🤖 Part 5: Giving Context & Examples

If the four ingredients from Part 4 are the frame, context is the picture inside it. The more an assistant knows about your situation — and the more you can show it what "good" looks like instead of just describing it — the closer its first answer lands to what you actually wanted. This part is about feeding the assistant well: the right background, and the single most underused move in all of prompting, the worked example.

📚 What You'll Learn

By the end of this part, you'll be able to:

Supply the context that matters — background, audience, tone, length, and constraints — without drowning the answer in noise
Use examples (few-shot prompting) to lock in a style: "here are two, now do a third the same way"
Give the assistant a persona or voice and reference material to work from
Recognize when too much irrelevant context hurts, and trim to just what's needed

In This Part

Context Is King

An assistant knows a great deal about the world in general and almost nothing about you in particular. It doesn't know your job, your audience, the tone your company uses, the deadline you're up against, or the three things your boss always complains about. Every one of those is a fact that would change the answer — and it's invisible until you say it. Context is simply the act of handing over the facts that only you have.

The remarkable thing is how little it takes to transform an answer. "Write a product description" is a shot in the dark. "Write a product description for a $12 hand-poured soy candle called 'Rainy Sunday,' sold on Etsy to people who love cozy, quiet moments; warm and a little poetic, under 60 words" is a bullseye. Same task — but now the assistant knows the price point, the platform, the buyer, the mood, and the length. It had all the skill it needed the whole time; it was just missing the facts.

Kind of context	What to tell it	Example phrase
Audience	Who will read or use this?	"for first-time home buyers who feel overwhelmed"
Background	The situation and any facts that bear on it	"we're a two-person nonprofit and the event is next Saturday"
Tone	The feeling you want	"encouraging, plain-spoken, never salesy"
Length	How much output you want	"about 3 short paragraphs" or "under 100 words"
Constraints	The must-dos and must-not-dos	"no jargon, don't mention price, include a call to action"

📖 Definition

Context: everything you tell the assistant about your specific situation so it can tailor the answer — audience, background, tone, length, and constraints. It's the difference between a stranger's guess and a colleague's help. A colleague gives better help because they know more; context is how you turn the assistant into that colleague, one sentence at a time.

Show, Don't Just Tell: Examples

Here's the move most people never discover, and it's a genuine level-up. Instead of describing the style you want, show the assistant an example of it and ask it to match. Describing tone is hard — "warm but professional, punchy but not gimmicky" — and easy to misread. An example removes all doubt: the assistant sees the pattern and copies it.

📖 Definition

Few-shot prompting: giving the assistant one or more worked examples of what you want before asking it to produce the next one. "Zero-shot" is asking with no example; "one-shot" gives one; "few-shot" gives a handful. More examples generally mean a tighter match to your style — because you've shown, not just told.

Suppose you're writing short, upbeat descriptions for items in an online shop and you want them all to sound the same. Don't try to describe the "voice" in the abstract. Show it two you like:

✅ Few-shot in action

Here are two product blurbs in our shop's voice:
1. "Meet the Nomad Mug — chip-proof, road-trip-ready, and just the right size for a proper coffee. Your adventures deserve better than a paper cup."
2. "The Sunday Throw is the blanket you'll fight over. Impossibly soft, machine-washable, and warm without the weight. Consider yourself warned."

Now write a third blurb in the exact same voice for a stainless-steel water bottle called the 'Trailhead.' Keep it to two sentences.

Because it can see the pattern — playful, benefit-first, a cheeky closer — the third blurb comes out sounding like a sibling of the first two, not a stranger.

The same trick works for structure, not just voice. Want every answer formatted a certain way? Show one filled-in example and the assistant will mirror the layout for the rest. This is far more reliable than describing a format in words, and it's the backbone of the How-To below.

graph TD A["📝 You describe the style
⟨warm, punchy, benefit-first⟩"] --> B["🤔 Assistant guesses
what you mean"] C["📎 You SHOW one example
of the style"] --> D["🎯 Assistant copies
the visible pattern"] B --> E["😐 Close, but off"] D --> F["😀 Matches on the
first try"]

Persona, Voice & Reference Material

Two more ways to feed the assistant well round out this part.

