Your First AI Conversation

35 min beginner Lesson 3

Learning Outcomes

  • Structure any request using a simple, repeatable four-part framework
  • Give the AI enough context to understand your situation without over-explaining
  • Set constraints — length, format, tone, and audience — so the first draft is usable
  • Assign a role and tone to shape how the AI responds
  • Iterate on a weak answer instead of starting over from scratch

Lesson Plan

Segment Duration Topic
Intro 3 min Why "just asking" rarely works first time
Framework 6 min The Task–Context–Constraints–Format model
Context 5 min How much background to give
Constraints 5 min Length, format, tone, audience
Roles & tone 5 min Asking the AI to "be" someone
Iterating 6 min Fixing a bad answer in a follow-up
Wrap-up 5 min Putting it together, next lesson

Before You Begin

Pre-work:

Shopping List:

  • A web browser with one AI chat tool open (the free tier is fine).
  • A short piece of your own text to practise on — a paragraph of notes or some bullet points.
  • Ten quiet minutes. This lesson is hands-on; you'll get more from it by typing along.
1 Why Vague Requests Fail — and the Four-Part Fix

Most people type something short, get a bland answer, and conclude the AI "isn't very good." Almost always, the problem is the request. A vague ask like "Write me an email about the project" forces the AI to invent the reader, purpose, and length. Compare a prompt that does the guessing for it:

> Write a short email to my manager: the website redesign is two
> weeks behind because the design sign-off was late, but reassure
> her we'll still hit the launch date. Under 120 words, warm but
> professional.

Same tool, wildly better result — because this prompt tells the AI what you want, why, and in what shape. Every strong prompt has the same four ingredients — Task, Context, Constraints, Format.

Part What it answers Example
Task What do you want done? "Summarise this report."
Context What background does the AI need? "It's a quarterly budget review for leadership."
Constraints What limits or rules apply? "Five bullets, plain English, no jargon."
Format What should the output look like? "A list I can paste into an email."

You won't always need all four — a quick question rarely needs a Format — but when an answer is weak, ask: which did I leave out?

NOTE
Key Insight
A conversation with AI is not a search query. With search you use the fewest words; with AI, the more relevant detail you give, the better the answer. Brief it like a capable new colleague who knows nothing about your situation — but don't overload it; it doesn't need your company history to fix a typo.
2 Giving Context — The Part Most People Skip

Context is the background that helps the AI tailor its answer to your situation — the most commonly skipped ingredient and the biggest difference-maker. Good context answers: who is this for? ("my CFO"), what's already true? ("the deadline is fixed"), and what's the goal? ("approve the spend"):

> Draft a reply to this customer complaint. The customer is a
> long-standing, usually-happy client, this is their first issue,
> and the delay was our fault. Apologise sincerely, explain
> briefly, and offer a goodwill gesture — without admitting liability.
>
> [paste the complaint here]

Without that paragraph the AI writes a generic apology. With it, it knows the relationship, history, and goal — and produces a reply you'd actually send.

NOTE
How It Works
The AI cannot see your inbox, files, or past conversations unless you paste them in. Every chat starts blank, so anything it doesn't know it assumes — generically. Context replaces guesswork with your reality.
3 Setting Constraints — Length, Format, Tone, Audience

Constraints are the limits that keep output usable. Without them, AI tends to be too long, too formal, or aimed at the wrong reader. Four cover most:

Constraint Ask for it like this
Length "Under 150 words," "three sentences," "one paragraph."
Format "As a bulleted list," "as a table," "as a checklist."
Tone "Warm and reassuring," "direct," "formal and neutral."
Audience "For a non-technical reader," "for executives," "for a frustrated customer."

A request with all four leaves little to chance: "Turn these notes into clear minutes — a numbered list of decisions, then action items with owners. Under 200 words, neutral, for people who missed the meeting." You're not telling the AI how to write; you're giving it the shape of what you need back. That's a constraint, not a script.

