Every time I open a fresh ChatGPT conversation, the AI starts blind — it has no idea what I do for work, the tone I prefer, or whether I want tight bullet points or a deeper walkthrough. I used to burn three or four messages just rebuilding that context before getting a single useful answer. ChatGPT Custom Instructions kill that warm-up tax permanently by storing your context once and applying it to every new chat.
Custom Instructions are a built-in feature on both free and paid ChatGPT accounts. You define two things — background about yourself and how you want ChatGPT to respond — and every new conversation inherits those preferences automatically. I set mine up in under three minutes, and my answers have started sharper ever since.
Quick Answer
Open ChatGPT, click your profile icon, then go to Settings, Personalization, and Custom Instructions. Fill the first field with your role and context, and the second with your preferred response style. Toggle Enable for new chats on, save, and every conversation you start after that uses your preferences automatically.
What Are ChatGPT Custom Instructions?
Custom Instructions are two persistent text fields stored with your OpenAI account:
- “What would you like ChatGPT to know about you?” — your profession, skill level, location, or ongoing projects.
- “How would you like ChatGPT to respond?” — tone (formal or casual), length (concise or detailed), format (bullets or prose), or any special preferences.
These act as a silent system prompt prepended to every new conversation. ChatGPT reads them before it reads anything I type, so I never repeat myself across sessions. When I added “freelance tech writer, plain-English voice” to mine, the AI stopped padding answers with disclaimers I never wanted.
Custom Instructions are a standing brief ChatGPT consults before your first word in every new chat.
Who Can Use Custom Instructions?
Custom Instructions are available on free ChatGPT accounts and ChatGPT Plus alike, so you do not need to pay to use them. The one exception is Temporary Chat mode, which is built for private sessions with no saved context and silences your instructions by design. The first time my answers felt oddly generic, I realized I had opened a Temporary Chat without noticing.
Every account tier supports Custom Instructions except Temporary Chat, which bypasses them on purpose.
How Do I Set Up Custom Instructions?
The path differs slightly between desktop and mobile, but both take a couple of minutes. I walk through each below.
On Desktop (Browser)
- Go to chatgpt.com and sign in.
- Click your profile icon in the bottom-left corner.
- Select Settings.
- Click Personalization in the left sidebar.
- Click Custom Instructions.
- Fill in both text fields (see the examples in the next section).
- Toggle Enable for new chats on, then click Save.
One thing tripped me up early: changes apply only to conversations started after you save, not to any chat already open. I now open a fresh tab right after saving to confirm the instructions took.
On the ChatGPT Mobile App (iOS and Android)
- Tap the three horizontal lines in the top-left corner.
- Tap your profile name at the bottom of the menu.
- Tap Custom Instructions.
- Fill in both fields and tap Save.
Setup lives under Personalization on desktop and your profile name on mobile, and only new chats inherit the changes.
What Should I Write in Each Field?
This is where most people stall. The key is specificity — vague instructions produce vague improvements. Below are examples I have tested across common roles so you can adapt one to your own work.
| Use Case | “Know About You” Field | “How to Respond” Field |
|---|---|---|
| Developer | Senior backend engineer, Python/Django, building B2B SaaS | Skip basic explanations, show code first, explain trade-offs briefly |
| Student | Undergrad biology student studying for exams | Use simple language, add analogies, end with a 2-sentence summary |
| Non-technical professional | Marketing manager, no coding background | Avoid jargon, keep responses under 200 words unless I ask for more |
| Writer | Freelance copywriter, B2B tech niche | Match professional tone, prefer active voice, flag weak word choices |
| Parent / Educator | Homeschooling parent, kids ages 8 and 11 | Explain as if to a curious 10-year-old, use real-world examples |
If your responses still feel generic after saving, confirm the Enable for new chats toggle is on and that you have opened a new conversation. Instructions will not retroactively apply to a chat that was already open when you saved. To go further on phrasing, my guide to writing ChatGPT prompts like a pro pairs well with a solid instruction set.
Specific role details and 2-3 clear response rules beat long, hedged paragraphs every time.
What Are the Common Mistakes to Avoid?
I made most of these before they became habits to skip. Here is what trips people up and the fix for each.
- Writing too much. Each field holds up to 1,500 characters, but cramming in every detail makes ChatGPT deprioritize what matters. Fix: stick to your 2-3 most important facts and 2-3 core response preferences.
- Contradicting yourself. Asking for “concise responses” in one field and “always include detailed step-by-step breakdowns” in the other creates unpredictable output. Fix: decide your priority and stay consistent.
- Never revisiting them. Your role, projects, and goals change, and stale instructions are almost as unhelpful as none. Fix: set a reminder to review them every few months.
- Storing sensitive data. Passwords, financial details, and private health information get sent on every prompt and stored on OpenAI’s servers. Fix: keep that data out of these fields entirely.
- Forgetting Temporary Chat silences them. A session opened in Temporary Chat bypasses your instructions by design. Fix: check the top of the chat window when answers feel impersonal.
To see the bigger picture of what ChatGPT stores and how it builds a profile from your chats, read my guide on what ChatGPT remembers and how to take back control.
Most failures come from overloading the fields or contradicting yourself, and each has a one-line fix.
Frequently Asked Questions
Do Custom Instructions work with GPT-4o and other models?
Yes, they apply regardless of which model you select. When I switch between GPT-3.5 and GPT-4o mid-project, the same instructions feed in as a system prompt before my first message every time.
Can people see my Custom Instructions if I share a chat link?
No, shared links display only the visible messages in that specific chat. I have shared dozens of conversation links with editors, and my instructions never appeared in any of them.
How are Custom Instructions different from ChatGPT’s memory feature?
Custom Instructions are static text you write yourself, while memory automatically saves facts ChatGPT picks up as you talk. I keep both on — my instructions set the voice, and memory remembers details like the project names I mention.
Will Custom Instructions affect image generation requests?
Partially — tone and style preferences will not change image output, but role context can shape how ChatGPT interprets an ambiguous prompt. When I ask for a “diagram,” my “tech writer” context nudges it toward clean, labeled visuals.
Can I have different Custom Instructions for different projects?
Not natively, since there is one global set per account. As a Plus user I use Projects, which let me override the global set inside a workspace — I keep a separate, code-first instruction set in my development project.
What happens if I delete my Custom Instructions?
ChatGPT reverts to default behavior as if none existed, and your open conversations are unaffected. I cleared mine once to troubleshoot a weird tone issue, then pasted them back in seconds.
Conclusion
Custom Instructions offer one of the best setup-to-payoff ratios in any AI tool — a few minutes of configuration that improves every conversation afterward. Fill both fields thoughtfully, save, and open a new chat to feel the difference right away. Still weighing your options? My ChatGPT vs Gemini vs Claude comparison breaks down which free AI fits your workflow next.