How to turn a filing cabinet of past exam papers into online practice tests

Almost every licensing school I talk to owns the same underused asset: years of past exam papers. Salesperson practice exams, air brakes packets, infection control review sheets, code question papers — built up by instructors, refined every intake, and calibrated to the exam far better than any generic question bank.

And almost all of it is stuck on paper, or in a Word document that a former instructor formatted in a way nobody else can maintain.

This post is about the route from there to online practice, and the parts of it that people usually get wrong.

Why "just retype it" fails

The obvious plan is to have someone type the questions into whatever platform you already have. It fails for a boring reason: a few hundred questions at a couple of minutes each is a week of someone's working time, and it has to happen again every time the material changes.

So it gets scheduled, deprioritised, and never done. The papers stay in the cabinet.

Step 1: pick the smallest set that proves the point

Do not start with everything. Pick one paper — one topic, thirty or forty questions — that you know your students struggle with.

The goal of the first pass is not coverage. It is to find out what your material looks like after it has been through extraction, while the stakes are low enough that you will actually finish.

Step 2: capture, do not transcribe

Modern extraction works from images and documents directly. Photograph the pages, or upload the PDF or Word file you already have. What matters is that the question stem, the options, and the marked answer are all legible in the source — extraction can only find what is actually there.

Two things that genuinely hurt accuracy, in rough order of severity:

• Missing answer keys. If the correct answer lives on a separate sheet, or in an instructor's head, no amount of processing will recover it. Get the key into the same document first.
• Shadowed or skewed photographs. A phone photo taken flat under normal light is fine. One taken at an angle in a dim room is not.

Handwriting is usually workable if it is high-contrast. Very old photocopies of photocopies are the hardest case, because the option letters bleed.

Step 3: treat every extracted question as a draft

This is the step people skip, and it is the one that matters.

AI extraction is good at structure — finding where one question ends and the next begins, separating the stem from the options, spotting which option is marked. It is not reliable at judgement. If your answer key is ambiguous, or two options are nearly identical, or the question refers to a diagram, it can produce something plausible and wrong.

A wrong answer key is worse than no question at all. It teaches the mistake, and students trust it because it came from the school.

So the workflow has to end with an instructor reading each question and approving it. Not spot-checking a sample — reading each one. It is fast, because you are reviewing rather than typing, but it is not optional. Any platform that publishes extracted questions straight to students without an approval step is handing you a liability.

Step 4: tag to your own structure, not a generic one

When the questions go in, tag them to the subjects and topics your course already uses — the national and state split, the endorsement, the code section, whatever your syllabus says.

This feels like admin. It is actually the entire reporting payoff. "This cohort is at 58% overall" tells you nothing you can act on. "This cohort is at 41% on the state law section" tells you what to re-teach on Thursday.

Get the tagging right on the first forty questions and every later import inherits the structure.

Step 5: check what the student actually sees

Open the result on a phone, because that is where it will be used. Students practice in the gaps — between yard time and road time, on a break on the clinic floor, on the bus. Long stems that require horizontal scrolling get abandoned.

What this does and does not fix

Getting your papers online will not raise pass rates by itself. Nothing does; anyone who promises you a pass-rate number is selling something.

What it changes is the timing of information. On paper, you learn who was going to fail after they have failed. With practice online and tagged by topic, you learn which students and which topics are behind while there is still a week of class left to do something about it.

That is a smaller claim than most software makes. It is also the one that actually holds up.

A note on buying content instead

There is a shortcut: license a ready-made question bank and skip all of this.

Sometimes that is the right call. But be clear about the trade. Your own past papers are calibrated to the exam your students sit and to the way you teach it. A generic bank is the same one the school down the road can license. If your material is part of why students choose you, replacing it with someone else's is worth thinking about carefully.

There is also a copyright question that is easy to get wrong. Exam questions have owners. If a platform hands you a large library of items that look a lot like real exam content, it is worth asking exactly where those items came from and what licence covers them — because the school's name is on the course, not just the vendor's.