🎭 Give it a persona

Beyond a simple "act as," you can describe a fuller persona to steer voice and priorities: "You're a no-nonsense but kind fitness coach who believes small habits beat big overhauls, hates fad diets, and always ends with one encouraging line." Now every answer carries that character. Personas are wonderful for anything you'll produce repeatedly — a consistent brand voice, a tutoring style, a customer-service tone — because you can reuse the same persona description at the top of each new chat.

📎 Paste in reference material

The assistant does its best work when it doesn't have to rely on memory. If your question is about a specific document, email thread, policy, or set of notes, paste it in and ask the assistant to work from that. "Here's our refund policy: [paste]. Draft a reply to this customer's complaint using only what the policy allows." Now the answer is grounded in your real material instead of a plausible guess — and it's far less likely to invent something. (For longer files and documents, Part 8 shows you how to upload them directly.)

✅ Tip — grounding beats guessing

Any time the right answer lives in a document you have, give the assistant that document and tell it to stick to it: "Base your answer only on the text I pasted; if it's not in there, say so." This one instruction both improves accuracy and gives the assistant permission to admit a gap instead of filling it with a confident guess — a direct antidote to the caution from Part 1.

⚠️ Important: Before you paste anything in, think about what's in it. Customer details, medical or financial information, passwords, anything confidential — pause. Grounding the assistant in your real material is powerful, but some material shouldn't leave your control. Part 13 covers privacy in full; until then, a good rule is: don't paste anything you'd be uncomfortable seeing stored on someone else's server.

When Context Backfires

More context helps — right up until it doesn't. The failure mode is irrelevant context: piling on background the assistant doesn't need, which buries the important details and pulls the answer off course. If you're asking for a two-line birthday message and you include three paragraphs about your family history, the assistant may earnestly try to work all of it in, and the result gets bloated and strange.

⚠️ Watch Out — signal vs noise

The goal isn't maximum context, it's relevant context. Every detail you add should plausibly change the answer. Ask yourself, "Would knowing this make the reply better?" If yes, keep it. If it's just color, cut it. A tight, on-point brief beats a long, rambling one — the assistant can only aim at the target you make visible, and noise blurs the target.

Helpful context (keep)	Noise (cut)
Who the message is for	Unrelated backstory that doesn't change the ask
The tone and length you want	Your whole life story when you need one paragraph
Constraints and must-avoids	Five past drafts when one clear example would do
The actual document to work from	Ten documents when only one is relevant

A practical habit: give a focused first brief, see the answer, and add context only where it fell short. That way you're feeding the assistant exactly the details it turned out to need, rather than guessing at everything up front. Refining like this is the whole of Part 6.

🛠️ How To: Lock In a Format With One Worked Example

What you'll do: use a single filled-in example to make the assistant produce every following item in the exact same shape — no fiddly format descriptions required.

Step by step

Decide the shape you want once. Pick the fields or structure — say, for meeting notes: Decision, Owner, Due date, Next step.
Fill it in by hand for one item. Write a single complete example exactly as you'd want every entry to look. This is your template.
Show it, then ask for more in the same shape. Paste your example and say "Follow this exact format for the notes below."
Hand over the raw material. Paste the messy meeting transcript or notes you want converted.
Check the first result against your example. If it drifts, point at the difference ("keep the Due date on its own line") — one nudge usually fixes it for the rest.

💡 Tip

Once a format sticks in a conversation, the assistant will keep using it for follow-ups in that same chat — so you only have to teach it once. If you want that format again next week, save your worked example; it's a reusable template you can paste into any new conversation.

📋 Templates & Examples

One-shot format lock — teach a structure with a single example:

Convert my raw meeting notes into action items. Use this exact format for each one:

Decision: Move launch to March 3
Owner: Priya
Due: Feb 20
Next step: Priya emails the vendor to confirm timing

Now do the same for these notes: [paste your notes]

Few-shot style match — teach a voice with two examples:

Write social captions in our brand voice. Here are two examples:
1. "New week, same great coffee. ☕ Swing by before the Monday rush hits."
2. "Rainy day? We've got a window seat with your name on it. 🌧️"

Now write three more in the same voice — short, warm, one emoji each — about our new autumn pumpkin latte.

