If you've ever typed something into ChatGPT, Claude, or Microsoft Copilot and thought "that answer was useless", this article is for you.
Here's the uncomfortable truth: most of the time, the AI wasn't the problem. The prompt was.
The good news? Prompting isn't a technical skill. There's no code, no jargon, no course you need to sit through. A good prompt is just a clear brief, the same brief you'd give a capable new employee on their first day. And every good prompt, whether you're writing an email, analysing a spreadsheet or drafting a proposal, is built from the same four components.
The four components of every good prompt
This framework comes from Microsoft's own guidance for Copilot, which matters, because it means the terminology you learn here is the exact language you'll see inside the tools you already use.
| Component | Microsoft/Copilot Term | What It Does |
|---|---|---|
| Instruction | Goal | What you want the AI to do |
| Context | Context | Who, why, and the situation — narrows the answer |
| Input Data | Source | The material the AI works from |
| Output Indicator | Expectations | Format, length, tone, audience |
Four things. That's the whole framework. Let's see it in action with a scenario every business owner recognises.
A real example: replying to a customer complaint
Imagine you run a restaurant. A customer named Sarah has emailed to say her main course arrived cold on Saturday night and she's especially disappointed because it was her anniversary dinner. You want AI to help you reply.
What most people type
"Write a reply to a customer complaint."
The AI has no idea who complained, what happened or how you want to handle it. So, it guesses. You get a stiff, generic apology that sounds like it came from a call centre — "Dear valued customer, we take your feedback seriously..."
Most people stop here and conclude AI isn't very good. They're wrong. They just haven't briefed it yet.
The same request, done properly
"Reply to the customer email below." — Goal (what you want done)
"The customer, Sarah, dined with us on Saturday for her anniversary and her main course arrived cold. She's a regular who visits monthly. I want to apologise sincerely, not make excuses, and invite her back with a complimentary dinner for two." — Context (the situation, who's involved, and the outcome you want)
"Customer email: [paste Sarah's email here]" — Source (the material the AI works from)
"Keep it under 100 words, warm and personal, sign off as 'Marco, Owner'. No corporate phrases like 'we value your feedback'." — Expectations (format, length, tone)
Same AI. Same request. Completely different result — a reply that mentions the anniversary, offers the free dinner and actually sounds like Marco wrote it.
The AI didn't get smarter between the two prompts. The human got clearer.
Why this framework is a diagnostic tool, not just a recipe
Here's what makes the four components genuinely useful: when an AI answer goes wrong, each missing component fails in a predictable way. That means you can diagnose a bad output instead of retrying at random.
- Remove the Instruction → nothing useful happens. The AI doesn't know what job to do.
- Remove the Context → you get a generic apology that could be from any business to any customer. No anniversary, no free dinner because the AI didn't know.
- Remove the Input Data → the AI invents the details of a complaint Sarah never made. This is where "hallucinations" come from.
- Remove the Expectations → you get 400 words of corporate waffle ending in "we value your feedback".
So, the next time an answer disappoints you, don't ask "why is the AI bad?" Ask "which of the four components did I leave out?" Nine times out of ten, one of them is missing.
The Copilot shortcut: sometimes you can skip the Source
If you use Microsoft Copilot inside Outlook, Word or Teams, there's a useful simplification: Copilot can often see your source material already. It can read the email thread you're replying to, the document you have open or your calendar.
That means the Input Data component can frequently be dropped as Copilot is "grounded" in your files. You still need the other three. In fact, they become more important, because Copilot with access to your entire inbox, but no Context about what you actually want is a very fast way to generate a very wrong email.
Three habits that make every prompt better
Once the four components are second nature, layer in these:
- Name the audience. "Explain this for a non-finance audience" shapes the vocabulary more than any adjective ever will.
- Quantify the output. "Under 100 words." "Exactly five bullet points." Vague length requests get vague results.
- Say what you want, not what you don't. "Warm and conversational" works better than "don't be too formal". AI models follow positive instructions far more reliably than negative ones.
What prompting really is
Here's the point we make in every training session and every client project at Gennovate AI:
The goal isn't to make people better at talking to AI. It's to make them better at thinking about what they actually want, the AI just happens to be the thing that punishes them for not knowing.
A vague brief gets a vague result, whether you give it to a machine or a person. The four components — Goal, Context, Source, Expectations are simply a checklist for clarity. Learn them once and you'll use them in every AI tool you ever touch, because they're not about the technology. They're about you.
Frequently asked questions
What is a prompt in AI? A prompt is the instruction you give an AI tool like ChatGPT, Claude or Microsoft Copilot. The quality of the answer depends directly on the quality of the prompt, a clear, complete brief produces dramatically better results than a one-line request.
What are the four components of a good prompt? Instruction (what you want done), Context (the situation and desired outcome), Input Data (the material to work from) and Output Indicator (format, length, and tone). Microsoft Copilot calls these Goal, Context, Source, and Expectations.
Do I need technical skills to write good prompts? No. Prompting is a communication skill, not a technical one. If you can brief a new employee clearly, you can write an excellent prompt.
Why does AI sometimes make things up? Usually because the prompt is missing Input Data. When the AI isn't given the actual material to work from the email, the report, the numbers, it fills the gap by inventing plausible-sounding details. Providing the source material is the single most effective way to prevent this.
Is prompting different in ChatGPT vs Microsoft Copilot? The framework is identical. The main difference is that Copilot is grounded in your Microsoft 365 data, so it can often see your source material (the email thread, the open document) without you pasting it in.
---
Gennovate AI builds custom AI applications for SMEs across the UK, US, EU and worldwide. We train teams to get real results from the AI tools they already pay for. If your team is typing one-line prompts and getting one-line value, book a 30-minute call.