Extract data from a messy spreadsheet into clean, separate columns
A column of addresses, product titles or notes typed by hand rarely splits cleanly with a formula. Upload the file, name the fields you want, such as City and Country, and get them as new columns for every row. When a cell does not contain a field, that cell stays empty instead of being filled with a guess. The first rows are free; the whole file is $5.
- Name up to six fields, separated by commas: City, Country, Postcode
- Each field becomes its own column after your last column
- A field the row does not contain is left empty, not invented
- Rows the model could not answer are listed in a warnings CSV
- Up to 20,000 rows for one payment of $5
Why formulas and splitting give up
Text to Columns and formulas need a pattern: a comma before the city, the size always last. Hand-typed data has no pattern. "Flat 2, 14 Rue Oberkampf 75011 Paris", "Paris (FR)" and "75011 PARIS FRANCE" all hold the same city in different places. A model reads each row as a person would, and the tool turns its answers into columns.
The instruction tells the model to use only what is in the row. It will not look up a missing country from a postcode database or browse the web, so a row that only says "Springfield" gets a city and an empty country, which is the honest answer.
Typical extractions
- City, region and country from one free-text address column
- Size, color and material from product titles such as "Linen shirt, sage, XL"
- Brand and model from marketplace listing titles
- Order number and requested action from customer emails pasted into a sheet
- Company name and job title from a "signature" or "bio" field
Getting clean values
Say the format you want in the instruction: "country as a full English name", "size as S, M, L or XL". If a field has a short fixed set of answers, switch to the Categorize task and give the set as allowed values; every answer is then checked against it.
How the tool works, what it costs and how your files are handled is on the AI for Every Row page; prices for every tool are on the pricing page.
Last updated
Questions and answers
Can ChatGPT extract the city from 8,000 addresses?
It extracts well from the rows it sees, but 8,000 rows do not come back in one reply, and it tends to stop or skip. This tool sends the rows in blocks of 40 and requires an answer for every row number.
What happens when a row does not contain the field?
The cell stays empty. The model is told not to invent details that are not in the row.
Can I extract several fields at once?
Yes, up to six in one job. Type the column names separated by commas.
Does it read every column?
By default it reads every column except IDs, links and images. Name the column to read, such as Address, to keep the model focused.
What does it cost?
Files of up to 100 rows are free in full. Bigger files are $5, paid once, for up to 20,000 rows.