Sources checked
What should an AI brief include?
Google’s prompt design guide emphasizes clear instructions and relevant examples. OpenAI’s model optimization guidance places prompting inside a process of evaluation and revision.
The template below is an original AiLookout worksheet. It is intended for everyday drafting and research; it is not a guarantee that any model will obey every instruction.
| Part | What to supply | Example |
|---|---|---|
| Task | The work to perform | Summarize this announcement for a nontechnical reader |
| Context | The relevant source material | Use the attached announcement only |
| Constraints | Boundaries and required qualifications | Flag missing facts; do not invent availability |
| Output | The shape of the result | One short answer and three distinct takeaways |
| Check | A testable completion rule | Every factual claim must be supported by the source |
Before and after: a news-summary prompt
Vague brief: “Write about this new AI tool.” This leaves the audience, sources, length, and treatment of uncertainty open.
Clearer brief: “Using only the announcement below, explain what changed for someone deciding whether to try the tool. Write a two-sentence answer and three distinct takeaways. Preserve availability limits. Attribute performance claims to the company. If a required fact is missing, say so.”
This is a hypothetical example, not a prompt tested across models. Its value is that each requirement can be checked against the result.
How do you improve a prompt?
Save a few representative inputs and define acceptable outputs. If the answer misses a limit, strengthen that specific requirement. If it invents facts, inspect both the evidence supplied and the request to handle gaps.
- Run the brief on a representative example.
- Label the failure: missing context, unclear task, wrong format, or unsupported claim.
- Change the relevant instruction or source material.
- Repeat with the same example and an unseen example.
- Keep the version that improves the actual task.
When is prompting not enough?
A prompt cannot supply facts you have not provided or verify an external action by itself. Add suitable sources or tools when the task requires them, and inspect the result.
Longer instructions can introduce contradictions. Remove requirements that do not help the reader or evaluator make a decision.
Common questions
Do I need a special expert role in every prompt?
No. A role can set context, but a concrete task, evidence, and completion rule are more useful than an unsupported claim of expertise.
Should I use the same prompt in every assistant?
Start with the same task brief for a fair comparison. Adjust it only when a particular application’s tools or supported inputs require a change, and record that difference.
What to remember
Make success observable. A short, specific brief is easier to test and improve than a long list of vague demands.
Sources & further reading
How this story was made
Written by Kristian Kostov with AI assistance and checked against the linked sources. Company performance claims are attributed to the company. Analysis reflects AiLookout’s interpretation; we have not independently tested the products discussed. Cover photography is illustrative and does not depict the specific announcement or product.
Our editorial standards
