Skip to content

Technique

Beginner

Zero-shot prompting

Ask the AI to do a task directly, with clear instructions but no worked examples. The simplest technique and the right starting point.

What it is

Zero-shot prompting means asking a language model to perform a task without showing it any examples of the answer you want. You describe the task in plain language and rely on what the model already learned during training.

"Zero-shot" doesn't mean "low effort." A good zero-shot prompt is still specific about the task, the audience, and the format. It just skips the sample input/output pairs.

When to use it

  • Common tasks the model already does well: summarizing, translating, rewriting, explaining, brainstorming
  • You need a quick first draft and can refine it afterwards
  • The output format is easy to describe in a sentence or two
  • You don't have good examples on hand, or examples would bias the answer too much

How it works

Why it works

Language models have seen enormous amounts of text, including countless summaries, emails, explanations, and lists. When you ask for "a three-bullet summary for a busy manager," the model already has a strong sense of what that looks like. Zero-shot prompting leans on that built-in knowledge.

The catch: the model fills every gap you leave with the most typical answer. Vague prompts get generic results. So the skill in zero-shot prompting is describing the task precisely enough that "typical" matches what you actually want.

The anatomy of a strong zero-shot prompt

A reliable pattern is task + context + constraints + format:

Before

Summarize this article.

After

Summarize the article below for a busy sales manager who has 30 seconds. Give exactly 3 bullet points, each under 20 words, focused on what changes for our sales team. Skip background history. Article: """ [paste article] """

The improved version names the audience (busy sales manager), the purpose (what changes for the team), the length (3 bullets, under 20 words), and what to leave out. None of that requires an example; it just requires being explicit.

Separate instructions from content

When you paste text for the model to work on, wrap it in clear delimiters such as triple quotes or a labelled section. This keeps the model from confusing your instructions with the material, especially if the material itself contains instructions or questions.

Tip: If a zero-shot prompt keeps missing the style or format you want after one or two rewrites, that's your signal to switch to few-shot prompting and show an example.

Limits

Zero-shot works best on familiar tasks. For unusual formats, house styles, or subtle classification rules ("what counts as urgent in our support queue?"), descriptions alone often aren't enough, and the model will guess. Remember too that any answer can contain confident-sounding mistakes, so check facts that matter.

Step by step

  1. 1

    State the task in one clear sentence using an action verb (summarize, rewrite, list, explain).

  2. 2

    Add the context the model can't guess: audience, purpose, and any relevant background.

  3. 3

    Set constraints: length, tone, what to include and what to leave out.

  4. 4

    Specify the output format (bullets, table, short paragraph, subject line + body).

  5. 5

    Put any pasted material inside clear delimiters, after the instructions.

  6. 6

    Review the result and tighten whichever part of the prompt the model got wrong.

Weak vs strong

Weak prompt
Write something about our new vacation policy.
Why it falls short: No audience, purpose, length, or format. The model doesn't know the policy details, so it will invent generic or wrong content and pick a random tone.
Strong prompt
Write a short, friendly announcement for our company Slack channel about our new vacation policy. Audience: all employees. Key facts: unlimited PTO starts January 1; managers must approve requests 2 weeks ahead; the old carry-over balance will be paid out in March. Keep it under 120 words, use a warm but professional tone, and end with "Questions? Ask HR in #people-team."
Why it works: It names the channel, audience, facts, tone, length, and closing line. No example is needed because a short announcement is a familiar format the model handles well when the details are supplied.

Practical use cases

  • Quick summaries

    Condense meeting notes, articles, or long email threads into a few bullets for a specific reader.

  • Rewriting for tone

    Turn a blunt message into a polite one, or a technical explanation into plain English for a customer.

  • Explaining concepts

    Ask for an explanation of compound interest, a code error, or a medical term "for a 12-year-old" or "in two sentences."

  • Brainstorming

    Generate 10 name ideas, gift ideas, or blog topics with a few constraints on style and audience.

Practice exercise

Upgrade a lazy prompt. Start with: "Help me write a thank-you note."

  1. Rewrite it as a zero-shot prompt that names the recipient, the reason for thanks, the tone, and the length.
  2. Run both versions in any AI assistant and compare the results.
  3. Note which added detail made the biggest difference, and keep that habit for your next prompt.
Practice in the PlaygroundAll techniques