Skip to main content

Extract

Extract key information like names, email addresses, and product attributes from unstructured text directly in Google Sheets. For example, you can extract name, weight, and price from bike model descriptions to create a catalog.

Run the bulk tool

Prerequisites
  1. Select Extensions > GPT for Sheets and Docs > Open.

    Open GPT for Sheets
  2. In the sidebar, select Bulk tools, and click Extract.

    Select Extract
    info

    If you open the Bulk tools for the first time, you are prompted to grant additional permissions. These permissions are required for GPT for Sheets to write in your spreadsheets.

    1. Click Sign-in with Google. A Sign in with Google window opens.
    2. Select the Google account with which you have installed GPT for Sheets.
    3. Click Allow to grant GPT for Sheets the required permissions.
  3. Set up your bulk tool run.

    Setup Extract
    FieldDescriptionExample
    1Column name row
    (optional)
    If your column names aren't in the first row, select the number of the row that contains the column names. The bulk tool will run on the rows below this one.2
    2ExtractEnter the entities you want to extract.
    Name
    Weight
    Price
    3From each cell in columnSelect the column that contains the text from which you want to extract entities.A: Description
    4Put results in columnSelect the column or columns to put the results in. Cells in these columns won't be overwritten with the results if they contain text.B, C, D
    5Extraction instructions
    (optional)
    Enter specific instructions for the extraction.
    Name: Extract only the model name, without manufacturer or year
    info

    The setup for the bulk tool run is stored in the History and can be reused for other runs.

    Bulk tool History
  4. Run the Extract bulk tool starting from the first empty cell in the results columns:

    1. Select a specific number of rows to run or select All rows.
    2. Click Run rows.

    Run Extract tool

You have set up and run the Extract bulk tool. If needed, try improving the results.

Extraction results

Improve results

Try different models

Find out which models work best for different use cases in our AI models overview. You can try different models by selecting them in the model switcher.

info

The model selection is a spreadsheet setting. It applies to all bulk tool runs and all GPT function executions in the current spreadsheet.

Provide extraction instructions

If your results lack precision, include incorrect extractions or inconsistent formatting, or omit information, you can improve them by providing context and specific rules for the AI to follow during extraction.

You can define the following types of instructions for the AI:

  • Extraction instructions are defined for the current bulk tool run.
  • Global instructions can also be defined in the spreadsheet settings and apply to all bulk tools and GPT functions in the current spreadsheet.
If both extraction and global instructions are defined, the AI uses both as a combined set of instructions. For consistent extraction results, make sure both sets of instructions are aligned and do not contradict each other.
For example, omit the global instructions and only define extraction instructions:
Set instructions

The following table provides examples of extraction instructions for different goals:

GoalDescriptionExample
Remove irrelevant informationDefine what should be excluded from the extraction to get cleaner results.
Name: Extract only the model name, without manufacturer or year
Normalize entitiesSet a consistent output for variations of the same entity in your text.
Weight: Extract weight in kilograms
Extract numerical values and unitsSeparate numbers from units to simplify their manipulation.Add Price value and Price currency as entities in the Extract field, then instruct:
Price value: Extract only the numerical value
Price currency: Extract only the currency code (for example USD, or EUR)

Perform open-ended extraction

Open-ended extraction helps you gather a wide range of information without specifying each entity. This approach works well when entities vary from one cell to another or when you want to avoid missing any extraction. For example, enter the following in the Extract field:

All relevant product details as a comma-separated list.
Open-ended extraction

What's next

Try another bulk tool.