Why does a ChatGPT table paste into Excel as one column?

Excel splits pasted text into columns on tab characters. A drag-selected ChatGPT table copies as plain text with pipe symbols or spaces between cells, not tabs, so everything lands in column A. You need the table converted to tab-separated (TSV) or CSV format first.

How do I convert a ChatGPT table to CSV?

Copy the response with the copy button under it, paste into ChatToWord's Excel converter, and export. It finds every table in the response, copies them in spreadsheet-ready format to your clipboard, and downloads a .csv backup that opens in Excel, Google Sheets, or Numbers.

Can I extract multiple tables from one ChatGPT answer?

Yes. A converter that parses the Markdown finds every table in the response and exports them together, stacked with their titles. That beats copying tables one by one, especially in long analytical answers.

How do I make ChatGPT give me data in table format?

Ask for it explicitly: 'Present the results as a Markdown table with columns X, Y, Z.' Also tell it to keep cells short and put units in the header row; both make the table far more usable in a spreadsheet afterward.

How to Get Tables From ChatGPT Into Excel or Google Sheets

Why pasting a ChatGPT table into Excel turns into one mangled column, and three ways to fix it: proper copy-and-convert, CSV export, and prompting for spreadsheet-friendly output.

5 min read · Updated August 26, 2026

You asked ChatGPT for a comparison, it produced a beautiful table, you copied it into Excel, and got one column of mush with pipe characters sprinkled through it. The fix takes ten seconds once you understand what's happening.

Why the paste breaks

Excel and Google Sheets split pasted text into columns wherever they see a tab character. A ChatGPT table isn't tab-separated. Under the hood it's a Markdown table, rows of text divided by pipe characters:

| Product | Price | Rating |
| --- | --- | --- |
| Alpha | $29 | 4.6 |

Paste that and the spreadsheet sees no tabs, so each row lands intact in column A, pipes and all. The fix is to convert the pipes into a format spreadsheets understand (TSV for the clipboard, or a CSV file), and nothing needs to be retyped.

Fix 1: Copy, convert, paste (free, 10 seconds)

Click the copy button under the ChatGPT response. The whole response is fine; you don't need to select just the table.
Paste into the ChatGPT to Excel converter and hit export. It parses out every table in the response, puts a tab-separated version on your clipboard, and downloads a .csv backup.
Click a cell in Excel or Sheets and paste. Real rows, real columns, numbers you can actually sum.

Analytical answers often contain three or four tables, and this pulls all of them in one pass instead of four fiddly copy jobs.

Fix 2: Export CSV straight from the chat

If you do this often, the ChatToWord extension adds an export button under every ChatGPT, Claude, and Gemini response; the dropdown next to it has Tables → CSV, which pulls every table from that reply into a .csv download without leaving the page. Same engine, zero tab-switching.

Fix 3: Prompt for spreadsheet-shaped tables

Conversion fixes the mechanics, but a table designed for reading isn't always designed for analysis. Two prompt lines fix that at the source:

Present results as a Markdown table.
Keep cells to single values (no sentences in cells),
put units in the column headers, one metric per column.

The units-in-headers line pays off the most: Price ($) as a header with 29 in the cells gives you numbers Excel can sum. $29 in every cell gives you text strings it can't.

What not to bother with

Excel's Text to Columns on pipe characters: works on one simple table if you also delete the | --- | divider row and trim whitespace everywhere. More steps than the converter, breaks on anything complex.
Asking ChatGPT to "output CSV": decent in a pinch, but commas inside values (like 1,299 or company names) silently shift your columns, and you lose the readable table in the chat. Converting the Markdown table is more reliable than asking the model to serialize data twice.
Retyping: more common than you'd think, and never worth it.

Tables are one case of a general rule: AI answers are Markdown, and everything gets easier once you copy the source instead of the rendering. The full picture is in the Markdown trick.