🤖 Part 19: Custom Instructions, Memory & Personalization

By now you've probably re-explained yourself a dozen times — "keep it short," "I'm a nurse," "write in American English." You shouldn't have to. Most assistants let you say that once and have it shape every reply, and some now quietly remember facts between chats. This part shows you how to set that up, how to see and clean out what's stored, and when personalization helps versus when it subtly steers you wrong.

📚 What You'll Learn

By the end of this part, you'll be able to:

Write custom instructions that tell the assistant about you and your preferences once, so every reply is tailored
Understand what memory features do — what they store, and how to view, edit, and clear them (names and rules vary by product and plan)
Set a sensible default tone and format without over-tuning
Spot when personalization helps versus when it quietly skews answers — and manage the privacy side

In This Part

Say It Once: Custom Instructions

Every fresh chat starts the assistant with a blank slate. It doesn't know who you are, what you do, or how you like answers unless you tell it — which is why, without any setup, you find yourself typing "keep this brief and skip the preamble" over and over. Custom instructions fix that. They're a small profile you fill in once, in settings, and the assistant reads them silently at the start of every conversation.

📖 Definition

Custom instructions (a.k.a. profile, preferences, or "about you"): a saved note you write once that the assistant applies to every new chat. It usually has two halves — who you are and what you're working on, and how you'd like it to respond. Think of it as briefing a new colleague on your first day so you never have to repeat the briefing.

The exact name and location differ by product, and both change often, so treat the labels below as a map, not gospel — the idea is the same everywhere even when the wording moves.

Assistant	What it's often called	Roughly where to find it
ChatGPT	Custom instructions / Personalization	Settings → Personalization
Claude	Personal preferences / "What should Claude know about you?"	Settings → Profile / Preferences
Gemini	Personalization / saved info (tied to your Google account)	Settings, or via your Google account controls

⚠️ Important: Feature names, locations, and even whether a given control exists on your plan change constantly. If a menu doesn't match what's written here, don't assume you're doing it wrong — open the assistant's settings, skim for anything like "personalization," "preferences," or "about you," and check the product's own help page. Throughout this part, take every specific name as "this is roughly what to look for," not "this is the exact button today."

What makes custom instructions worth five minutes is leverage: one good paragraph improves thousands of future replies. Tell it you're a small-business owner who wants plain language and no jargon, and every summary, email, and explanation from then on arrives shaped to you — without another word from you.

Memory: When the Assistant Remembers

Custom instructions are something you write. Memory is different: it's the assistant deciding, on its own, to hold onto a useful fact from a conversation and reuse it later — that you have two kids, that you prefer metric units, that you're learning Spanish. The next time it's relevant, it just knows, even in a brand-new chat.

📖 Definition

Memory: a feature where the assistant saves selected facts across separate conversations and draws on them later, so you don't have to re-establish context each time. It's distinct from a single long chat "remembering" what you said earlier in that same chat — memory persists between chats.

This is one of the fastest-moving, most plan-dependent areas in the whole guide, so here is the honest picture: whether memory exists, what it's called, how much it stores, whether it's on by default, and whether it's free all vary by product and tier and shift often. Some assistants have a mature memory system; some have a lighter "saved info" version; some tie it to your broader account. The right move is never to assume — it's to open settings and look.

graph TD A["💬 You chat normally"] --> B["🧠 Assistant notes a
reusable fact ⟨if memory is on⟩"] B --> C["🗂️ Saved to your memory
e.g. 'prefers short replies'"] C --> D["🔄 A new, unrelated chat"] D --> E["✨ It applies the fact
without you repeating it"] C --> F["👁️ You can view, edit,
or delete any entry"]

The single most important habit with memory is knowing it's there and looking at what it holds. Good memory systems let you see the list of what's been saved, edit or delete individual items, turn the feature off, and often tell it to remember or forget something on the spot ("remember that I work in healthcare," or "forget what I said about my address").

✅ Tip — talk to memory directly

On assistants that support it, you can manage memory in plain language mid-chat: "What do you remember about me?" often prints the list, "Remember that I prefer bullet points" adds an entry, and "Forget my old job title" removes one. It's the quickest way to keep the store accurate — but always confirm with the actual memory settings screen, since the chat's summary isn't always the complete record.

