Make it your diplomatic PA

The situation

You ask for a quick text to the swimming coach and get five paragraphs, three bullet points about “effective communication” and a closing line about self-care. So you reply “shorter”, then “warmer”, then “no lecture please”. Every single time. You can say it once, in settings, and it sticks.

By the end of this lesson you’ll understand how personal instructions work, what makes an instruction actually get followed, and you’ll have a tested set saved in your app.

Why it’s worth it

AI apps are built for everyone at once: students writing essays, programmers, retirees researching holidays. So their default style is a careful, thorough, slightly formal middle ground: the encyclopedia voice. It’s not wrong, it’s just not you. For family admin you usually want the opposite: short, warm, ready to send, and no unrequested advice.

You can fix it in each chat, as in Lesson 1.1 (“Shorter. Bullet points only.”), but that’s one more thing to remember every time, and remembering things is exactly the load you’re trying to shed. Personal instructions let you brief the app once, the way you’d brief a new assistant on their first day: who you are, how you like things done, and what drives you up the wall.

How personal instructions work

When you save personal instructions, the app quietly adds them to your conversations, as if you’d pasted them in at the top before typing anything. The AI reads them every time and shapes its answers around them. Nothing about the AI itself changes; it hasn’t “learned” you. It’s simply reading your briefing note before it starts.

Two things follow from that. First, instructions are guidance, not locks. The AI follows them most of the time, but in a long chat, or when your request pulls strongly the other way, it can drift back to its default habits. A quick “remember: short, no tips” in the chat pulls it back. Second, everything in your instructions travels into every conversation you have, which is why the template below gives context (ages, routine, country) but nothing that identifies your family.

You may also notice a memory feature in your app’s settings. That’s different: memory is the app picking up details from your chats and carrying them forward on its own. Instructions are what you deliberately write. This lesson is about instructions, because you control exactly what’s in them.

What makes an instruction work

The AI is a very literal reader. The instructions that work best are ones where it’s obvious whether they were followed:

• Specific beats vague. “Keep messages under 80 words” works; “be concise” means something different every time.
• Say what to do, not just what not to do. “Give me a ready-to-send draft” works better than “don’t write essays”, because it describes the result you want.
• One idea per line. A short list of clear rules is easier to follow than a paragraph of preferences.
• Context and style are separate. “About me” tells it your situation; “How to answer” tells it how to behave. Keeping them apart stops it from mixing your routine into every reply.

There’s one more line worth including in everyone’s instructions: asking it to say when it isn’t sure. It won’t stop confident guessing completely, but it noticeably increases how often the app says “I’m not certain about this date” instead of inventing one.

Where the setting lives

Apps rename and move their menus often, so if a name below doesn’t match exactly, look for the closest thing, or search the app’s help pages for “instructions”.

• ChatGPT: Settings → Personalization → Custom instructions (some versions of the phone app call it “Customize ChatGPT”). Check that customisation is switched on. On the free plan you can save up to 1,500 characters; the template below is about half that.
• Claude: Settings → look for Instructions (under “Your Claude” or “Account”, depending on your version).
• Gemini: Settings → Personal Intelligence → Instructions for Gemini → Add. You need a personal Google account; work and school accounts don’t have this.

Copy this

Copy this, replace the [square brackets], and paste it into the instructions box. Then save.

About me: I'm a mum of [number] kids, aged [ages]. I juggle [e.g. part-time work from home], school runs and activities. I live in [country], so use its date format and spelling.

How to answer:
- Be my calm, tactful personal assistant, not an encyclopedia.
- Keep it short. Give me the answer or the draft first, then stop.
- When I ask for a message, write a ready-to-send draft: warm, polite and confident. No grovelling, no over-explaining.
- Use plain words, and bullet points for plans and lists.
- Don't lecture me or add wellbeing tips unless I ask.
- If something important is missing (a date, a name, who it's for), ask me up to 2 short questions first.
- If you're not sure about a date, time, rule or fact, say so. Never make one up.

Why each part is there

“About me” gives it the context it would otherwise guess. Knowing the kids’ ages changes the advice (a 4-year-old’s bedtime isn’t a 14-year-old’s), and your country fixes date formats (is 03/04 March or April?) and spelling.

“Calm, tactful personal assistant, not an encyclopedia” names the role. AI apps respond strongly to a role: it sets the tone for everything else in one line.

“Give me the answer or the draft first, then stop” fights the most common annoyance: three paragraphs of preamble before the useful bit, and a summary after it.

“No grovelling, no over-explaining” matters for messages to schools, coaches and other parents. AI drafts tend to apologise too much, and over-apologising makes you sound less sure of yourself.

“Don’t lecture me or add wellbeing tips” switches off the well-meant advice AI apps add by default. You can always ask for it.

“Ask me up to 2 short questions” gives it permission to check rather than guess, and the limit stops it from interviewing you.

“If you’re not sure, say so. Never make one up.” is your standing guard against confident guessing.

What you’ll see: the app saves the text. Nothing else changes until you start a chat.

What you’ll get

Start a new chat and ask for something you need anyway. Here’s the same request before and after (examples: your app’s wording will differ).

You ask: Text the swimming coach that Ava will miss Tuesday, she’s got a cold.

Before instructions:

Here’s a message you could send to the swimming coach. It’s important to communicate clearly and give enough notice so the coach can plan the session. You might also want to mention when you expect Ava to return:

“Dear Coach, I hope this message finds you well. I am writing to let you know that unfortunately Ava will be unable to attend Tuesday’s swimming session as she has come down with a cold. We apologise for any inconvenience this may cause…”

Remember to look after yourself too, as caring for a sick child can be tiring!

After instructions:

Hi Coach, just letting you know Ava will miss Tuesday, she’s got a cold. Hoping she’s back for Thursday. Thanks!

Notice what changed, line by line. The preamble (“Here’s a message you could send…”) is gone, because the instructions said draft first, then stop. The formal “I hope this message finds you well” became a normal text, because you asked for warm and plain. The apology disappeared, because you said no grovelling. And the self-care line is gone. Same request, a fraction of the words, and it sounds like a person.

Before you rely on it

• Test it on something real. Ask for one message you actually need this week. Does it sound like you? If not, change one line and test again. Changing one thing at a time tells you which line made the difference.
• Read your “About me” once more. Ages, routine and country only. No surnames, schools or addresses: remember, it travels into every chat.
• Check it’s switched on. In ChatGPT, the instructions only work while customisation is on.

Make it yours

• Teach it your voice. Add: Here's how I write: [paste two or three messages you've sent before, with names removed]. Match this style. Real examples work better than describing your style, because the AI can copy the rhythm and word choices directly.
• Family languages. Add: I often need messages for family in [language]. Offer a translation when I ask. Useful when grandparents or in-laws prefer another language.
• One-off change. Instructions are the default, not a cage. In any chat you can say For this chat only, be more formal. Good for a letter to a landlord or an official complaint.

If it goes sideways

• It ignores your instructions. Some apps only pick up changes in new chats, so start a fresh one. Then check the setting saved and is switched on.
• Now it’s too blunt. “Keep it short” can tip into curt. Change it to “Keep it short, but warm.”
• It won’t let you save. You’ve probably hit the length limit. Cut the longest line, or drop the examples from “Teach it your voice”.
• It slips back into tips or long answers mid-chat. That’s the drift described above: instructions are guidance, and long chats weaken them. Say “No tips, just the draft” and it corrects for the rest of the chat.