Converting a PDF manual into searchable web docs

This matters more than any conversion tip, so it goes first. A PDF is a description of where marks go on a page. It does not, in general, record that a line is a heading, that a block of text is a list, or that two columns are meant to be read one after the other. All of that was known by whatever produced the PDF, and thrown away at export.

So a converter is doing inference. It looks at font sizes, positions and spacing and guesses what was structural. Good converters guess well on well-made PDFs and badly on everything else, and no converter can recover information that is not in the file.

This is why converting a .docx is a conversion and converting a PDF is closer to a reconstruction. If you can find the original Word or InDesign source, use that instead and stop reading here — you will get a better result in less time.

Worth checking before you spend an afternoon, because the answer is visible in about ten seconds. Open the PDF and try to select a paragraph of text with your cursor.

The scan case is worth being blunt about: if you cannot select the text, there is no text in the file, only pictures. You need OCR before any of this is relevant, and OCR output needs proofreading regardless of what you do with it afterwards.

1. Try selecting text, and find the original

Two checks, both quick. If text does not select, stop and run OCR. If the original .docx or source file exists somewhere, use it — converting a Word document loses far less.

2. Upload the PDF

Drop it into a project, up to 200 MB. It is converted at import on the same rule as a Word file: two or more top-level headings become a folder of pages, fewer become one page, and images are extracted and land beside the page with their links rewritten.

3. Expect one long page, and plan for it

This is the most common outcome and it is not a failure. If the manual’s headings were styled by size rather than marked as headings, the converter has nothing to split on. You get one page with all the content in the right order, which is a perfectly good starting point — you just do the splitting yourself.

4. Split by hand, top down

Work from the manual’s own table of contents. Create a page per chapter, cut the section across, and move on. It is dull and it goes faster than you expect, because you are moving text rather than writing it. Drag the pages into order as you go; the tree you are looking at is the navigation the site will use.

5. Publish, then fix in place

Do not wait for it to be perfect. A searchable site with rough edges beats a PDF nobody opens, and once it is published you can fix pages one at a time — every save is a version you can restore, and the published site only changes when you press Publish.

In rough order of how often it comes up.

• Hyphenated line breaks. Justified text often breaks words across lines. Search for a hyphen followed by a space and repair the joins.
• Headers and footers as body text. The company name and page number repeated every page, now scattered mid-paragraph. Search for the repeated string and delete the lot.
• Page numbers in cross-references. Same as any print document: "see page 41" has to become a real link.
• Tables flattened into text. Complex tables frequently arrive as runs of words. Rebuild them as Markdown tables, or as lists if they were never really tables.
• Captions detached from images. The image lands, the caption ends up above or below the wrong thing. Quick to spot, quick to fix.
• Ligatures and smart quotes. Occasional odd characters where the font did something clever. Worth a find-and-replace pass.

None of this is exotic and none of it requires a tool. It is an afternoon with a document you already know, and you end up with content that is finally editable rather than frozen.

Worth naming, because in the middle of fixing hyphenation it stops being obvious.

• Full-text search, built at publish time and served with the site — the thing a PDF cannot do until it has been downloaded and opened.
• A URL per section, so support can link to the exact instruction instead of naming a page number.
• One copy. You fix a mistake once, and every reader has the fix.
• It works on a phone, at a width the reader chose rather than at A4.
• It is editable Markdown now, so the next update is a paragraph rather than a new export cycle.
• You can still hand out a PDF. Nothing stops you exporting one; it just stops being the master copy.

A useful sequencing trick: publish the single long page on day one. It is searchable and linkable immediately, which is most of the value, and the splitting can happen over the following week without anyone waiting on it.