PDFs are where most companies' training material actually lives — the employee handbook, the security policy, the compliance pack from legal. They're also the least cooperative starting point, because PDF describes how a page looks rather than what it means.
Use the source document if you have it
Worth saying first because it saves the most trouble: if the PDF was exported from Google Docs, Word, or Confluence, convert from that instead.
The original has real structure — headings are headings, lists are lists, tables are tables. The PDF has coordinates and font sizes that a converter has to infer structure from. Starting from the source skips an entire category of problem.
People often reach for the PDF out of habit because it's the version that got circulated. The original is usually still there.
Which PDFs convert cleanly
Text-based, single-column PDFs convert well. Most policy documents, handbooks, and reports exported from a word processor fall into this category, and the result is usually accurate enough to use with light review.
Scanned PDFs are images. A scan of a printed document contains no text at all — just a picture of text. Nothing can extract from it without OCR first. If you can't select text in the document with your cursor, it's a scan.
Multi-column layouts are the most common source of scrambled output. Extraction may read straight across both columns, interleaving two unrelated sentences. Academic papers and designed brochures are the usual offenders.
Table-heavy documents lose structure readily. A table that reads clearly on the page can come out as a run of disconnected values.
Slide decks exported to PDF convert poorly in a different way — the text is there but the meaning was in the presenter's narration, so the extracted content reads as fragments.
What to check in the output
Always read the generated modules before assigning them. This takes a few minutes and catches almost everything:
- Headers and footers repeating. Page numbers and document titles often appear mid-sentence where a page broke.
- Sentences cut at page boundaries. Look for text that stops abruptly and resumes oddly.
- Table content flattened into a list of values with no relationship.
- Lost emphasis. Bold and italics carry meaning in policy documents — "must" versus "should" is sometimes signalled by formatting alone.
- Order. Check the modules follow a sensible sequence, particularly if the original had sidebars or callout boxes.
Splitting is the real work
Extraction is only the first half. A clean extraction of a forty-page handbook is still forty pages — the reason it wasn't read hasn't changed.
The conversion that matters is structural: splitting into modules short enough to finish, making each one understandable on its own, and adding a check at the end so completion means something. Publishing the extracted text as one long course reproduces the original problem in a new tool.
Keep the PDF as the source of truth
If the PDF is the canonical version — which it often is for anything legal has approved — treat the course as derived from it rather than as a second copy.
When the policy is revised, regenerate from the new PDF rather than editing the course. Two copies of a compliance document that disagree is a worse problem than either one being slightly out of date.
Doing it in one step
Poplearn handles extraction and splitting together. /courses new takes the document, divides the material into modules, and writes the quizzes as it goes — then /assign #support sends it to a person, a channel, or every new hire from that date forward, and /progress gives live completion by team, exportable on demand.
Modules arrive as Slack direct messages rather than sitting in a portal. Free under 20 seats, installs in about a minute.
Read the generated modules first, as with any conversion. It takes a few minutes and it's the difference between a course that works and one that quietly embarrasses you.
