Extract Variables

Use AI to pull specific, named values out of a block of text based on extraction instructions you define.

What It Does

Reads the text you provide and, for each variable you define, applies your extraction instructions to pull out a matching value. Each variable becomes its own output field, named whatever you called it. A value that cannot be found is returned as an empty string.

When To Use It

Use this to turn unstructured text, like a meeting transcript or an email, into individual workflow variables you can reference downstream, for example extracting a company name, a requested deadline, or a specific dollar amount. If you only need a single AI-generated decision or block of text, Run Prompt is simpler.

This node has no implicit context. The text to extract from must be passed in explicitly through Extraction Input, for example {{$triggerNodeId.meetingTranscript}}.

Configuration

Configure the node with these settings:

Field	What it controls
Model	Which LLM performs the extraction. Choose `default`, `lite`, `creative`, `analytical`, or a specific model. Defaults to `default`.
Extraction Input	The text to extract variables from. Can include workflow variables in mustache format. Required.
Variables to Extract	One or more variables to pull out, each with a `Variable Name` and `Instructions`. Required, at least one. Variable names can only contain letters, numbers, underscores, and hyphens.

Outputs

The node passes these values downstream. Check the Outputs tab on the configured node to confirm exact paths.

Configured variables

Variable	Type	Description	Example
`example_variable`	`string`	Example extracted value	`{{nodeId.example_variable}}`

Example Workflow

Pass a sales call transcript into Extract Variables with variables like budgetMentioned and decisionTimeline, then use those values to update CRM fields later in the workflow.

Run Prompt
Assign Property Value
Outcome Validator

Need More Help?

Reach the AskElephant support team by:

clicking the Support tab on the right side of your screen,
emailing [email protected],
or using @askelephant support in your dedicated Slack channel.