Custom instructions and memory: set up your AI once

The most under-used feature in mainstream AI is custom instructions. ChatGPT calls them custom instructions. Claude calls them user style preferences. Gemini calls them context. Whatever the name, the idea is the same: a saved set of facts about you and how you want the model to behave that get applied to every conversation, automatically.

Spending ten minutes setting these up is the highest-ROI thing you can do with AI as a beginner. From then on, you stop re-explaining who you are, what you do, and what you want, at the start of every conversation. The model receives that context automatically.

This article walks through how to do it, what to put in, and the most common mistakes.

Custom instructions are for stable, low-risk preferences. Memory is for facts you are comfortable storing. Do not use either one as a place to keep customer data, credentials, confidential project names, or sensitive personal details.

Where to find the settings

In ChatGPT: click your profile icon → Settings → Personalization. There are two relevant features: Custom Instructions (a free-form text area for who you are and how you want responses) and Memory (an automatic system that picks up facts across conversations).

In Claude: click your profile icon → Settings → Profile. You can set a profile description that applies to all conversations. Claude also has Projects, which are project-scoped contexts — more on that below.

In Gemini: the saved-info / "Gemini personalisation" controls live in the Gemini app's settings (and partially in Google Account → Data & privacy). The exact menu names move around — but the panel is there. This is less developed than ChatGPT's version and more focused on facts about you than instructions.

In Copilot (Microsoft 365): the personal-tier version has limited customisation; the enterprise version has admin-controlled context that your IT team sets up.

For the rest of this article we will use ChatGPT's structure as a default, but the same principles apply across all tools.

The two-part structure that works

ChatGPT's custom instructions split into two boxes:

1. "What would you like ChatGPT to know about you?" — the about me section.
2. "How would you like ChatGPT to respond?" — the behaviour section.

This split is useful because the two answer different questions. The first is facts. The second is style.

Part 1: About me

The goal here is to give the model the context it would otherwise have to keep asking you for. Some things worth including:

• Your role and field. "I am a senior product manager at a B2B SaaS company in Tallinn, Estonia."
• Your professional context. "Our customers are mid-market e-commerce companies in northern Europe."
• What you spend time doing. "Most of my AI use is around writing specs, analysing customer interviews, planning roadmaps, and drafting team comms."
• Your background, if relevant. "I studied design and was a designer for five years before switching to product."
• What languages you work in. "I write mostly in English; occasionally in Estonian; my Estonian is good but not native."
• Constraints. "I avoid storing customer data in personal AI tools — please remind me to use anonymous examples."

Keep it to 4–6 sentences. Longer is not better. The model uses these facts to calibrate; it does not benefit from a CV.

A useful template you can adapt:

I am [your role] at [your company / industry] in [location]. My work involves [main activities]. I have a background in [relevant prior experience]. I write in [languages]. I would describe my style as [adjective, adjective]. I care about [a thing or two you care about].

Part 2: How to respond

The second box is where most of the gains are. This is where you tell the model how to behave in every conversation. Some patterns that work well:

• Length and structure defaults. "Default to short responses. Use bullet points and short paragraphs. Skip preambles."
• Tone. "Be direct. No apologies, no hedging, no 'I hope this helps.' Match my conversational tone."
• Things to avoid. "Do not use the words 'leverage,' 'utilize,' 'going forward,' or 'in today's fast-paced world.' Do not begin responses with 'Great question.'"
• Calibration. "When you're not sure, say so explicitly. Tag uncertain facts with [unclear]. Don't fill in gaps with confident guesses."
• Specific habits you want. "If I ask about a tradeoff, present both sides before recommending one. If I ask about a decision, ask me clarifying questions first if anything important is missing."
• What to do with code or technical content. "When I share code, point out bugs and security issues directly. Don't add comments explaining what the code does — assume I can read it. Suggest improvements but don't rewrite unless I ask."

A reliable example template:

Respond in clear, plain English. Skip unnecessary preambles ("Great question," "I hope this helps"). Default to short replies unless I explicitly ask for depth.

>

Tone: direct, slightly informal, never apologetic. Use short paragraphs and bullet lists. Match my register if I write casually.

