Zero-shot prompting
Also called: zero-shot · direct prompting
By HubAI editorial team · Source checked: 24 October 2026What does it mean?
Asking a model to complete a task without supplying a worked example of the desired response in that prompt. You can still provide detailed instructions and source material.
Picture this
You ask an assistant to label a fictional enquiry as Billing, Delivery or Other, explaining the labels but showing no previously labelled enquiries. Adding sample input-output pairs would make it an example-based prompt.
Where you’ll come across it
Trying a clearly described task before deciding whether examples are needed. State your allowed labels, expected format and what to do when the supplied information is insufficient.
What it doesn’t mean
Zero examples in your prompt does not mean zero prior training. The model has learned from other material, and a direct request does not establish that its response is correct.
Start without examples, then change one thing
A short exercise · use fictional or non-sensitive material.
- Invent a short enquiry and ask for one of three clearly defined labels. Supply no worked example in the first request.
- Make a second request with the same task and a few labelled examples. Keep a different enquiry for testing.
- Compare the labels and output format. Check an ambiguous case too; do not infer overall reliability from one success.
Did examples improve the task you care about, or only make the formatting more consistent?
Did that make it clearer?
Likes are a private on-page choice, not a public rating. For a correction or complaint, send the pre-filled form to our editorial inbox. Nothing is sent until you submit it.Check the source
Written and explained by HubAI. The reference below lets you go deeper; it is not an endorsement of a particular product.
Technical reference: developers.google.com ↗