How to copy a ChatGPT table into Excel (3 ways that keep the numbers)
By the AI Text Formatter team ·
You asked ChatGPT for a comparison, it gave you a tidy table, you copied it, you pasted into Excel — and got one column of text with | characters in it. The short version: paste the table into the free ChatGPT table to Excel converter and either copy it back out for Excel or download a .xlsx. Here is why the straight paste fails, the three routes that work, and the handful of number formats that catch people out.
Why the table pastes as one column
What looks like a table in the ChatGPT window is markdown underneath: one line of text per row, cells separated by vertical bars, and a line of dashes under the header. When you select and copy, the clipboard holds that text. Excel has no idea the bars are column dividers, so it does what it does with any text — one line per cell — and you get a column of | Widget Pro | $12,400 | 8.5% | strings.
The copy button under the answer sometimes does better, because ChatGPT now puts a rich-text version on the clipboard as well. Whether Excel uses it depends on the Excel build, and when it does, the numbers tend to arrive as text with a non-breaking space glued on, which is why the column then refuses to sum.
Way 1: Copy for Excel (fastest for a sheet you already have open)
Paste the table into the converter. It reads the pipes, drops the dashes row, strips the hidden characters and the **bold** markers from the cells, and shows the grid. Click Copy for Sheets / Excel, click a cell in Excel, paste.
What Excel receives is an HTML table — the same thing it gets when you copy a table from a web page — so it reads it straight into cells and runs its own recognition on the text. 12,400 becomes a number, 8.5% becomes a percentage, 2024-03-01 becomes a date in your regional format. The header is on the first row. This is the everyday route: no file, no import dialog, cells where your cursor was.
Way 2: Download a .xlsx (best for a file, or for several tables)
Click Download .xlsx instead and you get a workbook built in your browser from the parsed cells. The typing is done by the tool rather than by Excel's guesswork: numbers are numeric cells with a thousands-separator format, percentages are stored as fractions and formatted as percentages, currency keeps its symbol as a number format so the column still sums, the header row is bold and frozen, and the columns are sized to their content.
The reason to prefer it over Way 1 is multiple tables. A ChatGPT answer with a summary table and a breakdown gives you a workbook with two sheets, in order, named Table 1 and Table 2. It is also the route when the destination is not Excel on your machine — a colleague's inbox, a system that imports workbooks, an archive.
Way 3: CSV, or the manual repair
Download .csv gives you one table as a plain comma-separated file with a byte-order mark at the start, which is the detail that makes Excel open accented names correctly instead of as mojibake. Use it when the destination is accounting software, a database or anything else that wants CSV rather than a workbook.
The fully manual route, for completeness: paste the raw text into column A, Data → Text to Columns with | as the delimiter, delete the empty first and last columns the outer pipes create, delete the dashes row, trim the spaces, then convert the number-shaped text to numbers. It works. It takes five minutes per table, and the number conversion is where it goes wrong.
The number gotchas
Spreadsheets are only useful if the numbers add up, and ChatGPT tables are written for reading rather than computing. The converter applies a few deliberate rules; they are worth knowing because the manual route applies none of them.
Thousands separators and currency. 12,400 is read as 12400 and shown with a separator. $12,400, €950 and £1,200 are read as numbers with a currency format, so the column sums and still displays the symbol.
Percentages. 8.5% is stored as 0.085 and formatted as 8.50%; 14% is stored as 0.14 and shown as 14%. This is how Excel expects percentages, and it means a SUM or an AVERAGE over the column gives a sensible answer.
Negatives. (1,234) in accounting style and −1,234 with the Unicode minus both become -1234.
Leading zeros and long IDs. 02134 is a ZIP code, not two thousand one hundred and thirty-four, and 1234567890123456 cannot be held exactly in a spreadsheet number. Both are kept as text on purpose. If you want a column of codes treated as numbers anyway, switch off Detect numbers and format the column yourself.
Dates. The .xlsx keeps dates as text, because 1/2 is a different day in Chicago and in Madrid, and typing it wrongly is worse than leaving it. Paste instead (Way 1) and Excel applies your regional date recognition. ISO dates like 2024-03-01 are recognised everywhere.
Units in cells. 12 kg is text. Ask ChatGPT for "numbers only, units in the column header" and the column types cleanly.
European decimals. Only the 1,234.56 convention is read as a number; 1.234,56 stays text.
Several tables, and tables from other assistants
Paste a whole answer and every table in it is detected; tabs above the preview switch between them, and the .xlsx download writes all of them. Prose between the tables is ignored, so there is no need to trim.
The same converter reads tables from Claude, Gemini, Perplexity, DeepSeek and Copilot — they all write markdown pipes — and there are dedicated pages for the quirks each one brings: Claude's CSV artifacts, Perplexity's citation markers inside cells, DeepSeek's reasoning block with its draft tables and Gemini's cells with units. For Google Sheets rather than Excel, the ChatGPT to Google Sheets page covers the locale and import differences.
Everything runs in your browser: the table is parsed, the clipboard copy is made and the .xlsx is built without anything being uploaded, which matters when the table is pricing, payroll or client data.