ChatGPT now remembers your preferences across chats

Every time you open a new ChatGPT window, you basically have to reintroduce yourself from scratch. Tell it what you do, how you like your answers, and what tone to use.

OpenAI is finally fixing that friction.

OpenAI rolled out Custom Instructions in beta. It gives you a way to set standing preferences that ChatGPT will remember across all future chats, so you only have to say it once.

The feature is available starting today for ChatGPT Plus subscribers and will expand to all users in the coming weeks. If you are in the UK or the EU, you will have to wait a bit longer due to local availability rollouts.

Set it once, keep it forever

OpenAI said the update came out of conversations with users across 22 countries, aiming to boost the steerability of the model.

The use cases are dead simple. If you are a third-grade science teacher, you do not have to keep typing your grade level every time you build a lesson plan. If you are a developer who writes anything but Python, tell it once, and it stops defaulting to Python.

It even works with plugins. If you mention where you live in your instructions, a restaurant reservation plugin can automatically use that location without you having to spell it out every run.

Turning it on is simple enough. On the web, click your name, go to Settings, find Beta features, and toggle Custom instructions. On iOS, you will find it under Settings and New Features.

OpenAI notes that during the beta, ChatGPT might occasionally overlook your rules or apply them when you did not mean to. On safety, instructions pass through the company's Moderation API, and OpenAI may use them to train models unless you opt out in your data controls.

Why it matters

We think this is one of the most practical usability upgrades OpenAI has shipped since plugins.

For power users, developers, and educators, it removes the annoying copy-paste ritual of pasting system prompts at the start of every session. It turns ChatGPT from an amnesic chatbot into an assistant that actually retains some working context about who you are.

Is it worth upgrading to Plus just for this? If you use ChatGPT multiple times a day for specialized work, the time saved adds up fast. If you are just asking casual questions, it is better to wait for the free rollout in a few weeks, especially since the beta still occasionally misses instructions.

Strategically, OpenAI is taking a smart, lightweight approach to personalization. Instead of burning expensive context window tokens by stuffing your entire history into every prompt, they created a persistent identity layer that sits alongside the model and plugins.

The main risk will be instruction drift. When a broad custom rule clashes with a hyper-specific task or a third-party plugin prompt, the model could easily get confused or deliver weird outputs. Still, making AI feel personal without making it creepy is the right line to walk.