Convert CSV to PDF

1. 1Add your CSVDrop a .csv, .tsv, .txt or .psv file onto the page, or paste the rows straight from your clipboard.
2. 2Check how it was readThe card shows the delimiter, the quote character and the encoding that were detected, with the first rows of your table. Override any of them if a column looks wrong.
3. 3Set the page and the tableChoose the paper size, orientation and margins, then the font size, column widths and whether the header row repeats on every page.
4. 4Convert, preview and downloadConvert, read the preview, and download. Turn on PDF/A-2b first if the file is going into an archive.

Build presentation-ready PDF tables from CSV with delimiter and encoding control, optional header styling, instant preview, and quick download. Everything below happens on your own device.

• Live preview — Read the finished document in the page before you download anything.
• Real table formatting — Column widths measured from the content, a styled header band, optional row shading and a grid you can turn off.
• Delimiter options — Auto-detection that parses the file rather than counting separators, or comma, semicolon, tab and pipe by hand.
• Encoding options — UTF-8, UTF-16 LE and BE, Windows-1252 and ISO-8859-1, for legacy and regional exports.
• First row as header — Style row one as a header band and repeat it at the top of every page.
• Page control — Six paper sizes, three orientations, four margin presets and a fit-to-width mode.

The obvious way to guess a delimiter is to split the first line on each candidate and take whichever produces the most pieces. It fails on a file every export produces: a quoted address containing commas. Given a semicolon-separated row like "a,b,c,d";"e";"f", counting says comma, because line one splits into five pieces on a comma and three on a semicolon — and every row then lands in a single column.

This tool parses a sample of your file with each candidate instead, using the same quote-aware reader that does the real work, and scores the result on how consistent the row widths are. A wrong delimiter produces ragged rows or one column; the right one produces a rectangle. The quote character is scored the same way, which is what stops a file full of apostrophes from being read as if the apostrophe were a quote. When the decision is close, the card names the runner-up so you can see the choice was not obvious.

• One line is not evidence — A header of Name;Address;City and a body whose first field contains a comma disagree, and the header wins by construction if only line one is read.
• Quotes are respected — A delimiter inside a quoted field is invisible to the scorer, because the scorer parses rather than counts.
• A single column is never a win — A one-column reading is perfectly consistent, so it scores zero rather than competing — otherwise the worst candidate would always win.

A printed table reads correctly when its numbers line up on the decimal point and its dates share a format. That needs the tool to know which columns are which, and the naive way to find out — try parsing every cell as a number, then as a date — quietly destroys data. The permissive date parser most converters lean on will pull a date out of almost anything: A-1 becomes 1 January 2001, 10-20 becomes 20 October, and a product code becomes a timestamp on a page that looks entirely correct.

So the grammar here is a closed list of anchored patterns, never a permissive parser, and three guards keep a value as text whatever else it looks like: a leading zero before another digit, because 007 and 00123 are identifiers and the zeros carry meaning; a leading plus, because +44 is a dialling prefix and not a positive number; and more than fifteen significant digits, because a double cannot hold a twenty-digit account number and the round trip is not reversible. If a column is still read wrongly, turning detection off renders every cell exactly as written.

• Numbers and dates right-align — The same rule a spreadsheet uses, and the difference between a printed table and a text dump.
• Text stays left — Product codes, names and cities keep the position a reader expects to scan.
• Date order is yours to set — 01/02/2026 is January in one convention and February in another. Auto reads the column for evidence; the dropdown overrides it.
• Decimal separators too — 1.234,56 and 1,234.56 are the same amount written two ways. Unambiguous values in the column decide which convention it uses.

A table with more columns than the paper holds has two honest answers and one dishonest one. The dishonest one is to squeeze the columns until they fit, which is what most converters do: the page looks complete and the text is unreadable. Fitting sixty columns onto one A4 landscape page needs a scale of about 32%, which sets nine-point text at under three points.

Fit to width here has a floor. It shrinks the table to fit the page, but never past an effective six points, and when the floor is reached the remaining columns continue on further pages instead — the same way a wide spreadsheet prints. The card says so when it happens, because a table arriving on twice as many pages as expected should be explained rather than discovered. Long tables break the same way vertically, with the header band repeated at the top of every page so a reader always knows what column they are in.

• Nothing is drawn past the margin — Every cell is clipped to its own box, so text cannot overprint the column to its right or run off the paper.
• Clipping is counted — When a value is wider than its column the card reports how many cells were affected, and the wrap control fixes it outright.
• Orientation can decide itself — More than six columns turns the paper to landscape unless you have chosen one.
• Rows taller than a page are split — A wrapped cell that cannot fit any page continues on the next, rather than being drawn off the bottom.

A converter that only embeds a standard Latin font cannot draw anything outside Western European text, and the way that failure presents is the worst kind: not an error, but a page of control characters and wrong glyphs where your data used to be. Any name, city or product description outside Latin-1 is simply destroyed.

