Skip to main content

Using the AI data processor

Learn when to use Process with AI for spreadsheet rows or document extraction, and how to write instructions that produce useful results.

Written by Jessica

The interface names the AI data processor Process with AI. Its role depends on the source: it can optionally transform rows in a Google Sheets source, or it can extract structured data from documents in a folder.

Choose the right AI path

Use the optional Process with AI processor when your source is a Google Sheets file and the data already exists in rows. It can transform that spreadsheet data based on your instructions.

Use the required Process with AI step for document extraction when a folder source has Content type set to Documents. On this path, Sheetgo reads PDF and Google Docs files and extracts their data. The AI step is selected automatically, cannot be changed, and no other processor can be used. See Extract data from PDF and text documents for the complete setup.

Process with AI is not available when a folder source has Content type set to Spreadsheets.

Process rows from a spreadsheet with AI

For a Google Sheets source, the Select a data processor step is marked Optional. You can choose Process with AI under TRANSFORM DATA or select Skip data processor when you do not need to transform the source data.

  1. Create a blank workflow.

  2. Choose Create an automation.

  3. Select Google Sheets as the source and configure the spreadsheet data you want to process.

  4. In the optional Select a data processor step, choose Process with AI.

  1. In the Process with AI panel, marked BETA, enter instructions that describe the task the AI should perform under INSTRUCTIONS.

  2. Use SOURCE COLUMNS to select a column from the source and add its header to the instructions without typing it manually.

  3. Under MOST POPULAR, choose a preset when one matches your task, then adjust its instructions for your data. The available presets are Summarize, Rewrite, Clean up & normalize, Categorize, Sort & rank, Detect sentiment, and Write email.

  1. Select a spreadsheet destination for the processed data.

  1. Select Finish.

By default, the AI data processor uses the GPT-4o mini model, which can process up to 128K tokens at a time. The panel also offers Connect now under "Connect your OpenAI account for unlimited AI processing" and explains: "Use your key to remove token and data size limits. It's fast, secure, and gives you full control." Connect an OpenAI account if you want to choose another AI model or access your GPTs.

Write effective instructions

Clear, specific instructions make the output more predictable.

  • Break a complex task into numbered steps or a short list.

  • Describe what the AI should do in positive terms. For example, write "Use descriptive language" instead of "Avoid filler words."

  • Explain the context and state the intended result precisely.

  • Use SOURCE COLUMNS to refer to source columns by their exact headers.

Test the automation with representative data. Adjust the instructions if the first result does not match what you need.

Extract data from documents with AI

For document extraction, start with a Folder source and set Content type to Documents. The second step is Process with AI, marked BETA, and states that AI processing is required for documents.

Enter extraction instructions in the INSTRUCTIONS box or start with one of these EXTRACTION PRESETS:

  • Invoices

  • Receipts

  • Contracts

  • Other documents

These document-extraction presets differ from the MOST POPULAR presets offered for Google Sheets rows: Summarize, Rewrite, Clean up & normalize, Categorize, Sort & rank, Detect sentiment, and Write email.

This required step extracts the data from each PDF or Google Docs file into structured rows in the destination spreadsheet. It is a different setup from using the optional AI data processor on rows that already exist in Google Sheets.

See Extract data from PDF and text documents for extraction instructions, output details, and the steps to combine document data.

FAQ

How many automation credits does the AI processor consume?

The basic AI processor consumes 1 automation credit per 100K tokens. When you use your own OpenAI account, it consumes 1 automation credit per processor.

How do I fix "Data is too large to be processed by AI: Try to use your own credentials and models."?

This message appears when a data source exceeds 128K tokens. Filter out data that the AI does not need to keep the input within this limit. You can also connect your own OpenAI account and choose a model that can process more tokens.

Need more help?

Send us a support ticket, visit the Sheetgo YouTube channel, or read the Sheetgo blog.

Related Articles
Did this answer your question?