AI Assistants at Work: Better Prompts, Checked Answers
Learn to write clear prompts for everyday work tasks, use examples to keep formats consistent, refine drafts with follow-ups, check facts, numbers and quotes before you send, and decide what not to share with an AI assistant. Six short lessons with practice on your own tasks.
- Language
- English
- Lessons
- 6
- Minutes
- 150
Price: · The subscription includes the whole catalog at no extra cost.
Course outline
Prompts That Get Useful Answers
- Anatomy of a Prompt: Context, Task, Format (25′)
- Show, Don't Describe: Using Examples in Prompts (25′)
- Iterate: Fixing an Answer Instead of Starting Over (25′)
Answers You Can Trust
- Fluent but Wrong: Where Assistants Make Things Up (25′)
- Check Before You Send: Verifying AI Answers (25′)
- Knowing the Limits: What Not to Ask, What Not to Share (25′)
Free first lesson
Anatomy of a Prompt: Context, Task, Format
Three parts, every time
Think of the last thing you typed into your assistant. You gave it a topic — "summarize this", "write an email about the delay" — and the rest was implied. Sometimes the answer came back fine, sometimes not, and you could not tell why.
A prompt that works carries three parts:
- Context: who is involved, what the situation is, who will read the result. The assistant knows nothing about your job, your boss or your customer unless you write it in or paste it.
- Task: one job, with a verb — summarize, draft, compare, list.
- Format: the shape of the answer: length, sections, list or table, tone.
Together they answer the assistant's three silent questions: what is this about, what exactly do I do, and what should the result look like? Leave one out and it guesses.
Not everything belongs in the context. The test: would the answer change if the assistant knew it? Your boss's deadline changes it; the full org chart does not. Two lines that do a lot of work: naming a role ("act as a customer-service lead") and naming the reader ("for my boss, who is not on this thread") — both set tone, depth and what counts as relevant.
Why does naming the reader change the answer, when the task stays the same?
From one-liner to full prompt
A task like one from your week: your boss must decide today, from a 40-message thread about a delayed shipment, whether to offer the customer a discount.
The one-liner:
> Summarize this thread. [paste]
What typically comes back: a wall of paragraphs, everything weighted equally, no bottom line — nothing told it there was a decision to make.
The full prompt:
> My boss was not on this thread and must decide today whether we offer the customer a discount. Summarize the thread for her: a two-sentence bottom line first, then bullets for what happened, what the customer claims and what we promised. Under 150 words, neutral tone. Thread: [paste]
What each part added. Context — the reader, her decision, the deadline — sets what matters: the messages about scheduling a call drop out on their own. Task keeps every bullet aimed at that decision. Format — bottom line first, bullets, word cap — makes the summary usable without editing.
And the question from the last section: naming the reader changes the answer because the assistant writes to someone. Depth, tone and relevance follow from who is reading, the same way they do in your own emails.
Your turn: the format line
A customer wrote asking why their refund has not arrived; you need a reply you can send after a small edit. Context and task are drafted:
> A customer emailed about a refund that has not arrived after 12 days. Our policy is refunds take up to 15 business days. Draft a reply that reassures the customer and explains the timeline.
Your step: add the format line — one sentence telling the assistant what the reply should look like: length, tone, structure, anything that must appear. Write yours, compare it with the model answer, then run it in your assistant.
Where prompts go wrong
- Assuming the assistant knows your situation. It has never seen your boss, your customer or your policy; every fact it fills in is a guess unless it is in the prompt.
- Several jobs in one prompt. "Summarize this, draft a reply and make a table" dilutes all three. Run jobs one at a time in the same chat.
Keep these three
- Every prompt states three things: context, task, format.
- The assistant knows nothing about your job unless you write it in.
- Name the reader: audience sets tone, depth and relevance.
Rebuild one of yours
Take a request you actually have this week. Write it as you would have before — the one-liner. Then rewrite it with context, task and format, run both in your assistant, and compare.
- prompt
- The full text you send the assistant: your instructions plus any pasted material.
- context
- The part of a prompt that says who is involved, what the situation is, and who will read the result.
- task
- The one job you want done, stated with a verb: summarize, draft, compare, list.
- format
- The shape of the answer you want: length, structure, sections and tone.