> ## Documentation Index
> Fetch the complete documentation index at: https://docs.asserto.ai/llms.txt
> Use this file to discover all available pages before exploring further.
# Quickstart – Test LLM Prompts Structured Output with Asserto AI
> Create a prompt and test cases in 5 minutes
In this quickstart we will create a prompt for classifying online reviews of movies as either `positive` or `negative`.
*Watch our quickstart demo to see how to create and test your first prompt in Asserto AI*
## 1. Create Your First Prompt
* Navigate to Prompt list page ([https://app.asserto.ai/app/prompt](https://app.asserto.ai/app/prompt))
* Set name to `Movie Review Sentiment`
You will see the Prompt Development Dashboard.
***
## 2. Define the Prompt
In the left panel of the dashboard, you can define the messages that make up the prompt.
Define the LLM's role and instructions. Specify the expected output format to make the response machine-readable.
```plaintext theme={null}
You are a sentiment classifier.
Classify the following movie review as 'positive' or 'negative'.
Respond in JSON format:
{"sentiment": ""}
```
Write a template for the user input using a placeholder for the review text.
Templates use Mustache syntax — wrap parameter names in double curly braces (`{{ }}`).
```mustache theme={null}
Review: {{review}}
```
## 3. Configure the Prompt
In the right-side panel, go to the **Config** tab. Here you'll configure how the LLM runs and define the inputs and outputs.
* Select a model (e.g., `gpt-4`, `claude`, etc.)
* Set temperature to `0` for consistent deterministic outputs
* Ensure **response format** is set to `JSON`
* Add a parameter named `review`
* Set its **type** to `text` to allow multi-line input
* This corresponds to the `{{review}}` placeholder in your prompt
* Add a result field named `sentiment`
* This matches the key in the expected LLM JSON response
* Used for validation and test automation
You can try this input in the playground later:
```json theme={null}
{
"review": "The plot was confusing, the pacing was terrible, and the acting was wooden."
}
```
## 4. Run Prompt in Playground
On the right-side panel, switch to the **Playground** tab.
Here you can run the prompt manually, inspect the LLM response, and make prompt edits.
Provide a sample value for the `review` parameter. For example:
**Positive review**
```json theme={null}
{
"review": "A visually stunning masterpiece with a moving story and brilliant acting."
}
```
* Click **Submit**
* Review the output JSON response
* Iterate on prompt instructions if necessary for better accuracy
Try different inputs to validate how well the prompt handles variations.
## 5. Add a Test Case
On the right-side panel, switch to the **Test Cases** tab.
Test Cases allow you to define fixed inputs and expected outputs that can be validated automatically. This is useful for regression testing and prompt versioning.
* Click **Add Case**
* Fill the name: `positive 1`
* For parameter values, set `review` to:
```text theme={null}
A visually stunning masterpiece with a moving story and brilliant acting.
```
* Click **Save**
* After saving the test case, click **Add Assertion**
* Select **property** `sentiment` as the output field to check
* Set the **operation** to `equal`
* Set **Expected** text output to `positive`
* Save the assertion
You now have a fully defined test case that will automatically validate that the prompt produces the expected output for the given input.
Repeat the steps to add a negative test case as well.
## 6. Run Test Cases
On the right-side panel, switch to the **Test Runner** tab to validate your prompt automatically using the test cases you've defined.
Click the **Run All** button to execute all your test cases.
Each test case will run the LLM with the given input and compare the output against the expected assertion(s).
If this is your first time running tests or if the API key is missing/expired:
* You'll be prompted to set the API key for your LLM provider (e.g., OpenAI, Anthropic)
* You can find this in the **Settings** panel if you need to update it later
* The test run session will show the **pass/fail status** of each test
* For failures, you can click to inspect the actual vs expected outputs
* Use this feedback to iterate on your prompt or test assertions
## 7. Next Steps
You now have a working prompt with automated test coverage! Here’s what you can do next:
* Try your prompt with **different models or settings** (e.g., temperature, top-p)
* Add more **test cases** to cover edge cases and ensure consistency
* Add **negative cases**, ambiguous inputs, or malformed reviews
* Explore **different assertion types** beyond `equal` — e.g., `contains`, `regex`, \`in list- Modify the prompt instructions to improve classification accuracy
## 8. Dive Deeper
* Learn about prompt versioning
* Learn about deployment options to integrate the prompts in your system
* Learn more about test cases and assertion types