Context-rich brief — all five kinds of context in one prompt:

Write a short announcement email.
Audience: our 300 newsletter subscribers, mostly hobby gardeners.
Background: we're launching a spring seed subscription box.
Tone: friendly and down-to-earth, a little playful.
Length: under 150 words.
Constraints: no hype words like "revolutionary," include one clear "Sign up" line at the end.

Grounding in reference material — work only from what you provide:

Here is our return policy: [paste the policy]. A customer wrote the message below asking for a refund after 40 days. Draft a polite reply using only what the policy allows. If the policy doesn't cover their case, say so and suggest they contact us directly. Customer message: [paste it]

Best Practices

✅ Do's

Show an example whenever style matters. One good sample beats a paragraph of describing the style you want.
Paste in the source material. Ground the answer in your real document instead of the assistant's memory.
State the audience and constraints explicitly. These reshape the answer more than almost anything else.
Give the assistant permission to say "not in the source." It reduces confident guessing.

❌ Don'ts

Don't bury the ask in irrelevant backstory — noise blurs the target and bloats the answer.
Don't describe a tone you could simply show — an example is clearer and faster.
Don't paste sensitive or confidential material until you've read Part 13 on privacy.
Don't assume it remembers context from a different chat — each new conversation starts fresh unless you re-supply it.

💡 Pro Tips

Keep a short "about me / about my work" blurb saved. Pasting it at the top of a new chat instantly gives the assistant the context it would otherwise lack.
When you love how something turned out, keep it as your example for next time — your best output becomes your best prompt.
If the answer wanders, the fix is usually to remove context, not add it. Try trimming to just the details that change the outcome.

📓 Learning Journal

Keep a journal as you work through this guide — digital or paper. After each part, jot down:

Key ideas you learned
Things that clicked for you
Questions or confusion points to revisit
Ideas you want to try
Your progress and how you feel about it

✍️ This part's prompt: Find something you've written that captures a style you like — an email, a caption, a note in your own voice. Paste it into an assistant as an example and ask it to write something new in the same voice. Then try the same request without the example, just describing the style in words. Compare the two results in your journal. How much did showing an example change the match? Was there any request where describing worked just as well as showing?

📝 Part Summary

🎓 Key Takeaways

Context is king. The assistant knows the world but not your situation — audience, background, tone, length, and constraints are the facts only you can supply.
Show, don't just tell. Giving one or more worked examples (few-shot) matches your style far more reliably than describing it.
Give it a persona for a consistent voice, and paste in reference material so answers are grounded in your real facts, not a guess.
More context helps only when it's relevant — irrelevant background buries the ask and pulls the answer off course.

🎉 What You Can Now Do

You can now feed an assistant the way a good manager briefs a talented new hire — with the right background, a clear example of what "good" looks like, and the real material to work from. That's the difference between answers that are technically fine and answers you can use as-is. You've also learned the restraint that separates pros from beginners: knowing what to leave out.

❓ Common Questions

How many examples should I give?

Often one is enough to lock in a format, and two or three tighten a style. Beyond a handful you get diminishing returns and you're just spending effort. Start with one, and add another only if the match isn't close enough.

Does the assistant remember context I gave it earlier?

Within the same conversation, yes — it keeps the thread, so you don't have to repeat yourself. A brand-new chat starts blank, though, so re-supply anything important. That's exactly why saving a reusable "about me" blurb and your favorite examples pays off.

Is pasting in my document safe?

For ordinary, non-sensitive material, it's one of the best things you can do — it grounds the answer in real facts. For anything confidential or personal, hold off until Part 13, which covers privacy, what gets stored, and how to keep sensitive information out of these tools.

🔭 Up Next

In Part 6: Iterating & Steering the Answer, we'll stop treating the first reply as the finished product. You'll learn to refine in place with quick follow-ups — shorter, warmer, "add a section on X," "give me three options" — and turn a decent draft into exactly what you wanted in a few turns.

📚 Additional Resources

🌟 Keep Going

You've unlocked the two moves that quietly do the most work in real, everyday prompting: giving good context and showing an example. Together they make the first answer land close to home far more often. Next, let's get great at the last mile — steering that first answer into exactly what you need. 🤖