Every time you open a new chat with an AI, you are a stranger to it.
It does not know your name, your job, your communication style, or what you were working on yesterday. It does not know your industry, your audience, your constraints, or your preferences. Every session, it starts with nothing and makes assumptions about all of it.
Most people do not notice this because they are asking one-off questions. But if you use AI for actual work — writing, analysis, strategy, research — you are quietly paying a tax on every session. You re-explain. You correct. You nudge the tone back to where it needs to be. You repeat yourself constantly without realising it.
The professionals who have solved this do not fight the limitation. They route around it.
The Fix: A Master Context Document
The idea is simple. Write a single document — a few paragraphs — that contains everything an AI needs to know about you to produce useful work immediately. Then load it at the start of every session, or pin it into a Project so it is always there.
Here is what goes in it:
Your role and context. Who you are, what you do, who you do it for. One sentence on the domain you operate in.
Your communication preferences. How you like outputs structured. Length. Tone. What you hate — bullet points, vague hedging, unnecessary caveats, corporate language.
Your working constraints. Things the AI would otherwise get wrong by default. Any recurring terminology, house rules, or non-negotiables.
What good looks like. A sentence or two on what a strong output actually looks like for your use cases. This is the single most effective thing you can add.
That document does not need to be long. A well-written 200-word context brief will outperform a chaotic 1,000-word one. The goal is clarity, not comprehensiveness.
Once you have it, use Claude or ChatGPT Projects to pin it permanently. From that point, every conversation starts at the right level — no re-explaining, no correcting, no lost context.
Why your AI keeps forgetting — and the permanent fix that isn't a tool
A clear breakdown of why LLMs are stateless by design, what that actually means for your daily work, and why the solution is not a plugin — it is a system. One of the most-shared AI workflow pieces this month.
The context document: what AI power users are quietly using to get better outputs
Interviews with a dozen professionals who noticed their AI outputs improve significantly after building a master context brief. What they include, how long it takes to write, and why most people skip this step entirely.
MIT Technology Review: the productivity gap between AI power users and casual users is widening
The gap is not about which model people use. It is about whether they have built systems around it. Context management is consistently cited as the highest-leverage habit separating the two groups.
Claude Projects / ChatGPT Projects
Both Claude and ChatGPT now have Projects — a way to create a persistent workspace where your context document, files, and preferences are automatically loaded into every conversation. You set it up once. It works forever.
Claude Projects let you pin a system prompt and upload files that stay in context across all sessions in that project. ChatGPT Projects work similarly, with the addition of memory that automatically captures things across conversations.
If you write a master context document and load it into a Project, you have effectively created a version of the AI that already knows you. That is the entire game.
Both are available on paid plans. Claude Pro is $20/month. ChatGPT Plus is $20/month.
Use this as your master context brief template. Fill it in once, paste it at the top of any new chat, or pin it to a Project:
ABOUT ME
I am [role] working in [industry/domain]. My primary focus is [main work area]. I work with [audience/team/clients].
HOW I WORK
I prefer outputs that are [concise / detailed / structured / direct]. Avoid [bullet spam / hedging / corporate tone / unnecessary caveats]. When in doubt, [be direct / ask before assuming / give me options].
RECURRING CONTEXT
[Any standing information: company name, key projects, recurring terminology, things the AI keeps getting wrong.]
WHAT GOOD LOOKS LIKE
A strong output for me is one that [specific standard — e.g. "could be sent immediately without editing" / "surfaces assumptions clearly" / "gives me a decision, not a list of options"].
When I start a session, confirm you have read this and are ready to work.
The last line — "confirm you have read this" — forces the AI to acknowledge the context before proceeding. It catches cases where the document was too long and got partially ignored.
Does your AI know your name?
Not because you told it once — because you have a system that ensures it always knows, every session.
If the answer is no, that is the single most valuable 20 minutes you could spend this week.
Reply and tell me what you put in your context document. I read every response.
If you want to build this properly — not just a rough paragraph but a real system with Projects, pinned context, and a weekly workflow that compounds — I put together two resources at tminusai.com.
The free Prompting Hub covers the fundamentals: 19 laws of prompting and 32 cheat codes that apply to any model. A good starting point if you are setting up your context document for the first time.
The Power Guide Pro (€59.99) covers the full system: personalisation setup, Projects configuration, workflow playbooks, and the exact structure I use to make any AI session productive from the first message. It is the permanent fix, documented.
More next week.
Kapish