Do you spend your days telling ChatGPT who you are and what you do? You feel like Claude could understand you better but don't know where to start? In this article I share what I've gradually put in place so these tools actually work for me. The goal is simple: stop re-explaining everything every single time while keeping control of my data.
I do all this for myself first. When I started using ChatGPT and Claude daily, I wasted an insane amount of time repeating the same information. My line of work, how I prefer to receive answers, the tools I use every day, the format I expect. Every new conversation, it was the same thing all over again. I felt like a regular customer forced to recite their order every single time.
Custom instructions and memory changed everything. These aren't gadgets reserved for experts. It's like giving your assistant a permanent brief. ChatGPT calls them "custom instructions" and "memory". Claude talks about "Instructions for Claude" and "Styles". Either way, the idea is simple. It's about making sure the AI knows you well enough to be useful from the very first message, without you having to teach it everything again every time.
And the best part? These features have been available on every ChatGPT plan since at least October 2026. Whether you're on the free, Plus, Team, or Enterprise version. Same on mobile and desktop. For Claude, "Instructions for Claude" work at the account level and apply to every new conversation. You have no excuse to skip them.
Personally, I see these settings as a form of investment. Ten minutes of setup today saves hours of repetition tomorrow. But there's a downside. The more you tell your AI who you are, the more of yourself you expose. And not everyone realizes what this means for privacy. That's exactly what I want to share with you here. How to properly set up these tools, what to put in them, and most importantly what to keep out of them.
With ChatGPT, you have two distinct tools to customize your experience. Custom instructions are guidelines you write yourself. You explain who you are, what you expect, how you want the tone adjusted. The model factors them into every response. This saves you from repeating the same information every time.
Memory, on the other hand, develops differently. Introduced in 2025, it keeps useful details and preferences across your conversations. Your preferred tone, recurring info about your business, the formats you like. It grows from what you tell it, while instructions stay fixed until you change them.
That's a key difference to understand. Instructions are a system prompt you control. Memory is a living summary the AI builds from your exchanges. To do this right, I recommend keeping instructions for your stable business context and letting memory catch the day-to-day usage details.
If you're new to prompt engineering, these settings are an advanced form of prompt writing. I cover this in my guide on how to write a good prompt when you're not a developer. And to discover other hidden settings in the tool, check out my hidden ChatGPT features.
Don't overload your custom instructions. I tend to want to cram everything into this field, but the more you write, the more you risk creating contradictions. Better three clear lines than a novel the AI has to decode every time.
ChatGPT's memory is what turns the tool from an amnesiac interlocutor into a coworker who remembers your last exchange. It stores preferences and facts you confirm or tell it directly.
You can manage it in your settings, under "Personalization > Memory". There you'll find a summary of what ChatGPT remembers. You can correct a piece of info, delete a specific entry that no longer fits, or wipe all memory at once if you want to start fresh.
I use this feature for two things. First, so ChatGPT knows I prefer short, structured answers. Second, so it remembers the broad strokes of my current projects. That way, when I ask it to review a text or analyze an idea, it already has context.
That said, this memory also affects the AI's context window. Meaning the space available to process your request. If you pile up too many useless memories, you reduce room for the useful content in your conversation. Which is why it pays to clean house regularly.
Claude works differently, but the principle stays the same: give it a permanent brief. "Instructions for Claude", previously called "Profile Preferences", are a system prompt set at the account level. It's factored into every new conversation. It complements project instructions and styles you can choose per chat.
These "Styles" are presets for tone, voice, vocabulary, and level of detail. You can generate them from a writing sample or a simple description. It's especially useful if you want Claude to adopt a register close to yours when drafting your emails or articles.
If you don't know the tool well, my article on what Claude is will give you the basics. Also note that Claude's system prompts evolve regularly. The official changelog runs from Claude 3 in July 2024 to Claude Opus 4.7 in April 2026. At the date of this article, the settings you configure remain valid from version to version, but it's worth checking that everything still works as expected after a major update.
