United Kingdom. Extraction routes, and the data protection question nobody asks first

Convert PDF to CSV: which route works for the file you have?

A PDF stores marks on a page, not rows and columns, so every conversion rebuilds the table. Three identical-looking files need three methods, one needing typing. Two questions say which you hold; a third, whether uploading it is wise.

Question 1

Open the PDF and try to select a line of text. What happens?

This one test separates the three fundamentally different cases, and it takes five seconds. Everything downstream depends on the answer.

PDF to CSV: which route works for which kind of PDF, 2026

Last updated 2026-08-15

PDF has no concept of a row or column. Every conversion reconstructs a table never stored as one, and how the file was made decides the result more than the tool.

PDF to CSV: which route works for which kind of PDF, 2026
What you have	How to tell	What works	The defect to look for
Text layer, ruled table	Text selects cleanly and the cells have lines around them	Any table extraction tool. The rules are objects in the file, so the tool has real evidence about the column boundaries	Merged cells and headers spanning two columns, which usually arrive shifted by one field
Text layer, whitespace-aligned table	Text selects cleanly and the columns are held apart by spacing alone	The same tools, with column positions checked by hand before the full run	A wide value closing the gap between two columns, merging them for that row only. Look for rows with fewer fields than the rest
Text layer, awkward	Text selects but grabs whole blocks or jumps around the page	Extraction with the reading order corrected, or a rule set written for the layout	Reading order. The output is complete and in the wrong sequence, which is easy to miss and hard to detect automatically
Scanned image	Nothing selects. It behaves like a photograph	Optical character recognition first, then extraction, then verification by eye	Character confusions in numeric columns: 0 and O, 1 and l, 5 and S, 8 and B, and decimal points that move or disappear
Fillable form	There are boxes you can click into and type in	Export the field data. It is already structured as name and value pairs, so nothing has to be inferred from position	Checkbox and radio values, whose stored value is frequently not what the form displays
Statement or invoice layout	It is not a grid: records wrap, totals sit between rows, amounts change column by sign	A rule set written for that specific layout, or getting the data from its source system instead	Plausible but unevenly wrong output. Subtotals arriving as transactions is the classic case
Any of the above, containing personal data	Names, addresses, account numbers, anything about an identifiable person	The same method, performed on a machine you control, or by a service your organisation has a processing contract with	Uploading it to a converter found in a search result. That service becomes a processor, and Article 28 requires a contract before the processing starts

PDF does not store tables: it stores marks on a page, so every PDF to CSV conversion reconstructs a structure that was never recorded.
One test separates the three fundamentally different cases: if text selects, the characters are in the file; if nothing selects, the page is an image.
gov.uk warns that a scanned document converted to PDF cannot be searched by users and will not be read by a screen reader, which is the same property that stops a converter reading it.
Whitespace-aligned tables fail row by row rather than wholesale: a value wide enough to close the gap merges two columns for that row only.
Optical character recognition confuses a predictable set of characters, and 0 with O, 1 with l, 5 with S and 8 with B are all invisible errors inside a numeric column.
A fillable PDF already stores structured data as named field and value pairs, so exporting it is easier and more reliable than extracting text from any document.
Uploading a file containing personal data to an online converter makes that service a processor, and UK GDPR Article 28 requires the processing to be governed by a contract.

How this was compiled, and how to cite it

The categories in this table are the structural cases a PDF can be in, distinguished by a test anybody can run in five seconds: whether text selects. The accessibility statements are quoted from gov.uk's guidance on publishing accessible documents, read on 15 August 2026, including its warning that a scanned document converted to PDF cannot be searched and will not be read by a screen reader, which is the same property that stops a converter reading it. The data protection row is taken from the text of UK GDPR Article 28 on legislation.gov.uk, which requires processing by a processor to be governed by a contract setting out the subject matter, duration, nature and purpose of the processing, the type of personal data and the categories of data subjects. No product is named and no accuracy rate is published anywhere on this page: extraction accuracy depends on the individual file, no authority measures it, and a percentage here would be a guess given the appearance of a benchmark. What the table gives instead is the specific defect each route produces, because that is what you check the output for.

Cite this page

“PDF to CSV: which route works for which kind of PDF, 2026”, CSV Convert, https://csvconvert.uk/ (updated 2026-08-15). The categories in this table are the structural cases a PDF can be in, distinguished by a test anybody can run in five seconds: whether text selects. The accessibility statements are quoted from gov.uk's guidance on publishing accessible documents, read on 15 August 2026, including its warning that a scanned document converted to PDF cannot be searched and will not be read by a screen reader, which is the same property that stops a converter reading it. The data protection row is taken from the text of UK GDPR Article 28 on legislation.gov.uk, which requires processing by a processor to be governed by a contract setting out the subject matter, duration, nature and purpose of the processing, the type of personal data and the categories of data subjects. No product is named and no accuracy rate is published anywhere on this page: extraction accuracy depends on the individual file, no authority measures it, and a percentage here would be a guess given the appearance of a benchmark. What the table gives instead is the specific defect each route produces, because that is what you check the output for.

