Articles: PDF tables from upload to the finished file

A PDF is not a “document with structure”, it is a list of instructions: “draw this glyph at these coordinates”. Concepts like “row”, “column” or “cell” usually do not exist in the file at all — they have to be reconstructed. We do it in two ways and tell you which one worked for every table we find.

By grid lines. When a PDF contains real lines and rectangles, the structure is rebuilt from them — including merged cells, which a naive conversion usually scatters. Such a table is labelled “Found by grid lines” in its card. This is the most reliable case: invoices, acts, price lists, reports and exports from 1C, SAP or Excel are almost always drawn with borders.

By text alignment. When there are no borders, columns are detected from the empty gaps between text: we look at where fragments on the same line start and end and group them into columns. This is how “invisible” tables are found — specifications, item lists, price lists without a grid. The card then says “Found by text alignment”.

The practical conclusion: the more honestly the grid is drawn in the source, the more accurate the result. If the document does contain a table but it looks like plain text with tabs, expect the second algorithm — and always compare the result with the original (see section 4).

1. 1Upload the PDF — drag the file onto the drop zone or click it and pick the document. Files up to 200 MB are supported; from roughly 40 MB processing takes noticeably longer.
2. 2Wait for the parse. Progress is shown per page (“Page 12 of 84”). Do not close the tab: the work happens in your browser, not on a server, so interrupting it means starting over.
3. 3See what was found. At the top — a counter of the tables found; below — cards with the number, the pages and the size (“p. 3–4 · 41 × 6”).
4. 4Compare with the original. Every card has an «Original» button: the PDF fragment is shown next to the result.
5. 5Set up the export. The “Header rows” and “Drop columns on the left” steppers decide what goes into the file; a text summary sits right next to them (“Export: without the header (2 rows) and 1 column on the left dropped”).
6. 6Fix cells if needed — a double click opens the text for editing.
7. 7Download or copy. «Excel»/«Word» for a single table, “Download all to Excel”/“Download all to Word” for the whole document (each table becomes its own worksheet or its own table). “Copy” puts the table on the clipboard, ready to be pasted into Excel, Word or Google Sheets as a real formatted table.
8. 8Need another document? “Another document” resets the result; you can also simply drop a new PDF on top — the current result is replaced.

• A text layer is required. The check is simple: open the PDF and try to select a word with the mouse. If it selects, the layer is there. If nothing can be selected, it is a scan or an image — and we have no OCR.
• Run scans through OCR first. Process the document with Adobe Acrobat, ABBYY FineReader or any other OCR tool and save the result as a PDF with a text layer; our service will then handle it normally.
• Remove the password beforehand. A protected PDF will not open. The quickest workaround is “Print → Save as PDF” without a password (if the document is available for viewing).
• One upload — the whole document. If the tables are scattered across a 300-page report, it is easier to upload it in one go: there is no page limit. On a weak device, though, large files take longer and need a lot of memory.
• A table inside a picture is not a table. A screenshot of a report pasted into a PDF stays a picture — there is nothing to extract from it.
• Do not “print to PDF” one more time than needed. Every extra pass through a virtual printer can turn real lines into a set of strokes and make the grid algorithm work worse.

The export is configured in the card of each table, and that is the main reason to review the results before downloading:

• Header rows. Multi-level headers (“Indicator / Unit / Value” on two lines) are often more useful kept in the file, while service rows on top are removed with the “Header rows” stepper.
• Extra columns on the left. The classic case is the row-number column. It is removed with the “Drop columns on the left” stepper; by default nothing is trimmed.
• Mouse selection. Select a rectangular range of cells and press Ctrl+C — only that range goes to the clipboard. Esc clears the selection.
• Continued on the next page. If a table is broken by a page break, “⤒ Merge with previous” joins the fragments into one table.
• A block that is not a table. Something the service mistook for a table (columns of plain text, for example) is removed with the cross — confirmed by a second click.
• Borders and numbers. The export settings hold the “Cell borders” and “Numbers as numbers” switches; they affect Excel, Word and the clipboard alike.

The .xlsx carries merged cells, borders, fills, bold text, alignment (left / center / right), column widths and row heights — so the table looks like the original instead of “text spread over cells”.

A word about data types. With “Numbers as numbers” enabled, values like 1,234.56 are written as real numbers: they can be summed, and formulas and sorting work on them. At the same time, article numbers, tax IDs, dialling codes and other identifiers with leading zeros (0045, 07734) stay text — otherwise Excel would silently eat the zero and the column would stop being a column of codes.

Worksheet names are localised and include the source pages: “Table 2 (p. 1-2)”. The file name is too — for example report_table_1.xlsx.

A .docx is the right choice when the table has to go back into a document: an act, a report, an explanatory note, a contract annex. We keep the column proportions, and wide tables are switched to landscape orientation automatically — otherwise they would be squeezed until unreadable. The caption under the table (“Table 3 (p. 7–8)”) makes it easy to reference the source when you double-check it.

The service reports specific errors, and each of them tells you what to do:

An honest list of limitations saves more time than promises do:

• No OCR. Scans, photos and tables-as-images are not extracted. You need an external OCR tool, after which the PDF can be uploaded here.
• No direct «PDF → CSV» conversion. But “Copy” plus pasting into Excel/Google Sheets and saving as CSV gives the same result in a few seconds.
• No handwriting recognition and no tables inside charts.
• No batch queue. Documents are uploaded one at a time — in exchange there is no sign-up, no limits and no waiting on a server.
• No recovery of collapsed merged cells in documents without lines. If the PDF physically has no grid, the merges have to be set by hand in Excel.

In your browser. The PDF is never uploaded to a server: reading the file, searching for lines and text, and generating the .xlsx and .docx all happen locally, on your device. Documents therefore never leave the computer and are not stored anywhere — which matters for invoices, contracts, bank statements, reports with personal data and anything under an NDA.

The only network request the page makes is a country lookup used to suggest an interface language (Russian, English or Spanish). It has nothing to do with your file and can be switched off by opening the site with ?geo=0.

The flip side of local processing is that the load lands on your device instead of someone else's cluster: a large document may take a minute or two, and that is normal.

Conversions are unlimited: by number of files, by number of pages and by number of exports. No registration and no payment. Hosting and development are covered by voluntary donations — Ko-fi (by card or through PayPal, no account needed) and Boosty (handy when cards of foreign services are unavailable). The buttons are on the main page: if the service saved you an hour of retyping a table, that is a way to say thank you and help keep it free.

A short list of things that are useful next to the task “get the data out of a document”:

• OCR for scans — Adobe Acrobat, ABBYY FineReader: they turn an image into a PDF with a text layer, which can then be fed to a table extractor.
• Google Sheets — accepts our table from the clipboard and lets you share the result quickly or build a chart on it.
• Cloud infrastructure and spend — zop.dev: resource inventory, costs and savings recommendations in one place, if your project, like ours, lives in the cloud and you would rather not pay for idle servers.
• This site — upload a PDF and check everything on your own document: it takes under a minute and needs neither an account nor installed software.