HTML Table to CSV / TSV / JSON / Markdown Converter

Real-world use cases

• Pull a Wikipedia comparison table into a spreadsheet in 5 seconds

  You're writing a research note and want the "List of countries by GDP" table from Wikipedia in a sortable spreadsheet. Wikipedia tables famously use `rowspan` for repeated values (the same continent spans 5 rows of countries) and `colspan` for grouped headers. Right-click the table in your browser, "Inspect", copy the outer `<table>` element, paste it here. The grid comes out correctly: each country gets its continent on the right row instead of falling off-by-one. Click "Download" — open the CSV in Excel or Google Sheets and start sorting. No browser extension, no scraping script, no Wikipedia API key.

• Extract a GitHub Insights table for a weekly report

  GitHub's Insights pages (Pulse, Contributors, Traffic) all render their data as HTML tables with merged-header `<thead>` rows. Open Pulse for your repo, save the page or copy the table element, paste here. Switch output to JSON with header mode "Auto" — you get an array of `{"week / stars": 412, "week / forks": 95, ...}` objects ready to feed into your dashboard generator. The double- header merge (top header has `colspan`, bottom header has the actual column names) is one of the most common reasons ad-hoc scraping scripts produce garbage; this tool gets it right because the grid model handles `colspan` correctly.

• Scrape a government open-data page when there's no CSV download link

  A lot of `.gov` and `.gov.cn` sites publish tables of budgets, permits, or rosters as HTML with no "download CSV" button. Open the page, copy the `<table>` (or save the whole page and paste the source), and you get a CSV with commas inside address fields properly quoted per RFC 4180. Check "BOM (Excel UTF-8)" before downloading if the data has Chinese characters — Excel needs the BOM to detect UTF-8, otherwise you get mojibake. The whole operation runs locally; you never have to upload a screenshot of the table to an OCR-then-CSV web service.

• Convert a forum-thread ranking table to Markdown for a wiki page

  A community forum has a "Top 50 contributors this month" table you want to embed in your team's wiki. The wiki uses GitHub-flavoured Markdown. Paste the HTML, switch output format to Markdown — the result is a valid GFM table you can paste directly into the wiki. Pipe characters inside cells (user signatures often contain them) are escaped to `\|`, multi-line cells get collapsed to a space so the Markdown stays parseable as a single-row-per-line table.

• Diff two versions of a "specs" table on a vendor site

  A vendor updates their published spec sheet and the only way to see what changed is to compare two HTML versions of the same table. Extract both as CSV with identical settings, then `diff` them — every row alignment will match even if the vendor added or removed `rowspan` cells in the new version, because the grid model normalises both to a rectangular layout. This is roughly impossible to do with a raw HTML diff (whitespace and attribute reordering drown out the actual changes).

• Bulk-export 20 nested tables from a saved expense report

  Some legacy CRMs export expense reports as one giant HTML page with a top-level table where each row contains a nested `<table>` of line items (receipts, taxes, mileage). Paste the whole page in. The tool returns 21 tables: the summary on top, then each nested line-item table as its own entry. The summary cells show `[table #1]`, `[table #2]` etc. so you can map line-item details back to which expense row they belonged to. Click "Download ×21" to get all of them as one concatenated CSV with row separators.

Common pitfalls

• Pasting only the `<tr>` rows without the wrapping `<table>` tag. DOMParser silently drops unwrapped `<tr>` because they are invalid outside a table context. Always include the `<table>...</table>` wrapper, even if you have to hand-add it.

• Forgetting to turn on "BOM (Excel UTF-8)" when the data has Chinese, Japanese, Korean, or emoji characters. Without BOM, Excel on a non-Western locale opens the CSV as the local code page and you see mojibake. The BOM checkbox fixes it without changing the data.

• Choosing "First row" header mode when the first row is actually data. JSON output then uses raw values as object keys (e.g. one key becomes `1`, another becomes `2.50`) which is technically valid but useless. Use "Auto" unless you specifically know the source has no `<thead>` and you have verified the first row is a header.

• Expecting nested tables to be flattened into one big table. They are not — each nested table becomes its own entry in the result list, and the parent cell shows `[table #N]`. Use the tab strip above the output to navigate between extracted tables.

• Trying to round-trip the resulting CSV back to HTML through the CSV-to-HTML direction and assuming `colspan`/`rowspan` will reappear. They will not: CSV has no concept of merged cells, so the round-trip loses the merge metadata. Keep the original HTML if you need that information later.

Privacy