What it does
AI Pre-Import Transform lets you write a plain-English instruction that automatically restructures uploaded files before they reach the column-mapping step. Instead of asking your users to clean up their files manually, the AI rewrites the data for them: renaming columns, flattening nested headers, reordering rows, splitting or merging columns, or whatever else the instruction describes.How to enable it
1
Open the sheet settings
Open your sheet’s Settings.
2
Go to the File Upload tab
Click the File Upload tab.
3
Enable the transform
Find AI Pre-Import Transform and set it to Enabled.
4
Enter your instruction
Enter your instruction in the Prompt field.
5
Save the settings
Save your settings.
Writing a prompt
The prompt is a plain-English description of what you want done to the uploaded file. Examples:- “Rename column ‘Item Description’ to ‘Product Name’ and ‘Unit Cost’ to ‘Price’”
- “Flatten the two-row merged header into a single header row”
- “Split the ‘Full Name’ column into ‘First Name’ and ‘Last Name’”
- “Remove any rows where the Quantity column is empty”
- “Add a column called ‘Status’ with the value ‘Pending’ for every row”
What happens during upload
When a user uploads a file, the AI reads the data and applies your instruction. The transformed result is what the user sees in the column-mapping step. The original file is not stored; only the transformed output is processed. If the AI can’t apply the transformation (for example, the file structure doesn’t match the instruction), the original data is shown unchanged.Credits
AI Pre-Import Transform uses a separate credit pool from OCR credits.
Monthly plan credits are used first each billing cycle. Purchased credits are used as overflow once the monthly quota is exhausted.
You receive an email notification when you reach 90% and 100% of your credit limit.
Limitations
- The prompt applies to every upload on that sheet. It can’t vary per user or per file.
- Very large files may use more credits due to a higher token count.
- Complex transformations (for example, multi-condition logic) may fall back to a direct AI call, which uses more tokens than a simple recipe.

