Teach it about you once, in settings

If you've used your AI tool (ChatGPT, Claude, Gemini, Copilot — whichever you picked) a few times, you've probably noticed the annoying part: every new chat, it has no idea who you are. You explain your business again. You say "make it less formal" again. You paste the same background again. It's like briefing a temp who's very smart and has no memory of yesterday.

There's a fix, it lives in settings, and almost nobody turns it on. Twenty minutes here means every conversation for the next year starts with the AI already knowing what you do, who you're talking to, and how you like things written. This is the highest-return twenty minutes in this whole track, and it costs nothing — the settings box is on the free plan in all four of the big tools, and if your tool is missing a piece, there's a free workaround below.

Two kinds of feature do the work, and they're different:

• Standing instructions — a short brief you write yourself. Stable facts and preferences. You control it completely. Every tool calls it something different: custom instructions, instructions for Claude, saved info. Same idea.
• Memory — things the AI notices and saves on its own as you chat. Convenient, occasionally wrong, and worth checking. All four tools have a version now, though what it's called and how much it keeps varies.

We'll set up the first one properly, then get the second one under control.

Step 1: Find the box

It's always in Settings, and in every tool you get there the same way: click your name, initials or profile picture (usually in a corner of the screen) and choose Settings. From there, as of this writing:

• ChatGPT: Settings → Personalization → Custom instructions. On the phone app, tap your name and look for Customize ChatGPT instead. You'll get two boxes — one for what it should know about you, one for how it should respond.
• Claude: click your initials (bottom left) → Settings → Profile. The box is called Instructions for Claude (older accounts may still see it worded as what personal preferences should Claude consider). One box.
• Gemini: the gear icon in the left sidebar → Saved info (some regions label it Instructions for Gemini). It's a list of short entries rather than one box, and you can add as many as you like.
• Copilot: your profile icon → Settings → Personalization. Look for a Custom instructions box next to the memory controls. If your version of Copilot doesn't show one, use the workaround at the end of Step 2.

These companies move menus around a few times a year, so if a name above doesn't match your screen, don't panic. Look for anything called Personalization, Instructions, Saved info or Profile, or just ask the tool itself: "where do I set standing instructions about myself in this app?" It can usually point you there.

Step 2: Write your standing brief

Open the box. Here's what earns its space.

What to include:

• What you do, in one line. Not your mission statement — your job.
• Who you're usually writing to. Customers? Suppliers? Yourself?
• Your tone, described concretely. "Warm but efficient" beats "professional."
• Recurring facts it would otherwise ask for every time — your services, your general pricing shape, your city if that matters to your work.
• Where facts should come from, if that matters to your work.
• One or two hard rules. Things you never want to see.

Here's a version you can adapt. Change the details, keep the shape:

I run a small service business. Most of what I write goes to customers and prospective customers — quotes, follow-ups, scheduling, social posts.

Tone: warm, direct, plain English. Short sentences. No corporate filler, no "we are thrilled to announce," no exclamation points unless I use one first.

Useful background: I handle everything myself, so drafts need to be close to sendable rather than a starting point I have to rewrite.

Sources: when you give me a fact, use primary or authoritative ones — the company's own site, official documentation, the agency that publishes the thing. Don't rely on crowd-edited or aggregator pages like Wikipedia, and tell me where the answer came from so I can check it.

Never: invent prices, invent policies, or make promises about timelines on my behalf. If you need a detail like that, ask me for it instead of guessing.

Where it goes, tool by tool: in ChatGPT, the first two lines go in the "what should it know about you" box and the rest in the "how should it respond" box. In Claude and Copilot, paste the whole thing into the one box. In Gemini, paste it as one saved-info entry, or split it into a few entries if it prefers short ones — either works.

Two of those lines are doing more work than they look like they are.

The sources line matters because an AI tool will happily answer from whatever it found, and a crowd-edited page is fine for casual curiosity and a bad footing for something you're about to tell a customer. Asking it to name where the answer came from is what makes checking possible — an unsourced claim and a sourced one look identical in the chat window.

The never line is the one people skip and shouldn't. A standing "ask me instead of guessing" instruction is the cheapest protection you can give yourself against a confident, wrong, sendable-looking draft — the failure mode we talked about in lesson one.

Save it, then open a brand-new chat and ask for something small — "draft a two-line reply telling a customer we're booked until the 14th." If the voice comes back closer to yours than it used to, your brief is working. If it comes back stiff, your tone line is too abstract. Go back and get more specific about what you don't want.

No settings box for this in your tool? Save the brief in a note and paste it in as the first message of any chat where it matters. One extra paste, works in every tool, free.

Step 3: Get memory under control

Memory is the other half, and it works differently: you don't write it, the AI does. It quietly saves things it thinks are worth remembering as you chat.

When it works, it's genuinely nice — it remembers your business name, your regular customers' situations, what you're working on. When it doesn't, it remembers something you said once in passing as though it were a permanent fact about your life.

