Context is the single most underused ingredient in prompting. Most people ask a question and expect the AI to fill in the blanks. The AI can't — it only knows what you tell it.
In this lesson, you'll learn what types of context matter, how to provide them efficiently, and the difference between prompts that get generic answers and prompts that get genuinely useful ones.
Why Context Matters So Much
Imagine asking a stranger: "Should I take the job?"
They have no idea what job. No idea what you currently do. No idea what you value in your career, what you're paid now, what the new opportunity offers, or where you want to be in five years. Their answer is going to be useless or so generic it might as well be a fortune cookie.
Now imagine asking a close colleague who knows your situation in depth. Same question, completely different answer — because they have context.
AI is the stranger. You need to give it the context the colleague already has.
The Five Types of Context
1. Who You Are
The model should know your role, expertise level, and relevant background. This changes not just the content but the vocabulary, depth, and assumptions it makes.
Without context:
Explain equity compensation.
With context:
I'm a software engineer with 5 years of experience who just got a job offer
with a startup. I've never had equity compensation before. Explain how it
works in a way that's useful for evaluating whether the offer is good.
The second prompt will get you a practical, level-appropriate explanation of vesting, cliffs, strike prices, dilution — the things you actually need — instead of a textbook definition.
2. Who the Output Is For
The intended audience changes everything. Writing for your CEO is different from writing for a technical team. Writing for a skeptical customer is different from writing for an existing fan.
Write an explanation of [topic].
Audience: [specific description — not just "non-technical" but "CFOs who understand
numbers but have no software development background and are skeptical of tech hype"]
3. The Goal Behind the Task
What are you actually trying to accomplish? Not just the immediate task, but the purpose behind it.
Surface task: "Summarize this report."
Real goal: "I have a 20-minute board meeting tomorrow and need to present the key findings from this report without reading it verbatim."
When you tell the model the real goal, it can shape its response around what will actually help you — not just technically complete the task.
4. Constraints and Boundaries
What can't you do? What must you include? What format are you working within?
I'm writing a LinkedIn post about [topic].
Constraints:
- Under 200 words (LinkedIn's algorithm prefers shorter posts)
- Can't mention competitors by name (company policy)
- Must end with a question to encourage comments
Without these constraints, you might get a beautifully written 400-word piece with a competitor comparison that you can't use.
5. Relevant Background Information
Any facts or data the model needs to do the task well but wouldn't otherwise know.
Context: Our product is B2B project management software. Our users are typically
marketing managers at agencies with 10–50 employees. Our main differentiator is
client-facing approval workflows, which our competitors don't have.
Given that context, write a one-paragraph product description for our website homepage.
The "Minimum Viable Context" Rule
You don't need to write a novel. You need to provide the minimum context that meaningfully changes the answer.
Ask yourself: What would the model most likely get wrong if I didn't tell it this?
That's what to include.
If you're asking for a recipe, your dietary restrictions are essential context. Your location, probably not. If you're asking for career advice, your industry and years of experience matter. Your shoe size doesn't.
Context in Practice: A Before and After
Before (no context):
Write me a cold email for my service.
Result: a generic cold email template with [YOUR NAME] and [COMPANY NAME] placeholders that has never convinced anyone of anything.
After (with context):
I run a freelance UX research service. I'm targeting product managers at mid-size
SaaS companies (50–500 employees). My differentiator is fast turnaround — I deliver
research insights within 5 business days, versus the typical 3–4 weeks with agencies.
The goal of this email is to get a 20-minute introductory call, not to close a deal.
The recipient has never heard of me. They're busy and receive a lot of vendor outreach.
Write a cold email that's under 120 words, leads with a relevant pain point (slow
research cycles blocking product decisions), and ends with a specific, low-commitment CTA.
Same task. Worlds apart in output quality.
Building a Context Habit
The fastest way to improve your prompting isn't to learn more techniques. It's to make context your default first step.
Before you type your actual request, write two sentences:
- Who you are (or who this is for)
- What you're actually trying to accomplish
That alone will improve 80% of your results.
As you get comfortable, add: 3. Any constraints the model should know about 4. Relevant background it can't guess
Key Takeaways
- AI models can only use what you give them — they don't infer intent
- The five context types: who you are, who it's for, your real goal, constraints, and background information
- Use "minimum viable context" — what would the model get wrong without this?
- Make giving context your first habit, not an afterthought
Next up: now that you know what to tell the AI, learn how to iterate when the first response still isn't right →