This converter collects every character your finished table actually needs — including symbols the formatting introduced, which are not in the source file — and embeds a subset of the faces required to draw them. Latin with accents, Cyrillic, Greek, Japanese, Chinese and Korean all render, and the text stays selectable and searchable in the PDF rather than being drawn as pictures. Right-to-left and Indic scripts are refused by name instead: the glyphs could be chosen correctly and would be positioned wrongly, and a page that looks like text and is not is worse than a clear refusal.

• Text stays copyable — One drawing operation per run of text in one face, so a PDF reader can join the words back into a sentence.
• Only what you use is embedded — A subset of each face, not the whole font, so a page of Japanese costs kilobytes rather than megabytes.
• Refusals name the script — Arabic, Hebrew, Devanagari and Thai are declined with the script named, not drawn incorrectly.

PDF/A is the version of PDF that archives accept: it forbids anything whose meaning depends on software you may not have in twenty years. Every font must be embedded rather than referenced, the colour space must be defined by a profile carried inside the file, the document must declare its own conformance level in metadata, and it must not be encrypted.

Turning it on here produces PDF/A-2b, and the claim is checked before it is made. After the pages are written, every font resource in the finished document is inspected for an embedded program; if any is missing you get an ordinary PDF and a note saying why, rather than a file that says it is archival and fails validation on ingest. A file that claims conformance it does not have is worse than one that claims nothing.

• Fonts embedded, not referenced — Including for plain ASCII, which is the case a conformance mode usually misses.
• An sRGB output intent — The colour profile travels inside the file, so the colours are defined rather than assumed.
• Metadata that agrees with itself — The title and dates in the metadata packet match the document information, which is itself a conformance requirement.

The rail is grouped by what each control changes: how the file is read, how the table looks, and what the pages are.

• Delimiter and quote — Auto, comma, semicolon, tab or pipe; quotes as ", ' or none. Changing either re-reads the file, because both decide where a field ends.
• Encoding — Auto, UTF-8, UTF-16 LE or BE, Windows-1252 or ISO-8859-1.
• Skip rows — Drop preamble lines — an export title, a date stamp — before the header row.
• First row as header — Apply header styling to row one and repeat it on every page.
• Detect types — Right-align numbers and dates, and keep identifiers as text. Turn it off to render every cell exactly as written.
• Font size and family — 5 to 16 points, in a sans, serif or monospaced face. Row heights follow the size.
• Column widths — Measured from the content, or equal for a uniform look.
• Scaling — None, or fit to width — which never shrinks text below 6 points.
• PDF/A-2b — Archival output with fonts embedded, a colour profile and the conformance metadata an archive checks.

Leave it on Auto first — detection parses the file rather than counting characters — then set it by hand if a column still looks wrong.

• Comma — The format's default and the most common delimiter.
• Semicolon — What Excel writes in locales where the comma is the decimal separator — most of continental Europe and Latin America.
• Tab — For TSV files and anything copied out of a spreadsheet or a database client.
• Pipe — Common in data pipelines and log exports, where the data itself contains commas.

Four stages, and each one is separate so that changing a setting does not re-read your file.

• Decode — The bytes are decoded, honouring a byte-order mark over the dropdown, and falling back to Windows-1252 when the file is not valid UTF-8.
• Parse — A quote-aware scan reads the rows: a delimiter inside quotes is data, a doubled quote is one quote, and a field may span lines.
• Type and lay out — Each column is sampled to decide whether it holds numbers, dates or text, then the rows are measured against the real font and flowed onto pages.
• Write — Pages become a PDF whose text stays selectable and searchable, with the fonts your data needs embedded.

Your file is read, laid out and written to PDF on your own device. Nothing is uploaded, and large files are handled off the main thread so the page stays responsive.

PDF is easy to share, print and archive, and its layout does not move between devices.

• Share and print — PDF keeps the layout identical on every device and every printer, which a spreadsheet does not.
• Archive-ready — PDF is the format long-term storage expects, and PDF/A-2b is the level archives validate against.
• Read-only by default — A table sent as PDF is a record of what you sent; a spreadsheet is an invitation to edit it.

Every layout control is free. The paper size, the orientation, the margins, the scaling, the column widths, the repeating header, the grid, the row shading, the font size, the page numbers and PDF/A are all available on the free plan at any size up to the cap — there is no quality axis on a table, and gating one would be gating whether the output is correct. What Pro buys is scale and delivery.

• Free — One file at a time, up to 25 MB, up to 300,000 cells. Every setting on this page, and no watermark.
• Pro — Up to 20 files queued, no imposed cell limit, and as large as your device can handle.
• Pro batch delivery — Download every converted file as one ZIP, or combine them into a single PDF — one document, one file per source, in order.
• Refusals are actionable — A file over a limit is refused before any work starts, with the actual number, the cap, and a link to the tool that fixes it.

• CSV to Excel
• CSV to JSON
• CSV Viewer
• Split CSV
• Add or remove columns
• Excel to PDF