AI for every row
Run one AI instruction on every row of a CSV or Excel file
Upload a spreadsheet, write one instruction such as "assign each product one of these categories" or "extract the city and country from the address", and get the same file back with new columns filled for every row. Up to 20,000 rows are $5 as one payment, with no subscription, and the first rows are free with no sign-up. A chatbot handles a pasted list of fifty rows well; on ten thousand rows it stops partway or starts guessing. This tool sends the rows in small blocks and checks that every row came back.
- Up to 100 rows free in full, first rows of a bigger file free, no account needed
- Whole file $5, one payment: up to 20,000 rows and 6,000,000 characters
- Ready-made tasks: categorize, extract fields, clean names, sentiment, translate and categorize
- Give a list of allowed values and every row gets exactly one of them
- Your file back with new columns at the end; every other cell unchanged
- Rows the model could not answer are left empty and listed, never invented
What you get back
You get your own file back, in the same format, with your new columns added after the last column, for example products.ai.csv. Before you pay, the first rows are shown on screen with their answers so you can check the instruction works.
- Every row in the same place and order, with the new columns filled. Cells you already had are copied from your file unchanged, and so is the header row.
- CSV: the same delimiter, quoting, line endings and encoding. Excel: the new columns go after the last used column of the sheet; formulas, styles and the other sheets stay as they were.
- With a list of allowed values (categories, yes or no, a scale), each answer is matched to the list and written exactly as you spelled it. An answer outside the list is asked again, then left empty.
- A warnings CSV with the row number and the reason for every row that was left empty, so you can fix those by hand.
A chatbot and this tool, row by row
ChatGPT, Claude and Gemini are good at the judgment itself: which category a product belongs to, which part of an address is the city. The trouble starts when the same judgment has to be made ten thousand times and written back into a file.
| Your situation | A chatbot | This tool |
|---|---|---|
| Fifty rows pasted into the chat | Usually does it well | Does it too, and gives you the file back |
| A sheet with 10,000 rows | The answer has to be as long as the file, so it tends to stop after the first part or ask you to continue | Sends the rows in blocks of 40, up to 20,000 rows per job |
| A fixed list of 20 categories | Drifts over a long list: new labels appear, spellings change | Checks every answer against your list and asks again if it does not match |
| A row with too little information | Often fills the gap with a plausible guess | Is told to leave it empty; empty rows are listed for you |
| Keeping IDs, prices and other columns | Rewrites the table, so any cell can change | Never rewrites your cells; it only adds the new columns |
| Doing it again next month | A new conversation, a slightly different result | The same instruction and list on the new file |
How to run AI on every row of a spreadsheet
-
1
Upload the file and write the instruction
Drop a .csv, .tsv or .xlsx file, pick a task such as Categorize or Extract fields, and write one instruction. Add allowed values or new column names if you want them.
-
2
Check the first rows free
The first block of rows is processed and shown next to the new columns. If an answer is off, start again with a clearer instruction. No account is needed.
-
3
Unlock and download
Happy with the first rows? Unlock the whole file for $5 and download it with the new columns filled for every row.
Why big sheets trip up general AI assistants
Classifying a row is a small task. Classifying ten thousand of them in a chat means the assistant has to write ten thousand answers in one reply, in order, without losing count. A reply has a length limit, so the assistant stops partway, asks you to type "continue", or quietly shortens the end. When it runs code over an uploaded file instead, it tends to fall back on keyword rules that miss anything unusual.
Consistency is the second problem. Over a long list the labels drift: "Home & Kitchen" becomes "Kitchen", a new category appears that was not on your list, and two spellings of the same city sit side by side. On a short list you would spot it. On a long one you find out when the pivot table looks wrong.
This tool never asks a model to write your file. It reads the rows itself, sends them in blocks of up to 40 with your instruction, and requires one answer back for every row number. Answers are checked against your list of allowed values, written in your exact spelling, and a row that is missing or off the list is asked again. If it fails twice, the cell stays empty and the row is listed. The new columns are then added to your original file.
