Few-shot prompting means putting a few examples of what you want inside your message to the AI, so it can copy the pattern. Zero-shot prompting means asking with no examples at all and relying on the AI to work it out from your instructions.
Imagine asking a new helper to label jars. You can just say “label these jars” (zero-shot). Or you can hand them three jars you already labelled and say “like these” (few-shot). The second way, they match your style first time. A “shot” here simply means one example.
In a sentence: “The summaries kept coming out too long, so I switched to few-shot prompting and pasted in three short ones I liked.”
What the words mean
The plain point: zero-shot is instructions only. Few-shot is instructions plus a handful of worked examples.
Google’s glossary defines both in one line each: zero-shot is to “Provide a prompt without any examples and rely on the model’s pre-existing knowledge”; few-shot is to “Provide multiple examples in the prompt to demonstrate the pattern or task that you want.”1
OpenAI adds the key point: the examples steer the AI without retraining it.2 The idea became famous with a 2020 research paper titled “Language Models are Few-Shot Learners”.3
A simple example
The plain point: the examples show the format and tone better than a description can.
Note: “Parcel came two days late but the jumper is lovely.”
Summary: Late delivery, happy with product.
Note: “Wrong size sent, I need a medium.”
Summary: Wrong size, wants exchange.
Note: “[paste the next customer note]”
Summary:
Leave out the two worked examples and you have the zero-shot version. It may still work, but the summaries will vary more in length and style.
Why it matters to you
The plain point: examples are one of the simplest ways to get the output you pictured.
- Give a few, not one. Anthropic suggests three to five varied examples.4 One example is easy to copy too closely.
- Vary them. OpenAI advises showing “a diverse range of possible inputs with the desired outputs”, so the AI learns the pattern, not one case.5
- Reasoning models may not need them. A reasoning model works a problem through before answering. OpenAI says to start without examples on these.6
If the AI keeps getting the format wrong, stop describing it and show it two or three examples instead.
Where the word comes from
The words come from machine-learning research, and the 2020 GPT-3 paper by OpenAI researchers carried them into prompting. It set out three terms by how many examples the model sees: “Few-Shot (FS) is the term we will use in this work to refer to the setting where the model is given a few demonstrations of the task”, one-shot allows “only one demonstration”, and zero-shot gives “only a natural language instruction describing the task.” The paper adds that the name “is related to few-shot learning as used in other contexts in ML”, the field’s older term for learning a task from very few examples.7
Often confused with
- Fine-tuning: retraining the model on hundreds or thousands of examples so it changes for good. Few-shot examples last only for that one message. See fine-tuning meaning.
- Templates: a fill-in-the-blanks prompt. A template can contain few-shot examples, but it does not have to.
- “Shots” in video tools: in AI video, a shot is a camera shot. That is a different word that happens to look the same.
Questions people ask
What is the difference between zero-shot and few-shot prompting?
Zero-shot is instructions only; few-shot adds a handful of examples. Google’s glossary puts it as: zero-shot means “Provide a prompt without any examples and rely on the model’s pre-existing knowledge”, and few-shot means “Provide multiple examples in the prompt to demonstrate the pattern or task that you want.”1 See zero-shot prompting.
How many examples should a few-shot prompt have?
A few, and varied. Anthropic’s advice for Claude is: “Include 3–5 examples for best results.”4 More examples make your message longer, so stop once the answers come out the way you want.
What is one-shot prompting?
One-shot means giving exactly one example. The GPT-3 paper describes it as “the same as few-shot except that only one demonstration is allowed”.7 One example is quick to write, but the AI may copy it too closely, which is why the advice above suggests several.
Learn more on this site
- Vendor advice side by side: Prompting for LLMs: ChatGPT, Claude and Gemini.
- The “one example is enough” myth: Prompting myths.
- Prompt basics: prompt meaning and Prompting 101.
- The basics: the AI glossary and AI explained for complete beginners.
