Why a scanned PDF is rarely ready to edit

A PDF can look perfectly tidy on screen and still be hiding a problem. Many of the documents that arrive for conversion are not true digital files at all — they are photographs of pages, saved as PDFs. There is no text layer beneath the image, so nothing can be selected, searched or copied. To make the words editable, recognition software has to look at the shapes on the page and decide what letters they represent.

That is a remarkable process, and it is right most of the time. But "most of the time" is not good enough when the finished text is going into a contract, a medical report, a legal bundle or a set of accounts. Every conversion we handle gets a careful read-through before it goes back to the client, because the faults that survive an automated pass are exactly the ones that cause embarrassment later.

The faults we look for first

Experience teaches you where a conversion tends to slip. These are the four issues we check on every job.

• Stray characters. Recognition software confuses shapes that look alike: rn becomes m, cl becomes d, and the humble 1, l and I are swapped with cheerful abandon. A reference like "Cl-1041" can quietly become "d-1041" and no spellchecker will ever flag it.
• Broken lines. Conversions often insert a hard return at the end of every visual line, so a single sentence arrives as eight short lines. Worse, words hyphenated across a line break stay split — "docu-ment" remains two fragments rather than becoming "document". Reflowing these by hand is fiddly but essential if the text is going to be pasted into anything else.
• Missing words. Faint print, a coffee ring, a staple, a highlighter streak or a stamp across the corner can all cause a word to vanish entirely. Punctuation disappears at the edges of pages. A missing "not" changes the meaning of a sentence completely.
• Merged and split words. Two columns read as one line, producing nonsense such as "the committee agreedthe proposal was sound". Spaces appear in the middle of words, and words run together where the original spacing was tight.

Tables, columns and layout traps

Prose is forgiving; tables are not. When a page is converted, the neat grid of a table often collapses into a stream of text with no indication of which figure belongs to which row. Numbers can shift one column to the left, and a total ends up sitting under the wrong heading.

Multi-column layouts cause similar trouble. A newsletter or academic paper with two columns may be read straight across, mixing the end of the left column with the start of the right. Headers and footers, meanwhile, get repeated in the middle of paragraphs, and page numbers land in places they have no business being.

The sensible approach is to rebuild the table structure rather than trust the conversion. Where the client needs the data in a spreadsheet, we will often set the table out again from scratch and check every figure against the original image, row by row.

Names, numbers and references need a second pair of eyes

Certain details deserve particular attention, because they matter more than the surrounding prose.

• Proper nouns. Surnames, place names and company names are frequently misread, especially unusual spellings or names with accents. We check them against the document itself and any correspondence supplied with the job.
• Dates and figures. A 3 read as an 8, or a 5 as a 6, is easy to miss and expensive to get wrong. Decimal points and commas in large numbers need checking too.
• Reference codes. Account numbers, case references, invoice numbers and postcodes are prime candidates for stray characters. Read them digit by digit rather than as a whole.
• Quotations. If a document quotes another source, the wording must match exactly. Even a small change can undermine the reliability of the piece.

A practical check before anything leaves the desk

A short, consistent routine catches the great majority of errors. Ours runs something like this:

• Spellcheck the whole document, then read it again — spellcheck will not catch a correctly spelled wrong word.
• Read the text aloud, or have someone else read it, to expose missing words that the eye skips over.
• Search for double spaces, stray line breaks and any symbol that looks out of place.
• Compare the first and last line of every page against the original scan, where problems often cluster.
• Check headings, bullet points and numbered lists have survived as distinct elements rather than being flattened into paragraphs.
• Confirm the formatting the client asked for: font, headings, page breaks and whether tracked changes or comments need preserving.

When retyping beats converting

Conversion is not always the fastest route to a clean result. If a document is faint, heavily annotated, handwritten, or laid out in a way that confuses the software, typing it out by hand from the images can be quicker than repairing a poor conversion. The same is true for short documents with dense figures, or anything where accuracy matters more than speed.

Knowing which method to choose — and being honest with a client about the trade-offs — is part of the service. Either way, the principle is the same: a converted document should read as though it were typed by someone who cared. That means the stray characters, broken lines and missing words have all been found and fixed before the file is sent, not after the client spots them.