Skip to main content

AI Smart Extraction

Describe what data you want in natural language, and AI automatically parses the page structure to return structured results.

AI Smart Extraction configuration panel

Configuration​

Variable Name​

Enter a variable name (e.g., products_table). Extraction results are stored in this variable and can be referenced downstream via {{variable_name}}.

Extraction Scope​

Limits which page area AI extracts from. Supports CSS selectors (e.g., .product-list) or semantic descriptions (e.g., "the product listing area"). Narrowing the scope prevents interference from other page regions.

Extraction Prompt​

Describe which fields to extract in natural language. Example:

Extract all products from the product list. Fields: product title, discounted price, original price, total sales, main image URL.

Three quick templates are built in — click to fill:

TemplateFields
🛍️ E-commerceProduct title, discounted price, original price, total sales, main image URL
💼 Job ListingsJob title, salary, company name, education requirements, benefits tags
💬 ReviewsUsername, review content, post time, interactions, IP location, sentiment

You can modify the prompt after selecting a template.

Preview​

Click the 👁 button next to the prompt input to preview extraction results on the current page in real time. Verify before running the full workflow.

Parameter Reference​

ParameterDescription
Variable NameRequired. Variable to store extraction results
Extraction ScopeRequired. CSS selector or semantic description to limit extraction area
Extraction PromptRequired. Natural language description of fields to extract

FAQ​

Extraction returns empty results​

  • Verify the "Extraction Scope" correctly targets the data container
  • Make the prompt more specific (e.g., "discounted price" rather than "price")
  • If data loads asynchronously, add a Wait Element node beforehand

Field names or types don't match expectations​

Specify field names explicitly in the prompt. Follow extraction with a Set Variable node and use the Data Transform Pipeline to convert types.