Most people use ChatGPT like a hotel receptionist with amnesia. Every conversation starts from scratch, you explain all over again that you live in Leeds, drive a diesel estate and can’t stand waffle, and it still answers like a motivational poster.
It doesn’t have to work that way. ChatGPT has two settings that let it keep track of who you are and how you like your answers: memory and custom instructions. Ten minutes setting them up will save you hours of repeating yourself. It’s also worth knowing exactly what you’re handing over before you start.
What is the difference between ChatGPT memory and custom instructions?
They sound alike but do different jobs, and mixing them up is why a lot of people give up on both.
Custom instructions are the standing orders. You write them yourself, they sit in your settings, and ChatGPT reads them before every new conversation. Think of them as the note you’d leave for a temp on their first day: who you are, what you do, and how you want things done.
Memory is more like a colleague who picks things up as you go. When you mention in passing that you’re vegetarian, that your daughter is doing her A-levels or that you’re selling a house in Kent, ChatGPT can save it as a fact and use it later without being reminded. There are two layers. Saved memories are specific facts it has written down, which you can view and delete one at a time. On top of that, ChatGPT can also draw on your past conversations more loosely to get a feel for your interests and habits.
Both live under Settings, in the Personalisation section. Both are available on the free plan, although paying subscribers get a fuller version of the chat-history memory.
How do I set up custom instructions in ChatGPT?
Open Settings, go to Personalisation and look for the custom instructions fields. You’ll be asked what ChatGPT should call you, what you do, anything else it should know about you, and what traits you’d like it to have. There is also a choice of preset styles if you want a quick shortcut, but writing your own is where the real gains are.
Fill these in the way you’d brief a sharp but literal-minded new hire. Vague requests get vague results. “Be concise” helps a bit. “Answer in plain British English, give the short answer first, use UK prices and UK rules, and don’t use bullet points unless I ask” helps far more.
A few lines that tend to earn their keep:
Where you are. “I live in Scotland” matters more than you’d think, because the answers on income tax, prescriptions, property law and plenty else differ from England. Say so and you’ll stop getting advice that only works south of the border.
What you know already. “I’m comfortable with spreadsheets but not programming” stops it either explaining what a cell is or throwing Python at you.
What annoys you. If you hate gushing openers, false enthusiasm or being told your question is “great”, say so. It will mostly listen.
How sure it is. Ask it to tell you when it’s unsure or guessing, and to point you to official sources for anything legal, medical or financial. It won’t be right every time, but it nudges the tone away from confident nonsense.
Keep it short. A tight paragraph of instructions works better than a page of rules that contradict each other.
Teaching it as you go
Memory is the lazier option, and you can be lazy on purpose. The easiest way to use it is to ask outright: “Remember that I’m a self-employed electrician” or “Remember that my car is a 2017 Skoda Octavia.” It will usually confirm that the memory has been updated.
It also works in reverse. “Forget that I’m planning a trip to Portugal” should remove it, and you can check the list of saved memories in Settings to make sure it has. Going through that list now and then is worth doing. ChatGPT sometimes saves odd things, such as a one-off question about haemorrhoid cream that you’d rather it didn’t bring up next time you ask for a dinner party menu.
The more ChatGPT knows about you, the more useful it becomes, and the more you have handed over to a private company. Decide where that line sits before it does.
If you use ChatGPT for a few distinct jobs, such as the business accounts, the allotment and a novel you’ll finish one day, Projects are worth a look. They group chats and files together with their own instructions, so your crime thriller doesn’t end up in the VAT queries.
Safe enough for ordinary preferences, if you’re sensible about it. Telling it you’re 58, like your answers short and support Aston Villa does no harm. Telling it your National Insurance number, bank details, passwords or the full details of a medical condition is a different matter, and the simple rule is: don’t.
There are three controls you should know about. First, you can switch memory and chat-history referencing off entirely in Personalisation, or leave them on and delete individual items. Second, Temporary Chat gives you a clean conversation that doesn’t use your memories or create new ones and isn’t used to train OpenAI’s models, which makes it handy for the one-off awkward question. Third, under Data controls there is a setting that decides whether your conversations can be used to improve the models. It is on by default for personal accounts. Turn it off if that bothers you.
Deleting a chat doesn’t automatically delete anything memory has saved from it. You need to clear the memory separately. It’s an easy thing to miss.
Because OpenAI processes personal data from UK users, it has to comply with UK GDPR. That gives you the right to see what data is held about you and to ask for it to be deleted. If a request goes nowhere, the Information Commissioner’s Office is the regulator to complain to.
Is it worth the bother?
For most people, yes, and it’s free. Custom instructions alone turn ChatGPT from a well-read stranger into something closer to an assistant who knows the house rules. Memory adds the details that save you typing the same background every time.
The catch is that personalisation makes ChatGPT more persuasive as well as more useful. An answer that knows your town, your car and your tone of voice feels more trustworthy, but it isn’t any more accurate. It can still invent facts, misquote rules and get dates wrong. For tax, health, legal or pension questions, use it to get your bearings, then check what it says against GOV.UK, HMRC, the NHS or a qualified professional.
Set it up once, look over what it has saved every couple of months, and keep anything you wouldn’t put on a postcard out of it.
In short
To get ChatGPT to understand you, fill in custom instructions in Settings under Personalisation with who you are, where you live and how you want answers written, then let memory pick up the details or tell it directly what to remember. Check and prune your saved memories regularly, use Temporary Chat for one-off private questions, and switch off model training under Data controls if you’d rather your chats weren’t used. Never share passwords, bank details or anything you couldn’t stand to see leaked.
Sources
Midcent Tech
Alex Innes explains technology for men 45 and over who have seen every gadget since the ZX Spectrum and are not easily impressed. AI and ChatGPT, phones, cars, the smart home, the subscriptions worth keeping and staying safe online – what it does, whether it's worth the money, and whether you need it at all.
One of Midcent's writers. Spotted something? Write to alex@midcent.uk.