Don't confuse account instructions and per-chat styles. The former apply everywhere. The latter are local to one conversation. If you want Claude to know you across all your projects, write it in "Instructions for Claude", not just in the current style.
Here's the serious part. Customizing your AI is good. Putting sensitive info in it is dangerous. And many entrepreneurs do it without knowing.
According to a Cyberhaven report from 2025, 34.8% of company data employees enter into AI tools is now sensitive. It was only 10.7% two years earlier. The risk is skyrocketing.
Why? Because on consumer ChatGPT accounts (Free, Plus, Go), your data can be stored, reviewed by humans, and used to train future models. Unless you explicitly disable this option in your settings. With Claude, the situation is similar for consumer accounts (Free, Pro, Max). Since September 2025, your conversations can be used for training if you leave the "You can help improve Claude" option on. It is on by default.
So what do I never put in my instructions or memory? Personally identifiable client data, exact revenue figures, non-public strategies, passwords, access codes. Nothing that could hurt me if it left the bubble.
If you're on pro plans, the game changes. ChatGPT Business, Team, and Enterprise encrypt data and don't use it for training by default. Admins have privacy and retention controls. On the Claude side, Team, Enterprise, and API plans don't use your data for training by default. And there's even a "Zero Data Retention" mode for sensitive industries.
But watch out. According to Cyberhaven Labs in 2026, 32.3% of ChatGPT usage happens through personal accounts, even when company workspaces are available. That means a good chunk of leaks comes from a simple login mistake.
I recommend reading my article on privacy issues with personal data and AI to dig deeper.
You're allowed to change your mind. Or realize you've shared too much. Here's how to take back control.
To clear ChatGPT's custom instructions, go to your settings. Disable the "Enable personalization" option. If you want, you can also delete the text from the corresponding fields. It's immediate.
For memory, it's in the same menu "Personalization > Memory". You can review the summary of what ChatGPT remembers. You have options: correct a piece of info, delete a specific entry that bothers you, or wipe all memory at once. I do a little cleanup every month. It takes two minutes and saves me from piling up obsolete memories.
With Claude, the logic is similar for account instructions. Go back to your preferences, edit or delete the text. For per-chat styles, you simply change the setting in the active conversation.
The most important thing is not to let this data linger if you switch computers, leave a job, or just want to test a new setting without baggage from the past.
Custom instructions are directives you write yourself. They stay fixed and the model applies them to every response. Memory builds up over conversations. It keeps details and preferences ChatGPT deems useful. It's a living summary, while instructions are a static brief.
On consumer ChatGPT accounts (Free, Plus, Go), your data can be stored and used for future training unless you disable the option. Same for Claude on Free, Pro, and Max accounts since September 2025, with the "You can help improve Claude" option on by default. On pro plans for both tools, your data is not used for training by default.
Yes. ChatGPT custom instructions are available on every plan and every platform since at least October 2026. With Claude, "Instructions for Claude" and per-chat styles are accessible without a paid subscription.
Disable the "You can help improve Claude" option in your settings if you're on a consumer account. If you handle sensitive data, move to Team or Enterprise plans, or use the API with appropriate retention options. And above all, never include confidential data in your account instructions or conversations.
You can do both. In the "Personalization > Memory" settings, you can delete specific entries one by one or wipe all memory in one go.
"Instructions for Claude" apply at the account level. So they're active in every new conversation. They complement project-specific instructions you can set per project and styles you choose per chat.
Customizing ChatGPT and Claude means saving time every day. But it's also a responsibility. I keep my instructions light and professional. I let memory catch the usage details. And I'm careful to never slip sensitive data in there, especially on a consumer account.
Take ten minutes tonight to check your settings. Read what your AIs remember about you. Correct, delete, adjust. It's a simple move that changes the quality of your answers. And protects you.