The Prompt Course

Think back to the last time AI gave you an answer that was... okay. Generic. Useful in the same way a Wikipedia entry is useful — correct, but not helpful for your specific situation.

Why did that happen? Most likely because you phrased the question as if you were doing a Google search.

The problem: AI answers like someone you talk to, and it needs context to give you a good answer.

Think of it like this: you're calling a consultant who is an expert on everything. If you just say "CEO letter quarterly report template" — what do they answer? "Which industry? How did the quarter go? Who is reading it? What is the tone?" They are asking for context. Not because they are difficult — but because they want to give you a good answer. AI does the same thing, just silently. If you don't provide the context — it assumes. And those assumptions are always generic.

The five elements of a prompt that always work

You don't need to remember all five every time. But the more you include, the better the answer.

1. CONTEXT — Who are you? Tell them who you are and what you do. "I am the CEO of a consulting firm with 12 employees in Gothenburg." That is all that's needed.
2. TASK — What needs to be done? Be specific about the task. Not "help me with a letter" but "write a draft of a quarterly letter to our board."
3. BACKGROUND — Why and what? Give it the relevant background. "We had a tough Q1 (-15% compared to Q4 last year) but have a concrete plan for Q2."
4. FORMAT — What should it look like? Specify length, tone, structure. "Max 400 words. Tone: direct but not panicked. No bullet points — running text."
5. EXAMPLE — How should it sound? If you have an example of the tone or style you want — paste it in.

The comparison that shows everything

Write a CEO letter about Q1

I am the CEO of a 12-person IT consulting firm in Gothenburg. We had a tough Q1 — revenue fell by 15% compared to Q4 last year, primarily because a large project was delayed until Q2. We have a strong pipeline and a realistic plan moving forward.

Write a quarterly letter to our board (5 people, experienced businesspeople). The tone should be direct and responsible — acknowledge the challenge without sounding panicked. Focus on the Q2 plan. Max 400 words, running text, no bullet points.

Run both prompts and compare the answers: the answer to the second one is specific to your company and ready to edit.

There is a common mistake that even experienced AI users make: they ask the AI to do the job without providing the raw material.

"Write a follow-up email to a potential customer."

Which email? Which customer? What was the meeting about? What do you want them to do next? What is the tone of your relationship? AI writes something. It looks like an email, but it says nothing about your customer.

The key is to understand: AI is only as good as the information it has to work with.

Three levels of context

I run an accounting firm. A client is late on payment for the third month.

I run an accounting firm. The client Lars Andersson at Andersson & Co is late on payment for the third month in a row — total debt 45,000 SEK. We want to maintain the relationship but must get paid. I have sent two previous reminders.

Level 2, plus: The email should be professional but firm. Not aggressive. Give him one last chance with a concrete deadline (10 days). Do not mention that we are considering debt collection — that is the next step if he doesn't respond. Max 150 words.

The technique: Paste the raw material

One of the most powerful things you can do: give the AI the actual raw material and let it work with it.

Instead of "Summarize the main points from the meeting" — do: "Here are my notes from the meeting: [paste notes]. Write a structured summary including: 1) Decisions, 2) Action items with responsible parties, 3) Open questions. Max half an A4 page."

It takes you 30 seconds to paste the notes. The result is far better.

The same principle applies to: email threads you need to reply to, contracts you need to understand, reports you need to summarize, articles you need to analyze.

You ask for something. You get an answer that is almost right. Then you accept it, or start from scratch with a new prompt.

There is a better way.

AI is designed for conversation. The best answer almost never comes on the first try — it emerges through dialogue.

Three ways to improve an answer

Good foundation. Make the third paragraph shorter — it's repetitive. And the tone in the introduction is too formal for our relationship with this customer.

This sounds too corporate. Here is an email I am happy with: [paste here]. Rewrite it using that tone.

You mention three options. Which one do you recommend for a company in our situation and why?

Use the conversation memory

Within one conversation, AI keeps what you have said in view, up to the limit of its context window. Some apps, like ChatGPT and Claude, can also remember things between conversations if you turn memory on. This means you can build up context iteratively without repeating yourself.

Example of an effective flow:

1. "I am the CEO of X. We have problem Y. Help me think through it."
2. AI reasons. You adjust: "What you call problem A is actually not a problem — it's solved. Focus on B and C."
3. AI adjusts. "Now give me a concrete action plan for B."
4. "Good. Now write a draft email to the board based on what we've come up with."

Four messages. The result is tailored to your situation in a way that a single large prompt can never be.

When to start over vs. continue

Continue in the same conversation when: the answer is on the right track but needs adjustment, you want to build on what was said, the tasks are related.

Start a new conversation when: the conversation has become long and scattered (50+ messages), you are switching to a completely new topic, you want a "fresh" perspective without previous assumptions.