⚠️ Watch Out — memory can carry over things you didn't mean it to

Because memory saves quietly, it can hold a one-off detail you'd rather it dropped — a passing mention of a health issue, a client's name, a temporary project. It may then surface that context in a later, unrelated chat, or shape an answer in a way you don't expect. This isn't sinister; it's the feature doing its job. But it's exactly why you should glance at the memory list now and then and prune anything that shouldn't linger.

Setting a Default Tone & Format

The highest-value thing to put in custom instructions is usually how you want replies, not just who you are. A few lines about tone and format save you the most repeated typing and make the assistant feel like it "gets" you from the first message.

Preference	What to say in your instructions
Length	"Default to concise answers; expand only when I ask for detail."
Tone	"Warm and plain-spoken; skip corporate jargon and hype."
Format	"Use short paragraphs and bullet points; avoid long walls of text."
Language / spelling	"American English spelling and everyday vocabulary."
Honesty	"If you're unsure or it's outside your knowledge, say so rather than guessing."

That last line is worth its weight in gold and echoes the golden rule from Part 1: you can literally instruct the assistant to flag its own uncertainty. It won't make it perfect — it can still be confidently wrong — but it nudges it toward telling you where to check.

✅ Tip — start small, then adjust

Don't try to write the perfect profile in one sitting. Put in three or four preferences you're sure about, use the assistant for a few days, and add a line each time you catch yourself giving the same correction twice. A profile that grows from real friction beats a long one you guessed at up front — and it stays shorter, which matters for the reason in the next section.

When Personalization Helps — and When It Skews

Personalization is genuinely useful, but it's not free of trade-offs. Because your instructions and memory shape every answer, they can quietly narrow what you get — and you won't always notice, since the reply still looks complete and confident.

✅ Where personalization helps	⚠️ Where it can quietly skew
Getting the right tone and length without asking every time	Reinforcing your existing view when you actually want a challenge
Skipping context you'd otherwise re-type ("I'm a beginner")	Over-simplifying because it "remembers" you as a beginner on a topic you've since learned
Consistent formatting across all your work	A stale fact in memory steering answers with outdated context
Faster, less repetitive conversations	Narrowing options because it's optimizing for "you" rather than the full picture

⚠️ Important: When you want a genuinely neutral or second opinion — a fresh take, a devil's-advocate view, an answer not colored by what it thinks you want — use a mode that ignores your profile. Many assistants offer a temporary or incognito-style chat that skips custom instructions and memory. Reaching for it is the antidote to the quiet-skew problem: personalize for daily convenience, go temporary when you want an unbiased look.

There's also a privacy dimension, and it connects straight back to Part 13. Memory stores things — by design. That's fine for "prefers metric units" and a poor idea for anything sensitive. Treat your memory list the way you'd treat a notebook someone else might read: don't let it accumulate personal health details, financial specifics, other people's private information, or passwords. Review it periodically and clear what doesn't need to be there.

⚠️ Watch Out — shared devices and work accounts

If you sign in on a shared or family computer, your custom instructions and memory travel with your account, not the machine — but anyone using your logged-in session sees a personalized assistant that knows your details. And on a work or school account, an administrator may control or disable these features entirely. When in doubt about who can see what, keep sensitive facts out of memory and out of the chat.

🛠️ How To: Set Up Your Custom Instructions

What you'll do: write a short, honest profile so every future reply is tailored to you — and confirm what, if anything, the assistant is remembering.

Step by step

Open settings and find personalization. Look for "Custom instructions," "Personalization," "Preferences," or "About you." If the wording differs, that's expected — skim for the closest match, or check the product's help page.
Fill in "about you" briefly. One or two lines: your role or situation and what you mostly use the assistant for. Example: "I run a small bakery and use this for emails, social posts, and plain-English explanations." Keep out anything sensitive.
Fill in "how to respond." Add three or four preferences — length, tone, format, spelling, and a line asking it to flag uncertainty (see the table above). Short and clear beats long and vague.
Save, then test it. Start a fresh chat and ask something ordinary. Does the reply match your preferences? Adjust the wording of any instruction it didn't follow well and try again.
Check your memory list. If the assistant has a memory feature, open its memory screen (or ask "What do you remember about me?"). Read what's there, delete anything that shouldn't be, and decide whether you want the feature on, off, or on-with-review.

