The useful result
Improve a bland output by giving one or two real examples and testing on a new input.
- Improve a bland output by giving one or two real examples and testing on a new input.
You ask for “a punchy product description” and receive the same punchy product description everyone else gets. Adjectives like punchy, friendly or professional mean different things to different readers, and a model falls back on the average of all of them.
The quickest cure is to show rather than tell. Google’s prompting guide says to always include examples, to keep their structure and format consistent and warns that too many can make the model overfit to them. Anthropic suggests a small set of relevant, varied examples. One or two real samples of your best work usually beats a paragraph of adjectives.
The examples below are illustrative exercises, not recorded results from any tool. Sources were checked on 24 October 2026; menus, limits and prices change, so confirm details on the linked pages.
Before you start
- A prompt that gives generic results
- One or two examples of output you are proud of
- A new input to test with
Review evidence
Documentation-based review · 2026-10-24
Primary documentation and official guidance checked on 24 October 2026. Examples, checks and exercises are original illustrations, not recorded tool tests. Cover artwork is a conceptual illustration.
Name what “generic” means for this task
Look at the bland output and say what you would change: too long, too formal, no concrete detail, wrong structure. Write one sentence for each. This tells you what your example must demonstrate.
Check your result
You have one sentence for each thing you would change.
Add one or two real examples in the same format
Pick one or two of your best real outputs and put them in the prompt with the same structure each time, labelled as examples. Say what they show, such as short sentences and a specific detail, and ask for a new output for the new input in the same format.
Write [the output] for the input below. Match the style, length and structure of these examples, but do not copy their wording or details.
Example input: [input 1]
Example output: [output 1]
Example input: [input 2]
Example output: [output 2]
New input: [your new input]
New output:Example / expected result
Illustrative example; not a recorded model result.
Two shop product descriptions of 40 words with one concrete detail each, then a third product as the new input.
Check your result
Your examples share one format and your prompt says what they demonstrate.
Test on a new input and watch for copying
Run the prompt on an input the examples did not cover. Check that the style matches and that the model has not copied phrases or details from the examples. If it copies, vary the examples or use a different one.
Check your result
The new output matches the examples’ style and repeats none of their specific wording.
When it goes wrong
Examples can lock the output into one pattern, so vary them. Do not include private or confidential text as examples. Too many examples make the model overfit and use more of your allowance.
Make it your own
Take a prompt you use often and add two examples. Compare five outputs before and after, and count how many you would use without editing.
Useful next resources
Sources and review
Last checked Oct 24, 2026. Features and account limits can change. Verify important results against your original material.
Did this help you complete your task?
Tell us what needs improving. No account or private output required.
Report incorrect information or a broken linkGenius Forges Editorial
Practical AI lessons with checked sources, labelled examples and human verification.
