Excel AI

Run one prompt on every row of a spreadsheet. Reference your columns as {{A}}, {{B}} and get the answers back in a new column - for translating, summarising, categorising or enriching data at scale.

3 min read · Updated 28 September 2026

Excel AI is for the job nobody wants: reading a spreadsheet row by row and writing something for each line. Upload an Excel or CSV file, write one prompt that refers to your columns, and the AI fills in a new column for every row. It runs row by row - the AI never gets the whole file at once - so each answer is focused on its own line. No formulas, no code.

Excel AI running a column-referencing prompt on every row.

How it works

  1. Upload File - an .xlsx, .xls or .csv file.
  2. Configure - write the prompt, choose the model and the output column.
  3. Preview - try it on a row and check the answer.
  4. Process - run it on every row.

First time here? Watch demo runs an automated walk-through that loads a sample file, writes a prompt and runs it for you.

Process a spreadsheet

  1. Upload your spreadsheet

    Go to Apps and choose Excel AI. Under Upload your spreadsheet, drop a file or select browse. Excel AI supports .xlsx, .xls and .csv.

  2. Write the prompt

    Under Configure your AI prompt, write what the AI should do for each row. Refer to a column by its letter in double curly braces: {{A}}, {{B}} and so on. For each row, {{A}} is replaced by that row’s value in column A.

    Translate {{A}} to French.
  3. Choose the model and the output column

    Pick the AI Model and the Output Column - the column where the answers should go, for example F. Choose an empty column so nothing is overwritten.

  4. Preview a row

    Test the prompt on a row before you run the whole file. Read the answer as critically as you would a colleague’s. If it is too long, too vague or in the wrong format, change the prompt and preview again.

  5. Process all rows

    When the preview looks right, process the file. Every row gets its answer in the output column.

Example: one action per store

This is the run in the recording above. Northwind’s operations team exports Store Performance Q3 2026.csv, one row per store:

ABCDEFG
1storeregionnet_sales_usdordersavg_basket_usdreturns_pctnps
2Austin DowntownAustin494166.24685272.124.434
3Denver LoDoDenver……………

The prompt:

Store {{A}}: returns {{F}}%, NPS {{G}}.
One action for the store manager this week, max 12 words.

With Output Column set to H, each of the twelve stores gets one concrete action in the new column - ready to paste into the weekly ops update.

History and downloads

The History tab lists every file you have processed, with its status, when it was created and finished, and the credits it used. Select Download to get the file back with the new column filled in.

You don’t have to keep the page open while a big file runs. When processing finishes, StickyPrompts also emails you a link to download the result:

The completion email for the run above.

Tips

  • Ask for a fixed format. “One sentence”, “one of: High, Medium, Low” or “a number from 1 to 5” gives you a column you can sort and filter.
  • Keep one job per run. For a category and a summary, run the file twice into two output columns.
  • Preview a tricky row. Test on a row with missing or unusual data, not just a tidy one.
  • Tell it what to do with blanks. “If {{C}} is empty, write N/A” avoids made-up values.
  • Clean the file first. One header row and one record per row gives the most reliable results.