Guide · Getting set up

Getting AI to know you

Most people learn to prompt and stop there — so the tool meets them as a stranger every time. The real unlock is the settings, the project spaces and the assistants you've never opened. Set them up once, and ChatGPT or Microsoft Copilot knows your role, your standards and your context, and acts on them without being told twice.

Where most people are

You can prompt. But the tool still doesn't know you.

Most people who use ChatGPT or Microsoft Copilot have learned to prompt — and stopped there. Every conversation starts from nothing, so they re-explain who they are, what they do and how they want things written, over and over. They treat a tool that could know them like a stranger they brief from scratch each morning.

The interesting part of these tools isn't the chat box. It's everything most people have never opened: a few settings, a project space, and the assistant you can build for a job you do all the time. None of it needs code.

The unlock

Three things you've never set up.

Each one takes the tool from generic to genuinely yours — and each exists in both ChatGPT and Microsoft Copilot, under slightly different names.

Settings

Make it know you

Tell it once who you are, your role, your standards and how you want it to respond — so every chat starts in context instead of from scratch.

ChatGPT — personalisation, custom instructions and memory
Copilot — Copilot Memory and personalisation, held under your organisation's controls

Projects

Give it a project space

A workspace that holds the instructions and documents for one strand of work, so the tool reasons over the right material — and keeps one client or matter separate from another.

ChatGPT — Projects — their own instructions, files and scoped memory
Copilot — Notebooks — grounded on the exact set of files you choose

Agents

Build a reusable assistant

Turn a job you repeat — onboarding, a particular kind of draft, a review against a checklist — into an assistant you configure once and reuse, or share with your team.

ChatGPT — a Custom GPT, built with no code
Copilot — a no-code agent, built in Agent Builder

A trick for the blank box

Don't know what to write? Let it interview you.

The hardest part of personalisation is the empty box — most people freeze in front of "tell me about yourself." So turn it around: ask the tool to interview you. One question at a time, each one shaped by your last answer — not a list of ten fired at once — until it has enough to write your profile for you.

Try this prompt

"Interview me so you can get to know me and work the way I need. Ask up to ten questions, one at a time — wait for my answer before the next, and let my answers shape what you ask. When you're done, write a short profile of me and a set of custom instructions I can paste into your settings."

Because each question builds on the last, it draws out the context you'd never think to volunteer — and it ends with something ready to paste. Then you do the part that matters: read it, tune it and own it. It works the same way in ChatGPT or Copilot. In a recent one-to-one with a finance director, this was the move that landed: ten minutes of being asked the questions, and the tool finally sounded like it knew him.

Stop briefing a stranger every morning. Set the tool up once, and it acts as you would — without being told twice.

Why it works

The real skill is direction, not prompting.

In Kramer Consulting's AI-fluency framework, the competency at work here is Direction — communicating a task clearly enough that the tool can act on it. Personalisation, projects and a custom assistant are how you give that direction once instead of every time: you encode your role, your context and your standards into the tool itself. It's the move from micro-managing each prompt to macro-managing the work — you set the goal, the rules and what good looks like, and let the tool carry them out.

Two cautions come with it. First, the tool only acts well if you've set it up well — vague in, vague out. Second, the instructions you write (often by asking the AI to help draft them) almost never behave perfectly first time; you review, tune and own the result. Setting the tool up doesn't take the human out of the loop — it makes your judgement count for more. And it's the groundwork for what comes next: once a tool knows you, the leap from answering questions to doing tasks is a short one.

Whichever your firm runs

ChatGPT or Copilot — the same three moves.

The moves are the same in both tools; the buttons differ — and so does one thing that matters more than any feature: where your data goes. A personal AI account and a business or enterprise tier are not the same animal. On the business tiers — ChatGPT Team and Enterprise, Microsoft 365 Copilot — your content is not used to train the models by default, and with Copilot it stays inside your organisation's tenant. On a personal account, the defaults are weaker.

For anyone handling confidential or client information — finance, legal and professional services in Luxembourg especially — that distinction is the whole point, which is why the setup sessions treat it as a first-class topic. The firm-wide version of the same question — which tools, what data, whose responsibility — is AI governance.

Where to start

Set it up properly — in a couple of hours.

Kramer Consulting runs short, hands-on intensives that do exactly this — in your own account, on your real work, one move at a time. Ordered directly and tailored to your firm; one-to-one or a small team.

ChatGPT