ME half in the first and the HOW TO RESPOND half in the second. One box? Paste the lot.
Pick a personality (optional)
Same screen lets you set a personality for ChatGPT — a nudge on overall style. Handy, but your pasted block is doing the heavy lifting.
Hit Save
Save and you're done. From now on every new chat reads your instructions first — no button to press, no reminder to give it.
Good to know
Custom instructions apply to new chats, not the one you're already in. Set them, then start a fresh chat to feel the difference. Already-open conversations keep their old behaviour.
What a great set includes
The generator prompt asks about all of this for you — but here's what it's fishing for, so you know why each answer matters. These six are what turn a generic assistant into one that gets you.
Your business
What you do, what you sell, what you're known for. The context every reply should assume without you spelling it out again.
Your audience
Who you're speaking with and what they care about. It's the difference between a reply for beginners and one for experts.
Your tone
Three to five words for how you sound — warm, blunt, funny, formal. This is what stops the beige, robotic default.
Your words
The phrases and spellings you always use — and, just as important, the ones you'd never use. The "never" list is what kills the AI smell.
Your format
Short and punchy or full detail? Bullets or prose? Emoji or none? Tell it once and stop re-formatting every answer.
Your rules + sign-off
The always-dos, the never-dos, and how you close things out. The small stuff that makes a draft feel finished, not first-draft.
The "never" list does the heavy lifting
Telling ChatGPT what to avoid is where the biggest jump in quality hides. List the words that make you cringe, the phrases that scream template, the formatting you hate — and watch how much more like you it starts to sound. The generator prompt asks for this on purpose.
Generic vs dialled in
Same request, two different ChatGPTs. Here's what actually changes once it knows you. The ask: "Write me a quick Instagram caption for my new time-management workshop."
🫥 Before (guessing)
"Introducing our brand-new Time Management Workshop! Designed to help you achieve your goals and boost your productivity, this workshop delivers real value for professionals of all levels. Sign up today and take the first step towards success! ✅📈"
💛 After (knows you)
"You don't need more hours. You need to stop letting your calendar run the show. My new workshop is 90 minutes of exactly that — the system I use to protect my mornings and still clock off on time. Thursday, 10am. Link's in my bio."
The difference is the setup, not the prompt
Notice the second one didn't need a fancier request — same six words in. The context you loaded once is what pulled a real caption out instead of a template. That's the payoff for ten minutes of setup you never have to repeat.
Keep them sharp
A few habits keep your instructions pulling their weight instead of going stale.
Don't pad to 5,000 just because you can
Every line competes for the model's attention. Five sharp instructions beat twenty vague ones. Keep the lines that visibly change the output, cut the rest.
Test it on a real task
Right after you save, ask for something you'd normally ask — a caption, an email, a summary. If the voice is off, tweak one line and try again. Two rounds and it's dialled.
Update when your business shifts
New offer, new audience, new positioning? Re-run the generator prompt or edit the block directly. It's a living doc, not a set-and-forget.
Use Projects for one-off context
Keep your global instructions general. For a specific client or launch, spin up a ChatGPT Project with its own instructions so you're not cramming everything into the one box.
Mind the character count
If you're bumping the 5,000 limit, cut the weakest lines first — the ones stating the obvious. Specific and short always wins over long and hedged.
One honest caveat
Custom instructions steer ChatGPT, they don't handcuff it. Now and then a reply will still drift — when it does, just remind it in the chat ("that's not my voice, remember my instructions") and tighten the weak line in your block afterwards. Treat it as a dial you nudge, not a switch you flip once.
Ready to go deeper?
If ten minutes of setup can do this, imagine what AI can do for the rest of your business.
Wright Mode Membership
Join a community of women using AI to take real work off their plate. Live calls, copy-paste prompts and workflows, and ongoing support.
Claude Masterclass
Learn to use Claude like a pro — from prompting fundamentals to building real workflows that save hours every week.
Claude Code Masterclass
Go beyond the chat box. Build automations, process data and create tools with Claude Code — no developer experience needed.