What counts as a markdown table?

Any block of pipe-delimited rows with a separator line of dashes under the header (for example `| a | b |` then `|---|---|`). The tool scans for these blocks anywhere in the text and ignores everything else.

Does it handle tables with misaligned pipes?

Yes. Leading and trailing pipes are optional, and extra spaces around cell content are trimmed. As long as each row uses pipes as the column delimiter, alignment does not matter.

How does JSON output map columns?

The first row is treated as the header. Each following row becomes an object whose keys are the header cells and whose values are the matching cell values, so one table becomes an array of objects.

What happens to cells containing commas in CSV mode?

Cells that contain a comma, quote, or newline are automatically wrapped in double quotes and internal quotes are doubled, following standard RFC 4180 CSV escaping so the file opens cleanly in spreadsheets.

Can it extract more than one table at once?

Yes. If the response contains several tables — common when an LLM compares options — each is detected, numbered, and converted independently.

Markdown Table Extractor

Extract tables from AI output

Large language models love to answer in markdown tables — comparisons, pricing grids, feature matrices. But you usually need that data somewhere else: a spreadsheet, a database seed, a config file. Copy-pasting a markdown table into Excel gives you one mangled cell. This tool finds every table in the response and converts it to CSV, TSV, or JSON you can actually use.

How it works

The extractor scans the text line by line looking for table blocks: a header row, a dash separator row, and one or more data rows, all delimited by |. Edge pipes are optional and whitespace is trimmed, so even sloppy LLM formatting parses:

| Model | Context |        ->   Model,Context
|-------|---------|             GPT-4o,128k
| GPT-4o | 128k |               Claude,200k
| Claude | 200k |

Each detected table is parsed into rows of cells, then serialised into your chosen format. CSV escaping follows RFC 4180 so commas and quotes inside cells survive.

Tips and notes

If a table is not detected, check that the separator line (|---|---|) is present — that dashed row is what distinguishes a real table from accidental pipe characters. For JSON, keep your header row meaningful since it becomes the object keys. When an LLM emits several comparison tables in one answer, each is extracted separately so you can take only the one you need without hand-editing.