# Prompt Step

The **Prompt** step runs a custom instruction against an LLM, with the lead's data as context. Use it to classify leads, generate personalized content, research accounts, or extract structured information for use in downstream steps.

<figure><img src="/files/vx76GHO3GTkxu2TnAboP" alt=""><figcaption><p>The Prompt step configuration panel with instructions, field definitions, and tools.</p></figcaption></figure>

## Configuration

### Instructions

Write the prompt that the LLM will execute. Instructions support variables — use the **Insert menu** (⚡, Variables tab only) to reference lead data fields.

Be specific about what you want the model to produce. If you're extracting structured data, describe the expected output clearly.

**Example instruction:**

```
Based on the lead's job title ({{ job_title }}), company size ({{ employee_count }} employees),
and the pages they visited on our website ({{ pages_visited }}), classify their buying intent
as one of: HIGH, MEDIUM, or LOW.

Return only the classification word with no explanation.
```

### Tools

Optionally enable one or both tools to give the LLM access to live information:

| Tool           | What it does                                                                |
| -------------- | --------------------------------------------------------------------------- |
| **Web Search** | Allows the LLM to search the internet to research the lead or their company |
| **Fetch URL**  | Allows the LLM to retrieve the contents of a specific URL                   |

These tools are useful for account research steps — for example, looking up a company's recent news or fetching their pricing page.

### Field Extraction

Save the LLM's output (or parts of it) into lead fields for use in downstream steps. These become available as workflow-scoped variables.

Add one or more **Fields**:

| Field           | Description                                                  |
| --------------- | ------------------------------------------------------------ |
| **Name**        | The field name (e.g., `buyer_intent`). Use snake\_case.      |
| **Description** | Describe what value should be extracted from the LLM output. |

## Example: Intent Classification + Personalized Email

{% stepper %}
{% step %}
**Prompt step — classify intent**

**Instructions:**

```
Review {{ first_name }}'s activity: they work at {{ company_name }} ({{ employee_count }} employees)
as a {{ job_title }} and visited {{ pages_visited }}. Rate their buying intent: HIGH, MEDIUM, or LOW.
```

**Field Extraction:**

* Name: `buyer_intent`
* Description: The intent classification (HIGH, MEDIUM, or LOW)
  {% endstep %}

{% step %}
**Condition step — branch by intent**

* Branch 1: `buyer_intent` equals `HIGH` → Call Phone
* Branch 2: `buyer_intent` equals `MEDIUM` → Send Email
* Branch 3: (catch-all) → Wait 7 days, then Send Email
  {% endstep %}

{% step %}
**Send Email step — reference the extracted field**

Subject: `Following up, {{ first_name }}`

Body: `Hi {{ first_name }}, based on your interest in our platform...`

The email content can vary per branch, using `{{ buyer_intent }}` if needed.
{% endstep %}
{% endstepper %}

## Tips

{% hint style="info" %}
**Be explicit in instructions.** LLMs produce more reliable structured output when you specify the exact format you expect. For classification tasks, provide the list of valid values.
{% endhint %}

{% hint style="info" %}
**Use Web Search for account research.** Enabling Web Search lets the model pull current information about a lead's company — recent funding, news, tech stack — to personalize downstream messaging.
{% endhint %}


---

# Agent Instructions: Querying This Documentation

If you need additional information that is not directly available in this page, you can query the documentation dynamically by asking a question.

Perform an HTTP GET request on the current page URL with the `ask` query parameter:

```
GET https://docs.spara.com/agents/workflows/prompt.md?ask=<question>
```

The question should be specific, self-contained, and written in natural language.
The response will contain a direct answer to the question and relevant excerpts and sources from the documentation.

Use this mechanism when the answer is not explicitly present in the current page, you need clarification or additional context, or you want to retrieve related documentation sections.