Shaping the request · Lesson 3 of 8
Zero-shot prompting
Get good results from instructions alone, with no examples — and recognise when a task needs more than a zero-shot prompt.
You're learning as a guest — everything is open. Sign in to save progress, earn XP and pick up where you left off.
What "zero-shot" means
A zero-shot prompt asks the AI to do a task without showing it any examples of the answer you want. "Translate this into French," "Classify this review as positive or negative," and "Summarise this article" are all zero-shot prompts. The "shot" refers to an example; zero shots means none.
Most everyday prompting is zero-shot, and that's fine. Modern language models have seen so many tasks during training that they can often handle common requests from a plain description. The skill is in writing that description well — and knowing when it isn't enough.
Why zero-shot often works
Common tasks like summarising, translating, rewriting, explaining and brainstorming are familiar to the model. You don't need to demonstrate what a summary is. What you do need to supply is everything that's specific to you: the audience, the length, the format, the categories, the edge cases.
Is this review positive? "Delivery was late but honestly the blender is great."
Classify the customer review below as exactly one of: POSITIVE, NEGATIVE, or MIXED. - Use MIXED when the review contains clear praise and clear complaints. - Reply with the label only, no explanation. Review: "Delivery was late but honestly the blender is great."
Both prompts are zero-shot, but the second defines the labels, explains the tricky case and fixes the output format. That's the heart of good zero-shot prompting: replace examples with precise instructions.
A checklist for strong zero-shot prompts
- Name the task with a clear verb — classify, summarise, extract, rewrite, compare.
- Define the output space — the allowed labels, the number of items, the length.
- Handle the edge cases you can predict — what to do with mixed, missing or unclear input.
- Specify the format — label only, a bullet list, a single sentence.
- Add context that matters — who will read it and why.
Extract every date and deadline from the email below. Return a bulleted list in this form: - [date as written] — [what happens on that date] If a deadline is relative (like "next Friday"), write it as written and add "(relative)". If there are no dates, reply "No dates found." Email: """ [paste email] """
When zero-shot falls short
Zero-shot struggles when the thing you want is easier to show than to describe:
- A particular style or voice — "write like our brand" is hard to define in words.
- Unusual or custom formats — a house style for product descriptions, a specific JSON shape.
- Subjective labels — what your team means by "urgent" or "high quality."
- Consistency across many items — classifying 200 support tickets the same way every time.
If you've rewritten a zero-shot prompt several times and the results still drift, that's a strong signal to add examples. That's few-shot prompting, covered in the next lesson.
Tip: Start zero-shot. It's faster to write, uses fewer words and avoids the risk of the AI over-copying your examples. Move to few-shot only when you can see a specific gap.
Asking for reasoning in a zero-shot prompt
For multi-step problems — a tricky word problem, a scheduling puzzle, a pricing comparison — adding a line like "Work through this step by step before giving your final answer" can improve results. It gives the model room to lay out intermediate steps rather than jumping straight to an answer.
Note: The written steps are generated text, not a guaranteed view of how the model actually reached its answer. They're useful for spotting mistakes, but check the final result on its own merits. Some AI tools also do their own hidden reasoning, so the extra instruction may matter less there.
Common mistakes
Mistake: Assuming zero-shot means "short and vague."
Fix: Keep the no-examples approach but make the instructions precise: labels, format, edge cases.
Mistake: Leaving the allowed answers open when you need consistent labels.
Fix: List the exact categories and say what to do with ambiguous cases.
Mistake: Endlessly rewording a zero-shot prompt for a style it can't capture.
Fix: If you can show it more easily than describe it, add one to three examples.
Practice exercise
Sharpen a zero-shot classifier
You manage a shared inbox and want the AI to sort incoming messages. Improve the starter prompt so it stays zero-shot (no examples) but produces a consistent, usable label every time.
Your draft stays in this browser tab — it isn't sent anywhere.Try your draft in the Playground
Hints
You can do the whole exercise as a guest.Sign in to save progress
Knowledge check