>

When you are uncertain about a fact, say so explicitly. Mark uncertain claims with [unclear]. Do not fabricate sources.

>

When I ask about a decision or tradeoff, ask one or two clarifying questions if a key detail is missing; otherwise present both sides briefly and then give a recommendation with a confidence level.

>

When you draft writing for me, produce 2-3 variants by default. End every email draft with a clear next step. Do not start emails with "I hope this finds you well."

Save it. Try it on your next ten conversations. Edit it the next time something specific irritates you.

ChatGPT Memory

Memory is a separate, automatic system. It picks up facts across conversations ("Andres works at AI Expert," "He is vegetarian," "He prefers metric units") and applies them to future conversations.

The pros: less re-explaining, more personalised responses.

The cons: it remembers things you might not want it to, it does not transfer if you switch accounts, and reviewing what is in there takes some attention.

A useful approach:

• Turn memory on if you want a more personalised assistant.
• Periodically review (Settings → Memory) and delete anything sensitive, outdated, or accidentally captured.
• Tell ChatGPT explicitly when you want it to remember something important: "Remember that I lead the AI team at [company]." That gets stored as a fact.
• When you do not want something remembered, say "Don't remember this for later" or use a Temporary Chat (the icon next to the model selector) — temporary chats are not saved and do not feed memory.

If you are highly privacy-conscious, leave memory off and use custom instructions only. They give you 80% of the benefit with full control.

What belongs where

Use this split:

| Context type | Where it belongs | Example | | --- | --- | --- | | Stable response preferences | Custom instructions | "Use short paragraphs and flag uncertainty." | | Stable non-sensitive personal context | Optional memory | "I prefer metric units." | | Project-specific context | Project/workspace instructions | "This project is for our public AI training site." | | Sensitive work details | Approved work conversation only | Customer issue, contract clause, internal draft | | One-off private details | Temporary chat or do not use AI | Personal medical, legal, family, or financial detail |

That split keeps the convenience without turning your global settings into a messy database of things you forgot you disclosed.

Claude Projects: a different model

Claude's approach is slightly different. Instead of a single set of instructions that apply globally, Claude has Projects — folders that contain a custom instruction set, knowledge files, and the conversations inside that project.

The pattern: create a project per major area of your work (e.g., "Personal," "Client X," "Internal Strategy"), give each one its own context, and start conversations inside the right project. The model carries that project's context into every conversation in it.

This is genuinely useful for anyone who switches between unrelated domains. Your "personal" project does not need to know about your client work and vice versa, but each context is rich within itself. The downside is the overhead of organising yourself into projects.

Common mistakes

A few patterns to avoid.

Instructions that are too long. Three pages of instructions are worse than three paragraphs. The model handles 200–500 words of instructions reliably; beyond that, it starts ignoring some.

Stacking contradictory instructions. "Be friendly and direct. Avoid filler but be warm. Keep it short but cover everything." Pick a side. Contradictions confuse the model and produce inconsistent outputs.

Instructions that lock you in. "Always respond in Estonian" or "Always use bullet points" can be annoying when you actually want English prose. Make defaults rather than absolutes — "default to bullet points unless the task is conversational."

Forgetting that they apply everywhere. If your custom instruction says "be terse, no warm preamble," every conversation will be terse — including the one where you wanted a friendly birthday card draft. The fix is to override per-conversation when needed: "Ignore my default style for this — I want something warm and chatty."

A useful experiment

Set up your custom instructions today. Then over the next week, every time you find yourself typing the same context into a new conversation — "I work in product," "I prefer short replies," "Don't apologize" — pause. That repeated context belongs in your instructions, not in every prompt.

By the end of the week, the instructions will have evolved into something that fits your real use. Re-read them once and edit. After that, you will probably touch them every couple of months as your work shifts. The total time investment is small. The output quality lift is permanent.

The takeaway

Custom instructions are one of the highest-ROI defaults in AI. They are free, they take ten minutes, and they materially improve every conversation you have. Almost no beginners use them. The ones who do operate at a different baseline — every conversation starts further down the runway, and the cumulative time saved across a year is large.

Go set yours up. The next prompt you send will start producing better answers, and you will not have to do anything cleverer to make it happen.