AI Text Generate Step

Step Guide: AI Text Generate

Knowledge

Generate text or structured output with AI

Overview

The AI Text Generate step sends data to a large language model and returns text, structured JSON, or email content. It supports three actions: Text generates a single output string in Text, HTML, or Markdown format; Structured generates user-defined output fields parsed as JSON; and Email generates a subject line and HTML body with purpose-aware prompting. Supports OpenAI and Anthropic providers.

When to Use

Summarize, rewrite, or translate text from previous steps
Extract structured data from unstructured input (e.g., parse an email into fields)
Generate content like product descriptions, reports, or social media posts
Classify or score data based on criteria described in the prompt
Generate email content (subject line and body) with purpose-aware prompting

How It Works

The step builds a message to the LLM by combining your prompt with all input field values as a context block. Input fields are user-defined. You add whatever fields you need (e.g., company_name, revenue) and map them from upstream steps. The prompt is required, and execution fails if it's empty.

Actions

Text

Generate text, HTML, or Markdown with AI

How it works

Write a prompt with {{stepId.fieldName}} references to upstream data. Choose an output mode: Text (plain text), HTML (formatted HTML), or Markdown. The LLM returns a single string written to an "output" field. The difference between modes is the format you instruct the LLM to produce.

Use HTML mode when the output will be rendered as HTML (e.g., feeding into a Send Email body)
Use Markdown mode for content that will be displayed with formatting or converted later

Structured

Generate structured data with AI using user-defined output fields

How it works

Define custom output fields in the output field set. The JSON schema is passed to the LLM for structured generation, and the response is parsed into your custom fields. Use this when you need multiple distinct values from a single LLM call.

Name output fields clearly so the LLM understands what to generate for each one
Switching from Structured to Text mode replaces the output field set — you'll lose any custom output fields

Generate email subject line and body with AI

How it works

Configure a purpose (e.g., Marketing Campaign, Follow Up, Cold Outreach), describe the audience and what the email should communicate. The LLM generates a subject line and HTML body following best practices for the selected purpose. Optional settings control greeting lines, signatures, and additional guidance.

Wire the output to a Send Email step: {{stepId.subject_line}} for subject, {{stepId.body}} for body
The purpose setting controls default tone and length — override with additional guidance if needed
When multiple email steps run in sequence, previous subject lines are automatically provided to avoid duplication

Tips

Keep prompts specific. Describe exactly what output you want and in what format
Input fields are appended to the prompt as key-value context, so name them clearly so the LLM understands their meaning

Related Steps

Step Guide: Form

Collect user input with custom forms

Step Guide: Evaluate

Run calculations on numeric data

Step Guide: Send Email

2 actions

Send emails from Formt or your own custom domain