1.2 Few-Shot Prompting

Tip

Few-shot prompting includes examples in your prompt to teach the model what you expect. You are essentially training the model for a specific task by providing sample input-output pairs. Especially suitable for classification and labelling tasks.

When to use it

Classification and labelling

You need the AI to sort, tag, or categorise items and you can show it a few correct examples.

Consistent formatting

You want the output to follow a specific pattern and showing examples is easier than describing the rules.

---

Best practices

1. Task – Clearly state what you want the model to do.
2. Context – Provide the relevant context necessary to carry out the task.
3. Demonstration – Give at least one example of what you want the model’s reply to look like.
4. Consistent formatting – Format your examples consistently so the model picks up the pattern.

---

Ready-to-use prompt

Prompt: Label NIOO publications by research theme

At the Netherlands Institute of Ecology (NIOO), all publications
are labelled with one or more of our research themes.
The themes are Biodiversity, Climate Change and Sustainable Use
of Land and Water.
In rare cases, we opt for 'No theme'.
The publications below have already been labelled with the
appropriate themes:

- https://doi.org/10.18194/ws.00345, Biodiversity
- https://doi.org/10.1111/geb.13901, Climate Change
- https://doi.org/10.1007/s13593-024-00991-3, Sustainable Use of Land and Water
- https://doi.org/10.1002/2688-8319.12303, Biodiversity & Climate Change
- https://doi.org/10.1016/j.scitotenv.2024.173699, Biodiversity & Sustainable Use of Land and Water
- https://doi.org/10.1016/j.watres.2024.122193, Climate Change & Sustainable Use of Land and Water
- https://doi.org/10.3390/microorganisms12040760, Biodiversity & Climate Change & Sustainable Use of Land and Water
- https://doi.org/10.1186/s40462-024-00489-3, No theme

Assign one or more of NIOO's research theme(s) to this
publication: https://doi.org/10.1111/brv.13164.

Note

The examples at the top act as training data — the model learns the pattern and applies it to the new publication at the bottom.

---

Sources

Based on materials from Prompt Engineering Guide and Software Guide.