The Blank Slate Problem
Claude starts every single conversation knowing absolutely nothing about you. No memory of last time. No assumptions about your situation. No idea what a good answer looks like for your life. Understanding this one fact — and acting on it — is the single biggest unlock in using AI well.
Why most people get mediocre answers
Think about how you'd ask a question if you'd just met a stranger at a braai. You'd give them some background. You'd explain your situation. You'd tell them what you've already tried. You'd be specific about what you actually need.
Now think about how most people use Claude. They type one sentence. They get a generic answer. They think AI isn't that useful.
The stranger at the braai would also give you a generic answer if all you said was "how do I save money?" They'd give you the same advice they'd give anyone. But if you said "I earn R18,000 a month, I have R8,000 in clothing account debt at 22% interest, no emergency fund, and I want to start building towards a house deposit in the next 3 years" — now they can actually help you.
Claude is the same. The context is the instruction.
"Most people give Claude the task. The people who get great results give Claude the situation first."
What Claude doesn't know about you
Unless you tell it, Claude has no idea:
- — Who you are or what you do
- — Your income, debts, assets, or financial situation
- — Your constraints — time, money, skills, location
- — What you've already tried and why it didn't work
- — What a "good answer" looks like for your specific life
- — Your goals, values, or what you're trying to build
- — That you're in South Africa, dealing with SARS, or earning in Rands
When you don't supply this, Claude fills in the gaps with assumptions about the most likely person asking that question. That person is usually some average, generic human being — not you.
What changes when you give context
Here's the same question asked two ways. The difference in output quality is not small — it's the difference between advice you could have found on Google and advice that could actually change what you do:
Claude can now apply the avalanche vs snowball method to your specific debts. It can calculate exact payoff timelines. It can tell you whether your car payment is the actual problem. It can recommend whether to put the R5,500 surplus toward debt or emergency fund first — based on your interest rates and real risk profile. It can give you a month-by-month plan.
The generic version can't do any of that. It can only tell you that debt is bad and you should pay it off. You already knew that.
The five things to include in any serious prompt
You don't need all five every time. But whenever you're asking Claude for something genuinely important, check this list:
The shortcut: ask Claude what it needs
If you're not sure what context to give, just ask. This is one of the most underused techniques in AI:
Claude will ask you exactly what it needs. You answer those questions. Then you have a tailored brief and Claude has the context to actually help. This takes 2 minutes and the difference in output quality is enormous.
The permanent fix: Claude Projects
If you use Claude regularly, you'll quickly get tired of re-entering your context every single conversation. The fix is Claude Projects — a feature in Claude Pro that lets you set a persistent context that applies to every conversation in that project.
A good project context looks something like this:
Always use South African Rands. Always reference current SARS rules when tax is relevant. Always be direct — I don't need things softened. I prefer tables and structured output over paragraphs. Flag any assumptions you make about my situation.
With that set, every conversation starts already knowing who you are. You never have to re-explain your situation. Claude can give you specific, calibrated advice from the very first message.
This is the jump from "using AI" to "AI that works for you specifically."
Open Claude and try it →