# Get more useful answers from the Agent

> Give a clear task, relevant evidence and boundaries the assistant can work with.

A good prompt explains what you are deciding and supplies the facts needed to help. You do not need complicated prompting language.

![Start a focused conversation in the message field and attach relevant evidence when needed.](https://sellify.ai/images/docs/walkthroughs/agent-new-chat.webp)

## Use four ingredients

1. **Product and customer:** what you sell and who you want to reach.
2. **Evidence:** verified features, real questions or actual performance numbers.
3. **Task:** the decision or output you need.
4. **Constraints:** what must stay accurate and what the answer should avoid assuming.

## Example: compare angles

> Here are three angles for this product: \[paste]. My customer is \[describe]. These are the verified product facts: \[facts]. Compare the angles by relevance and claim support. Tell me which assumptions I should check before making an ad.

## Example: review copy

> Review this headline and primary text: \[copy]. Flag unclear wording, unsupported product claims and a mismatch with this offer: \[real terms]. Suggest one revision that preserves the facts.

## Example: interpret a test

> During \[date range], version A had \[figures] and version B had \[figures]. The budget, audience and destination were \[details]. What can I reasonably learn, and what information is missing?

Use real sales figures and clearly label any information you do not have. Check that the answer follows from the data you supplied.

## Improve the next reply

If an answer misses the point, identify the gap: “That assumes insulation, which this bottle does not have. Use only capacity, material and the carry loop.” Ask for a concise revision rather than restarting a useful conversation unnecessarily.

Keep promises and claims grounded in the product. Use [support](https://sellify.ai/docs/support) for account or software failures that require investigation.