- Related: Prompting Techniques Explained.
QUICK CHECK · 3 QUESTIONS
Pick an answer to see if you got it. Nothing is recorded.
1. What is few-shot prompting?
- Putting a few examples of what you want inside your message
- Asking the AI the same question a few times
- Retraining the model on thousands of examples
Show the answer
Answer: Putting a few examples of what you want inside your message. Few-shot prompting means adding a handful of worked examples to your message so the AI can copy the pattern.
2. How many examples does Anthropic suggest for a few-shot prompt?
- Exactly one, so it is easy to copy
- At least a hundred
- Three to five varied examples
Show the answer
Answer: Three to five varied examples. Anthropic suggests three to five varied examples; one example is easy to copy too closely.
3. How is few-shot prompting different from fine-tuning?
- Few-shot prompting changes the model for good
- Few-shot examples last only for that message; fine-tuning retrains the model
- They are two names for the same thing
Show the answer
Answer: Few-shot examples last only for that message; fine-tuning retrains the model. Fine-tuning retrains the model so it changes for good, while few-shot examples steer it only for that one message.
Related terms: Context rot · Context window · Deep research · Grounding · Knowledge cutoff · Latency
- 1Google Cloud, Generative AI glossary, last updated 24 Sep 2026, read at source 1 Oct 2026zero-shot is to “Provide a prompt without any examples and rely on the model’s pre-existing knowledge”; few-shot is to “Provide multiple examples in the prompt to demonstrate the pattern or task that you want.” ↑azero-shot is to “Provide a prompt without any examples and rely on the model’s pre-existing knowledge”; few-shot is to “Provide multiple examples in the prompt to demonstrate the pattern or task that you want.” ↑b
- 2OpenAI, Prompt engineering, undated guide, read at source 1 Oct 2026“Few-shot learning lets you steer a large language model toward a new task by including a handful of input/output examples in the prompt, rather than fine-tuning the model.” ↑
- 3Brown et al., Language Models are Few-Shot Learners, arXiv:2005.14165, 28 May 2020, read at source 1 Oct 2026“tasks and few-shot demonstrations specified purely via text interaction with the model.” ↑
- 4Anthropic, Prompting best practices, undated documentation, read at source 1 Oct 2026“Include 3–5 examples for best results.” ↑a“Include 3–5 examples for best results.” ↑b
- 5OpenAI, Prompt engineering, undated guide, read at source 1 Oct 2026.↑
- 6OpenAI, Reasoning best practices, undated guide, read at source 1 Oct 2026“Try zero shot first, then few shot if needed”. ↑
- 7Brown et al. (OpenAI), Language Models are Few-Shot Learners, arXiv:2005.14165, 28 May 2020, full text (v4) read at source 12 Oct 2026“Few-Shot (FS) is the term we will use in this work to refer to the setting where the model is given a few demonstrations of the task at inference time as conditioning [RWC+19], but no weight updates are allowed.” “One-Shot (1S) is the same as few-shot except that only one demonstration is allowed”; “Zero-Shot (0S) is the same as one-shot except that no demonstrations are allowed, and the model is only given a natural language instruction describing the task.” and “few-shot learning as described here for language models is related to few-shot learning as used in other contexts in ML”. ↑a“One-Shot (1S) is the same as few-shot except that only one demonstration is allowed”. ↑b
Brown et al., Language Models are Few-Shot Learners, arXiv:2005.14165, 28 May 2020, read at source 1 Oct 2026.
The comparison on this page (labelling jars for a new helper) and the customer-note example are this site’s own, not claims from these sources.
Free printable AI cheat sheets, and AI news every week.
Sixteen one-page AI guides and checklists in one free PDF, no sign-up. And if you want to keep up, the newsletter covers what changed in AI each week, in plain English, with the sources.
SUBSCRIBE FREE ↗ GET THE CHEAT SHEET PACK