What people use it for
- Categorize 10,000 products into a fixed list of store categories or a marketplace taxonomy
- Extract city, country, size, color or brand from messy free-text columns
- Standardize company names before a CRM import: one spelling per company, legal suffixes removed
- Label support tickets or emails by topic and urgency
- Score sentiment for thousands of comments, then filter the negative ones
- Translate a column and categorize it in the same pass, for example foreign-language reviews
What it does not do
It reads only what is in each row. It does not browse the web or look up company registries, so "find the website of each company" will produce empty cells, not research. Very long cells are shortened to their first 1,000 characters for the model, and each row to 3,000; your file keeps the full text.
One file per job, up to 100 MB; in an Excel file the first sheet with a header row is used. One payment covers up to 20,000 rows and 6,000,000 characters of input; Unlimited covers up to 100,000 rows. Rows beyond the limit are left empty, and the screen tells you how many before you pay.
It is an AI model making a judgment on each row. Check a sample before you rely on the result, especially for anything with legal, medical or financial weight.
Getting a good result
Write the instruction for one row, as you would for a new colleague: what to look at and what to write. Give allowed values whenever the answer should come from a fixed list; it is the single biggest gain in consistency. Leave out columns the model does not need, such as internal codes, by listing the columns to read.
Use the free rows as a test. If an answer is wrong, the instruction is usually ambiguous; add a short rule ("a mug is Kitchen, not Dining") and start again. Rows the model could not answer stay empty rather than guessed, so a quick filter on the empty cells shows what needs a human look.
By language and situation
Categorize 10,000 rows of a CSV with AI
Upload the CSV, paste your category list, and every row comes back with one category from that list in a new c...
Read more →Classify rows in Excel with AI
Upload the .xlsx workbook, describe the classes, and get the same workbook back with a new column after the la...
Read more →Extract data from a messy spreadsheet
A column of addresses, product titles or notes typed by hand rarely splits cleanly with a formula. Upload the...
Read more →Standardize company names in Excel or CSV
A CRM export with "ACME, INC.", "Acme Inc" and "acme" for the same customer breaks every count and every merge...
Read more →Run a ChatGPT-style prompt on every row of a spreadsheet
Write the prompt you would type for one row, name the columns you want back, and upload the file. The prompt r...
Read more →What it costs
Small files, first rows
Free
A file of up to 100 rows is free in full, downloads included. Bigger files: the first rows on screen, no account. 1 job a day, or 3 a day with a free account.
Whole file
$5
One payment for one file: up to 20,000 rows and 6,000,000 characters. No subscription, never expires.
Unlimited
$19 / week
Every tool on the site, with files of up to 100,000 rows.
Questions and answers
Can ChatGPT categorize every row of a big CSV?
It can categorize a few dozen pasted rows well. A file with thousands of rows has to come back in one reply, so a chatbot tends to stop partway, and nothing checks that every row got an answer from your list. This tool sends the rows in blocks of 40, checks every answer and adds the results to your file.
How many rows can I process?
One payment of $5 covers up to 20,000 rows and 6,000,000 characters of input. Unlimited covers up to 100,000 rows per file. Empty rows do not count.
Can I make it choose only from my own categories?
Yes. Put the allowed values in the box, one per line. Every answer is matched to the list and written in your spelling; an answer that is not on the list is asked again and then left empty and listed.
Can I add several columns at once?
Yes, up to six. For example City, Country and Postcode from one address column, or a translation and a category in one pass.
Which files can I upload?
CSV and TSV files with a comma, semicolon, tab or pipe delimiter, and Excel .xlsx files, up to 100 MB. In Excel, the first sheet with a header row is processed and the new columns are added after its last column.
Will my other columns change?
No. Your cells are copied from your file as they are; only the new columns are added at the end of each row.
What happens when the model is unsure?
It is told to leave the cell empty rather than guess. Empty answers, missing rows and answers outside your list are listed in a warnings CSV with the row number.
Is it free?
A file of up to 100 rows is free in full. For bigger files the first rows are processed free and shown on screen, with no account. The whole file is $5 paid once, or Unlimited at $19 a week covers every tool on the site.
Are my files private?
Only you can open your job. The cells of each row are sent to an AI model provider for the sole purpose of answering your instruction, through routes that do not use your data for training. Jobs uploaded without an account are deleted after 7 days, and you can delete any job yourself at any time.
Updated October 25, 2026 · APPFACT STUDIO LTD