💡 Tip

Keep a copy of your custom-instructions text in a notes app. Products reset, plans change, and you may want the same profile across ChatGPT, Claude, and Gemini. Having it saved means you paste it into each one in seconds instead of rewriting it from memory — and you can tweak one master version over time.

Best Practices

✅ Do's

Write instructions the way you'd brief a new colleague — who you are, and how you like the work delivered.
Review your memory list now and then and prune anything stale, sensitive, or no longer true.
Ask it to flag uncertainty right in your profile — it's a one-line safety net you get on every reply.
Use a temporary / incognito chat when you want a neutral answer untouched by your profile or memory.

❌ Don'ts

Don't put sensitive details in instructions or let them pile up in memory — health, finances, passwords, or other people's private info.
Don't assume the feature works the same everywhere — names, availability, and plans vary and change often; check the official page.
Don't over-tune. A giant, contradictory profile confuses more than it helps; keep it short and edit from real friction.
Don't forget it's on. If answers feel oddly narrow, your own personalization may be the cause — check it before blaming the model.

💡 Pro Tips

Keep two profiles in your notes — a "work" version and a "personal" version — and paste whichever fits when you switch contexts.
Every few weeks, ask "What do you remember about me, and what are my current custom instructions?" It's a fast audit that catches drift and stale facts.

📓 Learning Journal

Keep a journal as you work through this guide — digital or paper. After each part, jot down:

Key ideas you learned
Things that clicked for you
Questions or confusion points to revisit
Ideas you want to try
Your progress and how you feel about it

✍️ This part's prompt: Draft the custom-instructions profile you'd actually use. Write two or three lines about who you are and what you use an assistant for, then three or four lines about how you want replies (length, tone, format, and a request to flag uncertainty). Now read it back with a skeptic's eye: is there anything in there you wouldn't want stored or surfaced in a later chat? Note what you'd keep out of memory, and whether you'd rather have memory on, off, or on-with-review.

📝 Part Summary

🎓 Key Takeaways

Custom instructions let you say who you are and how you like replies once, and the assistant applies it to every new chat.
Memory is the assistant keeping useful facts between chats — powerful, but you should view, edit, and clear the list, and know it can be off, on, or on-with-review.
A few lines on tone and format — plus "flag your uncertainty" — deliver the most value; keep the profile short and grow it from real friction.
Personalization helps daily work but can quietly skew answers; use a temporary chat for neutral takes, and keep sensitive details out of memory (see Part 13).

🎉 What You Can Now Do

You can set an assistant up so it stops making you repeat yourself — tailored tone, length, and format from the first message — and you know the difference between what you deliberately tell it and what it quietly remembers. Just as importantly, you can look inside its memory, prune it, and reach for a neutral mode when you want an answer that isn't shaped by your own profile. That's personalization with the lights on.

❓ Common Questions

Is memory the same as the assistant remembering earlier in a long chat?

No. Within one conversation, the assistant naturally has the earlier messages in view — that's just context. Memory is the separate feature that carries facts across different chats. You can have context without memory, and you can turn memory off while long single chats still work normally.

Do I have to pay for custom instructions or memory?

It varies and changes. Custom instructions are often available on free tiers; memory's availability and limits depend more on product and plan, and both move over time. Don't take any specific claim as permanent — open the assistant's settings and its official help page to see what your account actually has today.

How do I completely wipe what an assistant knows about me?

Look in settings for the memory screen and clear it there, and separately clear or blank your custom instructions. Some products also let you delete individual chats and manage data retention (Part 13). If you're unsure a chat-level "forget" fully removed something, confirm on the actual settings screen rather than trusting the in-chat reply.

🔭 Up Next

In Part 20: Reusable Assistants — Projects, Custom GPTs & Gems, we'll go one step further than a personal profile: packaging a whole setup — standing instructions plus reference files — into a reusable assistant you can open, share, and reuse for a task you do again and again.

📚 Additional Resources

🌟 Keep Going

You've taught the assistant who you are without lifting a finger on every future chat — and you know how to keep it honest and tidy. That's the quiet upgrade that makes daily use feel effortless. Next, let's turn a setup you reuse into something you can build once and open forever. 🤖