> ## Documentation Index
> Fetch the complete documentation index at: https://docs.getsmelt.io/llms.txt
> Use this file to discover all available pages before exploring further.

# Testing Templates

> Test your templates before running full jobs

Test your templates with real data before running full jobs to catch issues early.

## Why Test First

* ✅ Verify prompt produces good output
* ✅ Check quality flags for issues
* ✅ See cost estimates
* ✅ Catch variable problems
* ✅ Iterate quickly without wasting usage

<Warning>
  Template tests count toward your monthly usage! Each test consumes 1 row of your allowance.
</Warning>

## How to Test a Template

<Steps>
  <Step title="Open the Template">
    Go to Templates and click on the template you want to test
  </Step>

  <Step title="Select a Test File">
    Choose a file from your uploaded CSVs to use as test data
  </Step>

  <Step title="Click Test">
    Click the **Test** button
  </Step>

  <Step title="Review Results">
    See generated output, quality flags, token usage, and cost
  </Step>

  <Step title="Iterate if Needed">
    Adjust your prompt and test again until satisfied
  </Step>
</Steps>

## What You'll See in Test Results

| Field                | Description                                             |
| -------------------- | ------------------------------------------------------- |
| **Generated Output** | The AI's response for the test row                      |
| **Quality Flags**    | Any issues detected (too short, forbidden word, etc.)   |
| **Token Usage**      | Input and output tokens consumed                        |
| **Cost**             | Estimated cost for this generation                      |
| **Input Data**       | The row data that was used                              |
| **Built Prompt**     | The final prompt sent to the AI (with variables filled) |

## Testing Unsaved Templates

You can test a template **before saving it**:

1. Fill in the template form (name, prompt, etc.)
2. Select a test file
3. Click **Test**
4. Review results
5. Then save if satisfied

This lets you iterate on your prompt without creating multiple saved versions.

## Interpreting Test Results

### Good Results

* ✅ Output is relevant and personalized
* ✅ No quality flags (or only minor ones)
* ✅ Length is appropriate
* ✅ Matches your intended tone

### Warning Signs

* ⚠️ Output is too generic
* ⚠️ Quality flags present
* ⚠️ AI ignored your instructions
* ⚠️ Variables didn't resolve correctly

### What to Do If Results Are Poor

<AccordionGroup>
  <Accordion title="Output is too generic">
    Add more specific instructions. Include examples. Reference more data columns.
  </Accordion>

  <Accordion title="Output is too long/short">
    Add explicit length constraints: "Keep under 100 characters" or "Write 2-3 sentences"
  </Accordion>

  <Accordion title="AI ignored instructions">
    Make instructions clearer. Put them at the end of the prompt. Use "IMPORTANT:" prefix.
  </Accordion>

  <Accordion title="Variables show as empty">
    Check that column names match exactly. View your file to verify column names.
  </Accordion>

  <Accordion title="Contains forbidden words">
    The AI might not follow forbidden words perfectly. Make the prohibition clearer in your prompt too.
  </Accordion>
</AccordionGroup>

## Testing Multiple Rows

The test feature runs on sample rows from your file. This gives you a representative output without processing the entire file.

<Tip>
  If testing consistently produces good results, you can confidently run a full job.
</Tip>

## Cost of Testing

Each test:

* Uses 1 row of your monthly allowance
* Consumes tokens (billed if BYOK)
* Counts toward usage tracking

**Recommendation:** Test 3-5 times to iterate, then run a small batch (10-20 rows) as a final verification before processing thousands.