Where the switch and the saved list live, as of this writing:

• ChatGPT: Settings → Personalization → Memory. A toggle, plus a Manage memories button that shows the list.
• Claude: Settings → Capabilities → Memory. A toggle, plus a view where you can read and edit what it has kept.
• Gemini: Settings → Personal context → Your past chats with Gemini is the switch. The things it has explicitly saved about you live back in Saved info, where you can edit or delete them directly.
• Copilot: your profile icon → your profile name → Memory → Personalization and memory. The switch and the saved list are together.

Do this now:

1. Turn it on if it's off.
2. Read the saved list. This takes two minutes and is the part nobody does.
3. Delete anything wrong, stale, or none of its business. Deleting is harmless.

You can also manage memory in conversation, which is easier than digging through settings:

Remember that my business is closed on Mondays.

Forget what you saved about my pricing — it's changed.

What do you currently remember about me?

That last one is worth running every couple of months — the fastest audit there is. If your tool says it doesn't keep memories between chats, your brief from Step 2 does the job instead.

Step 4: Decide what never goes in

Personalization means putting real information about yourself into a company's product, so be deliberate rather than nervous. My working rule: treat it like a postcard. Useful for anything you'd be comfortable having a stranger read over your shoulder. Not for anything else.

Keep out of both your standing instructions and memory:

• Passwords, and anything that would let someone into an account.
• Full card numbers, bank details, Social Security numbers.
• Customer information they'd expect you to keep private — health details, financial details, anything covered by a professional obligation.
• Anything you'd need to un-say later. Deleting a memory is easy; unsending information is not.

None of that is a reason to avoid these features. It's a reason to fill them with the boring, useful, professional facts that make your drafts better, and nothing else.

Step 5: Use a temporary chat when you don't want it remembered

Some conversations shouldn't join your permanent record. Most tools have a temporary or private chat mode that doesn't save to your history or feed memory:

• ChatGPT: the Temporary chat toggle near the top of a new chat.
• Claude: the small ghost icon in the header of a new chat. You'll see an Incognito chat label and a dark border when it's on. (It doesn't work inside Projects.)
• Gemini: open the menu and, next to New chat, choose Temporary chat. Personal Google accounts only — work and school accounts don't get it.
• Copilot: I couldn't find a temporary mode when I wrote this. Delete the chat from your history when you're done, which does the same job one step later.

Use it for one-off sensitive things: a difficult email about a specific person, a question you'd rather not have shaping every future answer. Then close it and your ordinary setup is untouched.

What you should have now

• A standing brief that means you stop re-explaining your business — in settings, or in a note ready to paste.
• Memory switched on, read through, and cleaned of anything wrong.
• A clear line about what you'll never put in either.
• A private mode, or a delete habit, for the conversations that shouldn't stick.

Do the one thing that makes this real: go start a fresh chat and ask for something you'd normally have to set up with three sentences of background. That saved effort is what you just bought, and it compounds every single time you open the app.

FAQ

What's the difference between memory and standing instructions?

Standing instructions (custom instructions in ChatGPT, instructions for Claude, saved info in Gemini) are a brief you write and control — stable facts about your work and how you want things written. Memory is what the AI saves on its own as you chat — convenient but occasionally wrong. Write the first one carefully; check the second one every so often.

Do I need a paid plan for this?

No. The standing-instructions box is on the free plan in ChatGPT, Claude, Gemini and Copilot, and memory is too in each of them as of this writing, though free plans may keep less of it. If a box is missing in your version, paste the brief in as your first message — free, works everywhere.

Where exactly is the setting in my tool?

ChatGPT: Settings → Personalization → Custom instructions. Claude: Settings → Profile → Instructions for Claude. Gemini: Settings → Saved info. Copilot: Settings → Personalization. Menus move, so if yours doesn't match, look for anything called Personalization, Instructions, Saved info or Profile, or ask the tool where it keeps standing instructions.

How do I see what my AI tool remembers about me?

Two ways. Read the saved list in settings (ChatGPT: Personalization → Memory → Manage memories; Claude: Capabilities → Memory; Gemini: Saved info, plus the Personal context switch; Copilot: Memory → Personalization and memory). Or ask in a chat: "What do you currently remember about me?"

How do I make it forget something?

Delete the entry from the saved list in settings, or say it plainly in a chat — "forget what you saved about my pricing." In tools that don't keep memory between chats, there's nothing to forget; just update the brief you paste in.

What shouldn't I put in my settings?

Treat it like a postcard: fine for anything you'd let a stranger read over your shoulder, not for anything else. No passwords, no card or bank numbers, no private customer details. Boring professional facts are exactly what belongs there.

It's still not writing in my voice. What now?

Your tone line is probably too abstract. "Professional" tells it nothing. Describe what you don't want — no corporate filler, no exclamation points, short sentences — and give one real example of a sentence that sounds like you. Concrete beats flattering every time.