TIP
Tip
Length is the single most valuable constraint. AI defaults to long, so keep it under 100 words turns a rambling reply into something you can paste and send. Word counts are approximate — trim the result yourself if you need an exact figure.
4 Roles and Tone — Asking the AI to "Be" Someone

You can change an answer dramatically by telling the AI what perspective to write from — a role, set in one sentence at the start of your prompt:

> Act as an experienced operations manager reviewing this proposal.
> Point out the three biggest practical risks a leadership team
> would want addressed before approving it.
>
> [paste the proposal here]

The role focuses the AI's attention — a more grounded critique than no role at all, because you've told it whose eyes to use. A few worth keeping:

Role Good for
"Act as a sceptical executive reviewer" Stress-testing a proposal before you send it
"Act as a friendly editor" Tightening writing without changing your voice
"Act as a financial analyst" Explaining numbers in plain English

Pair the role with a tone instruction. Tone is how it sounds; role is who it sounds like.

NOTE
Key Insight
A role is a shortcut, not magic. It nudges the AI toward a useful style, but it is not actually an operations manager and lacks real experience — treat its output as a first draft to judge, never as expert sign-off.
5 Iterating — Fixing a Weak Answer Instead of Starting Over

The habit that separates confident AI users from frustrated ones: when the first answer isn't right, they don't retype everything in a new chat. They reply with a small correction and the AI adjusts, because it remembers the conversation. If a draft is too long and formal, say:

> Good start. Now cut it to half the length and make it sound
> warmer — less corporate, more like a real person wrote it.

You're sculpting, not rebuilding. Common fixes:

Problem Follow-up to fix it
Too long "Cut this to under 100 words."
Too generic "Make it specific to a software company."
Wrong tone "More formal — this is going to a client."
Missed the point "Focus only on the budget impact."

Only start over when your original request was fundamentally wrong — you asked for an email but actually need a slide outline. Otherwise, iterate.

TIP
Tip
If an answer is 80% right, never throw it away — tell the AI exactly which 20% to change. If after five or six rounds it's still drifting, start a fresh chat with one clean prompt.
6 Putting It All Together — One Real Task

Let's combine everything into one realistic prompt. You have rough notes from a meeting and need a clean status update for stakeholders. Instead of a bare "Write a status update from these notes," use the full-framework version — all four parts plus a role:

> Act as a project manager writing for senior stakeholders. Turn
> the rough notes below into a status update. Context: a monthly
> software-rollout update; slightly behind but the launch date is
> still safe. Under 200 words, confident, no jargon. Format: a
> one-line headline, a "Progress" paragraph, a bulleted "Risks"
> list, and a one-line "What's next."
>
> [paste your rough notes here]

Run that, read the result, and apply Step 5 to perfect it. That's the whole workflow — it works for emails, summaries, plans, and any writing task ahead. In short: Role → Task → Context → Constraints → Format → iterate.

NOTE
Build a Library
Save your best prompts in a notes file. You'll reuse the same shapes constantly, and a small library saves real time — we build on this in Lesson 9.

Questions & Answers

Q: Isn't writing a detailed prompt slower than doing the task myself?
For a one-line reply, sometimes. But for anything you'd spend more than a couple of minutes on, a 30-second prompt that returns a usable draft is a clear win — and the framework becomes second nature quickly.
Q: What if I don't know what I want until I see something?
That's common. Ask for a quick first attempt with minimal instructions, then react: "more formal," "shorter," "focus on cost." Iterating is a valid way to discover what you want.
Q: Is it safe to paste my work documents into the chat for context?
For practice, use low-sensitivity material. Whether you can paste real work content depends on your tool's data policy and your employer's rules — covered in Lesson 2 and Lesson 10. When in doubt, strip out names and numbers first, or check your AI policy.
Q: The AI confidently told me something that turned out wrong. How do I trust it?
AI can state incorrect things with total confidence — a known limitation, not a bug you caused. For anything factual or consequential, verify it independently. Good prompting improves writing and structure; it does not guarantee accuracy. Lesson 8 covers verification.

Key Takeaways

  1. Vague in, vague out — A bland answer usually means a bland request. Fix the prompt, not the tool.
  2. Use the four-part framework — Task, Context, Constraints, Format. When an answer disappoints, find the part you skipped.
  3. Context is the multiplier — The AI starts each chat knowing nothing about your situation. Tell it who, why, and what's already true.
  4. Constrain the shape — Length, format, tone, and audience turn a rambling reply into something you can send.
  5. Roles and tone steer the voice — One opening sentence ("Act as a friendly editor") reshapes the whole answer.
  6. Iterate, don't restart — When a draft is 80% right, fix the 20% in a follow-up. The AI remembers.

Next Steps: Lesson 4: AI for Document Work