Which tool to choose
Look for two things in your assistant. The first is a personal instructions section (often called instructions or customization): a field where you describe who you are and what you prefer, and that applies to all chats. The second is automatic memory, where the AI saves a few facts that come up as you talk. The first you control and is the most reliable; the second is handy but opaque. If you find neither, no problem: you can paste your preferences at the start of the conversation.
How to do it
- Open your assistant's personal instructions, in the profile settings.
- Write your tastes concretely: what you love, what you can't stand, practical constraints (for example allergies, budget, the tone you prefer).
- Add how you want the answers: short or detailed, with examples or without, in which language.
- Save and try with a normal request: you'll see the answer arrives already tailored, without having to explain it again.
A concrete example
Marco is vegetarian, hates cilantro and cooks for two in half an hour in the evening. Tired of repeating it every time he asked for a recipe, he wrote all this in his personal instructions: no meat or fish, never cilantro, recipes for two portions ready in thirty minutes. Since then, every time he asks "give me an idea for dinner" the suggestion is already right on the first try. He didn't have to teach it to the AI conversation after conversation: he wrote it once, in the right place.
When it does NOT work (and how to fix it)
If the AI seems to ignore a preference
Often it's written too vaguely. "I like to eat healthy" says little; "no fried food, little sugar, lots of vegetables" is an instruction the AI can follow. Rewrite the preference so it's an actionable indication, not a generic wish.
If your tastes change
Preferences aren't carved in stone. When something changes, go back to the instructions and update them: remove what no longer applies, add the new. It's a field you can change at any time, and it's right to keep it alive.
If the assistant has no field for preferences
Keep your preferences in a short document of your own and paste it at the start of the chat when you need it. It works everywhere and doesn't depend on the features of the individual platform.
A tip from someone who really uses it
Distinguish between stable tastes and tastes of the moment. In the personal instructions put only what applies almost always — allergies, values, the tone you prefer — because that has to hold in every chat. Passing whims ("today I feel like something spicy") say directly in the conversation, where they're needed and nowhere else. If you fill the instructions with temporary details, they become a messy warehouse and the AI struggles to figure out what really matters.
Frequently asked questions
Where do the preferences I write end up?
They stay in your assistant profile and apply to subsequent conversations, until you change them. They're different from automatic memory: the instructions you write and control yourself, word for word, while automatic memory is what the AI decides to note down on its own as you talk.
Can I have different preferences for different uses?
Yes, if the assistant lets you create separate assistants or spaces: one for cooking, one for work, each with its own instructions. Where this isn't available, it's best to keep general instructions and specify the context in the individual request.
What if the interface changes and I can't find where to write the preferences?
Assistant screens change often, so the exact button may move. The way that always works is to ask the AI itself: write in the box "from now on bear in mind I'm vegetarian and I want short answers" and carry on. To make it stable, then look in the settings for the instructions or customization item.
Is it enough to use it a lot and the AI will learn my tastes on its own?
No, and it's the idea that wastes the most time. Automatic memory gathers details in a fragmentary and unpredictable way: sometimes it notes something said in passing and misses the important one. You don't know what it has stored until it gets something wrong. Writing the preferences yourself, once, in a field you control, gives an immediate and predictable result. Waiting for it to learn them "by dint of using it" means leaving your tastes to chance.