Scope of this checker

The 6 things it checks

Text-layer PDFs, scanned image PDFs and the awkward middle case
Ruled tables against whitespace-aligned columns, which need different tools
PDFs that are really forms, where the data can be exported directly
Statement and invoice layouts, which are not tables at all
The data protection question: what uploading a file to a web converter actually is
Route selection only. Nothing is uploaded to this site and nothing is converted here

CSV Convert is an information site. It is not affiliated with, endorsed by or connected to any software vendor, the Information Commissioner's Office or any government body, and it is not a law firm or a data protection adviser. This site does not convert, process or store your documents and never asks you to upload one: it tells you which route your file needs so you can choose a tool deliberately. Nothing here is legal advice on a data protection question. We publish no extraction accuracy figure anywhere, because accuracy depends on the individual document and no authority measures it. Every requirement quoted is taken from the source cited beside it and read on the date shown at the top of this page.

Have a lot of these, or one that will not behave?

Tell us what the documents are and how many. Data extraction suppliers who handle that kind of file will contact you directly.

What are the documents?

How many?

Do they contain personal or financial data? (optional)

When do you need this?

Next: where to reach you.

Free, no obligation
We never ask you to upload a document here
No marketing lists, ever

Frequently asked

How do I convert a PDF to CSV?

It depends how the PDF was made. Try to select a line of text. If it selects, a table extraction tool can rebuild the columns. If nothing selects, it is an image: optical character recognition must invent the characters, then be verified. Boxes you can type into mean a form, whose field data exports directly, the easiest case.

Why do PDF tables convert badly?

Because PDF does not store tables, only where characters and lines go. Every conversion infers a grid never recorded. Ruled tables convert better: the lines give the tool evidence of column boundaries. Whitespace-aligned tables are inferred from horizontal positions, until a value is wide enough to close the gap between columns.

Is it safe to upload a bank statement to an online PDF converter?

Decide deliberately, not by clicking. A statement holds personal and financial data, and uploading it makes the converter your processor. UK GDPR Article 28 requires a contract setting out the subject matter, duration, nature, purpose, data types and data subject categories. A free converter from a search result has none. Converting locally, on a machine you control, removes the question.

How accurate is OCR on a scanned table?

We publish no accuracy figure: it varies by document and no authority measures it. Predictable errors: 0 and O, 1 and l, 5 and S, 8 and B, and moving or vanishing decimal points. In numbers they are invisible. Treat output as a draft, check totals, and validate known formats such as dates and account numbers.

My PDF is a bank statement, not a table. What now?

Statement and invoice layouts have no consistent grid. Records wrap, amounts change column by sign, and running totals look like records. Output is plausible but unevenly wrong, worse than an obvious failure. Use a rule set for that layout, or get the data from its source. For UK bank data, ask for a machine-readable feed before writing a parser.

How do I check whether the conversion worked?

Six checks, about five minutes. Count rows against the source, wrapped ones included. Rows with fewer fields mean a merged column. Sort a numeric column to confirm numbers, and sum it against any source total. Check first and last rows for headers as data and off-by-one errors. Spot check five mid-document records by eye for reading-order errors.

Is there a better option than converting at all?

Frequently, and try it first. Something generated the PDF from a database, spreadsheet or report. One email asking the supplier, bank or public body for a CSV or data feed succeeds more often than expected. Extraction only reconstructs, worse, what already exists in a better form.

The detail

Sourced, dated, kept current.

Why converting a PDF table to CSV is harder than it looks
PDF stores marks on a page, not rows and columns. What that means for extraction, and why two identical-looking files behave completely differently.
Scanned PDFs: what OCR gets wrong, and how to catch it
Recognition errors in numeric columns do not look like errors. The predictable confusions, the checks that catch them, and why input quality dominates.
Before you upload a PDF to an online converter
A free converter processing your file is a processor under UK GDPR, and Article 28 requires a contract. What that means in practice, and the simpler alternative.
Checking a PDF to CSV conversion before you trust it
Every route has a signature defect. Six checks that take five minutes and catch nearly all of them, in the order worth doing them.

Sources

This site's own datasets, every figure quoted from its source and dated:

The Converter File Handling carries 14 published figures read from 11 sources' own pages; the 11 published figures given without a breakdown run from 7 to 90 day, read 11 November 2026.
The PDF to CSV Converter Index carries 36 published figures read from 30 sources' own pages; the 30 published figures given without a breakdown run from 5 to 1,500 USD (43 sources publish a figure), read 11 November 2026.

One selection test, and you know the route

What happens when you select the text, what the data looks like, and whether it is personal. Three answers.

Talk to a specialist Run the checker

Put this on your site

Embed it anywhere; attribution stays on. Copy this:

<script src="https://csvconvert.uk/embed.js" async></script>

Talk to a specialist