Providing Context: From Vague to Precise
You learn how context drastically improves AI output quality.
Why this lesson matters
You've learned the framework. You know you need to give a persona, a task, a format and constraints. But there's one building block that makes the difference between a "fine" answer and an answer you can use right away: context.
Without context, AI does what it always does: give the most average answer. With the right context you get answers that feel as if someone knows your business, your customer and your situation.
Why AI answers generically without context
AI models don't know who you are, where you work, or what you've done before. Every conversation starts from zero. Without context, the model picks the answer that's statistically most likely, and that is by definition generic.
Compare it to a consultant who doesn't know your business. If you say "give advice on customer retention", you get the standard advice found in every marketing book. But if you first explain who your customers are, what your current retention rate is, and which actions you've already tried, you get advice that matters.
AI has no memory between conversations (unless you set that up explicitly). Every time you start a new conversation, the AI starts with a blank slate. Context is how you fill in that slate.
Types of context you can give
| Type of context | What it does | Example |
|---|---|---|
| Background | Places the problem | "Our company supplies software to hospitals" |
| Audience | Sets tone and level | "The reader is a CEO without a technical background" |
| Tone | Steers the writing style | "Write formally but accessibly" |
| Examples | Shows what you expect | "Here's an example of how I want it" |
| Limits | Rules out irrelevant information | "Focus only on the Dutch market" |
| Data | Gives facts to work with | "Our revenue in Q3 was €4.2M" |
Example: without vs. with context
Without context:
Write a LinkedIn post about AI.Keep reading
Leave your name and email address and you can read the rest
You get the whole lesson right away, and every other lesson stays open after that. No password, no confirmation email.
Result: A generic story about "AI is changing the world" that you've seen a thousand times.
With context:
Write a LinkedIn post (max 200 words) for me as HR director of an accountancy firm with 200 employees. Topic: how we use AI to speed up our recruitment process. Tone: enthusiastic but realistic. Give a concrete example: we've cut time-to-hire from 45 to 28 days with AI screening of CVs. End with a question to the network.Result: A personal, credible post that fits your situation and is ready to publish.
Few-shot prompting: learning from examples
One of the most powerful ways to give context is few-shot prompting: you give the AI one or more examples of the result you want.
Why does this work so well? Because an example communicates more than a hundred words of instruction. From your example the AI learns the style, structure, length and tone, all at once.
Example: product descriptions
Write product descriptions in the same style as this example:
Example:
Product: CloudSync Pro
Description: Sync your files between all your devices in seconds. No fuss with cables or USB sticks: everything is automatically ready where you need it. Ideal for teams working across multiple locations.
Now write a description for:
Product: TaskFlow AI
Category: project management tool with AI planningFew-shot works best when you give 2-3 examples. One example could be a coincidence, but with several examples the AI picks up the pattern reliably.
Zero-shot vs. few-shot vs. many-shot
| Method | What it is | When to use |
|---|---|---|
| Zero-shot | No examples, only an instruction | Simple tasks, standard formats |
| One-shot | Give one example | When you want to steer the style |
| Few-shot | Give 2-3 examples | When consistency matters |
| Many-shot | Give 5+ examples | For very specific patterns |
More examples isn't always better. With more than 5 examples the AI can become too rigid and stop varying creatively. Start with 2-3 and only add more if the output isn't consistent enough.
In practice: building context in layers
You don't have to cram all the context into one prompt. You can build it up:
Step 1: Give the background
I work as a controller at a logistics company with 500 employees and €80M revenue. We report monthly to the board.Step 2: Give the specific assignment
Write the management summary for the November monthly report. Revenue: €6.8M (budget: €7.2M). Margin: 18% (last year: 20%). Main cause: rising fuel costs.By building context in steps, you can reuse the first part for several assignments.
Quiz
What is few-shot prompting?
Which context is MOST valuable to give an AI when you want it to write an internal memo?
Want to keep your progress, get the practice files and sign up for the free evening? Create a free account. All you do is click a link in your email.
Create a free account