How to Make ChatGPT Remember How You Work

The tiresome part

By your second week something starts to grate. Every new chat, you explain yourself again. That you run a small bakery. That you write in British English. That you loathe exclamation marks. That your emails go to customers, not colleagues.

It is not being difficult. Each new chat genuinely starts blank, in the way a very capable temp starts blank on their first morning. There are two ways to stop doing the induction every time, and they solve different halves of the problem.

Custom instructions: your standing brief

Custom instructions are a small box, sitting in settings, where you write once what you would otherwise say at the top of every chat. Every new conversation then begins with it already known.

Look under your account or profile settings for something named custom instructions or personalisation. The exact wording and location shift as the app is redesigned, so hunt by meaning rather than by remembered position.

Two things go in there. Who you are, and how you want to be answered.

About me: I run a small independent bakery in Leeds with three staff. Most of my writing is to customers, suppliers or my local council. I am not technical.
How to respond: British English. Plain words, short sentences, no exclamation marks and no corporate phrasing. Give me the answer first and the reasoning after, only if I ask. If a question is unclear, ask me rather than guessing.

The single highest value line in most people's instructions is the last one. Left alone, it fills gaps with plausible invention. Told to ask instead, it asks, and you stop finding fictional details in your own emails.

Memory: what it picks up along the way

Memory works differently. Rather than a brief you write, it quietly saves things you mention while talking. You say your daughter is allergic to peanuts, and weeks later a recipe suggestion accounts for it without your asking.

When it is working well this feels like being known. When it is working badly it feels like being haunted by a passing remark from March. Both happen.

The important part, and the reason this lesson exists: you can look at the list. In settings, under personalisation, there is somewhere you can read everything it has saved about you, delete individual items, or clear the lot. Go and read yours once. People are routinely surprised, sometimes pleasantly, sometimes not.

What do you currently remember about me? List it plainly, and tell me which of those things you are drawing on most often.

You can also steer it in the moment. Say "remember that I always want measurements in grams" and it will save that. Say "forget what I told you about my old job" and it will drop it.

Please remember this from now on: my business is a bakery, I write in British English, and I never want exclamation marks in anything you draft for me.

Why this is a privacy control too

Here is the bit that gets skipped. Memory is a convenience feature and a privacy setting at the same time, and it is worth seeing it as both.

Everything memory holds is stored on the company's systems, not on your device. Which means the useful question is not "is it accurate," it is "am I comfortable with this being kept." A note about preferring metric measurements is nothing. A note about your health, your finances, your family or your employer is a different weight of thing entirely.

You are allowed to be selective. Memory can be switched off completely if you would rather it forgot everything the moment you close the tab. You can leave it on and prune the list occasionally. You can keep memory on for ordinary work and use a temporary chat, covered in lesson 6, for anything sensitive.