Custom Instructions

Why customize ChatGPT?

Without configuration, ChatGPT responds generically. It doesn’t know your profession, preferences, or context unless its memory has retained details from past conversations. The result is predictable: you spend the first three messages of every chat re-explaining who you are, what you do, and how you want the answer formatted. Multiplied by ten conversations a week, this wasted time adds up—and the day you forget to provide context, the response misses the mark.

Custom Instructions change the game. You configure ChatGPT once in your settings, and it applies those rules to every conversation. They’re available on all plans, across the web, desktop app, iOS, and Android. This is the first building block of personalization—the foundation this course relies on before covering memory and creating custom GPTs.

Where to set them up

On the web and desktop app, go to Settings > Personalization; on iPhone and Android, tap Customize ChatGPT. Ensure the Enable customization toggle is on, then write your guidelines in the Custom Instructions field. Space is limited: 1,500 characters on Free and Go plans, 5,000 on paid tiers. The same screen offers two related settings: Base style and tone, which selects a personality like Default, Professional, or Friendly, and Characteristics, which adjust warmth, enthusiasm, formatting, and emoji usage.

Two important notes. In a Project, Project-specific instructions override your global ones. Also, if you use third-party plugins, OpenAI warns that parts of your instructions may be shared with their developers: never include anything you wouldn’t want them to see.

Two questions to answer

Good instruction text answers two distinct questions. Confusing them is the most common mistake.

The first part answers: What should ChatGPT know about you? Describe your profession and industry, your expertise level on the topics you cover, your working context—team size, tools used, target market—and any specific constraints, whether regulatory, linguistic, or format-related. Consider a concrete example: if ChatGPT knows you lead a small team targeting the French market, it stops suggesting strategies calibrated for a US multinational. Contextual information filters out unsuitable responses from the start.

The second part specifies the desired response format: preferred tone (formal, direct, educational), length (concise or detailed), format (bullet points, paragraphs, tables), and elements to consistently include or exclude. This section is often neglected, yet it transforms the experience most visibly. Asking once for responses under 300 words with a concrete example saves you from repeating that request in every message—and guarantees a consistency that occasional reminders never achieve.

Your turn

Your instructions in two parts. Here’s how a B2B marketing manager might write theirs; use this as inspiration to draft your own, then test them on three everyday requests.

Who I am: “I’m a marketing manager at a 50-person B2B SaaS SME. My markets are France and Benelux. I use HubSpot, LinkedIn Ads, and Notion daily.”

How to respond to me: “Respond directly and actionably. Use bullet points. Always provide a concrete example. Keep responses under 300 words unless I ask for more. Avoid English jargon when a French equivalent exists.”

Notice the logic: every sentence provides actionable information. “50-person B2B SaaS SME” sets the maturity level for advice; mentioning HubSpot lets ChatGPT suggest actions directly executable in that tool; the 300-word limit prevents rambling responses. With these instructions in place, every new conversation accounts for your context without you having to repeat it.

Pitfalls to avoid

The first pitfall is vagueness. “I work in marketing” tells ChatGPT nothing: marketing what, for whom, with which tools? Compare that to the example above—the difference in precision shows up word for word in response quality. Conversely, overload is equally counterproductive: ChatGPT struggles to apply twenty guidelines, especially if some contradict each other. Aim for a few clear rules, say five to eight, and reserve exceptional instructions for specific prompts.

Also consider expiration. Your January instructions might not reflect your role, tools, or priorities by June; a manager who changed teams but keeps receiving responses calibrated for their old scope won’t understand why ChatGPT “gets it wrong.” Make it a habit to review your instructions quarterly. Finally, avoid the most common mistake: carefully describing who you are but saying nothing about how to respond, even though that’s what changes the outcome most.

Official sources: ChatGPT Custom Instructions (OpenAI Help Center: path, activation, limits, third-party plugins) · Customizing your ChatGPT personality (Base style and tone) · Characteristics in ChatGPT · Projects in ChatGPT (Project instructions)

Key takeaways

• Custom Instructions automatically apply to all your conversations
• Settings > Personalization (Customize ChatGPT on mobile): 1,500 characters on Free and Go, 5,000 on paid plans
• Precise, concise instructions beat long, vague ones—aim for 5 to 8 rules
• Cover two questions: who you are, and especially how to respond to you; Base style and Characteristics complement these
• Review your instructions quarterly to keep them aligned with your reality