Zero-Shot Prompting

Zero-shot prompting means asking the model to do something without providing any examples of what you want.

It's the default — what most people do when they first open Claude or ChatGPT. The word "zero-shot" just gives a name to this baseline approach.

When it works

Zero-shot works well for tasks the model has seen many times in training: summarizing text, explaining a concept, writing a generic cover letter, answering factual questions.

Clear task + enough context = zero-shot works.

Vague task + no context = zero-shot produces whatever the model guesses you meant.

Compare:

• "Make it better." → the model has no idea what "it" is or better by what standard
• "Rewrite this paragraph in a more casual, conversational tone. Keep it under 100 words: [paste paragraph]" → specific, actionable, context-provided

The second prompt is still zero-shot — no examples — but it's much more likely to return something useful.

The three AI platforms

Most of this camp uses three models. They're free to sign up:

Create accounts on all three. Send "Hello!" to each and confirm you get a response.

Comparing models

One of the most useful things you can do with zero-shot prompts is run the same prompt on all three models and compare the results. Different models have different defaults around tone, length, formatting, and accuracy.

Try a few prompts from this list, each on all three platforms:

• "Explain compound interest to a 10-year-old in under 100 words."
• "Give me five startup ideas for reducing food waste in a high school cafeteria."
• "Write the opening paragraph of a cover letter for a first-time barista job."
• "Suggest a four-colour palette for a website about the night sky. Give me the hex codes."
• "Write a short rhyming rap verse about the water cycle."

For each, jot down:

• Which model felt most useful for that task?
• Any differences in tone, length, or formatting?
• Did any model produce something that seemed wrong?

There's no universal answer. The right model depends on the task, your rate limits, and your own preferences — and those preferences are worth forming from actual evidence.

What zero-shot can't do

Zero-shot has no way to know your preferred style, format, or audience unless you describe them. When you need output in a specific shape — a particular HTML pattern, a defined JSON structure, a tone that matches something you've already written — you need to show an example. That's one-shot.