How to Convert Old Question Papers into an Import-Ready CSV with AI

Every coaching institute has them: folders of old question papers saved as PDFs, chapter tests typed up in Word, answer keys on the last page. They’re full of good questions, and almost none of them are in a form your exam software can use.

Retyping them is slow and error-prone. An AI assistant such as ChatGPT, Claude or Gemini can do the restructuring for you in minutes: take messy text from a PDF or Word file and turn it into clean rows that match your question bank’s CSV template. This guide shows the exact prompts we use, how to handle maths and answer keys, and, most importantly, how to check the output so mistakes don’t reach your students.

Before you start: what you need

• Your question bank’s CSV template. Download it from your exam software’s import page. The AI needs the exact column names.
• The source file. A PDF, Word document or even a photo of a printed paper.
• An AI assistant. ChatGPT, Claude or Gemini all handle this well. A paid plan helps with long files, but the free versions work in small batches.
• The right to use the questions. Your own papers or material you’re licensed to use. Don’t upload papers that contain students’ names or marks.

Step 1: Get the text out of the PDF or Word file

How you start depends on the kind of file you have:

Scanned papers need the most care. OCR often confuses similar characters, such as O and 0 or l and 1, and drops symbols like √ and π. Plan to check those questions more closely in step 4.

Step 2: Use a prompt that matches your template

Vague prompts like “convert these questions to CSV” give you a CSV, just not one your importer will accept. The fix is to hand the AI your exact header row and one row you’ve filled in yourself. Here’s the prompt we use. Replace the parts in square brackets:

You are helping me move exam questions into my question bank.

Convert the questions below into CSV rows that match this header exactly:
[paste your template's header row]

Here is one correctly filled example row:
[paste one row you filled in yourself]

Rules:
- Copy every question and answer choice word for word. Do not rephrase, correct or shorten anything.
- One question per row. Keep the questions in their original order.
- Use keys in this pattern: [for example PHY-CH3-001, PHY-CH3-002 ...]
- Put the correct answer only where the source clearly gives it. If the answer is not in the source, leave that cell empty and add NEEDS-ANSWER in the notes column. Never guess an answer.
- Write maths and chemistry in LaTeX, for example \(x^2\) and \(\ce{H2O}\).
- If a question refers to a diagram or image, add NEEDS-IMAGE in the notes column.
- Put any text that contains a comma inside double quotes.
- Return only the CSV in a single code block, with no explanation.

Questions:
[paste 20–30 questions here]

If your template has no notes column, ask the AI to list the NEEDS-ANSWER and NEEDS-IMAGE questions separately after the code block instead.

Step 3: Handle answer keys, maths and diagrams

These three cause most of the trouble, so deal with them on purpose.

Answer keys at the end of the paper

Many papers print the answers on the last page (“1-B, 2-D, 3-A…”). Paste that key after the questions and add one line to the prompt: “Use this answer key to fill the correct answer column, matching by question number.” Then spot-check five or six rows against the original key, especially near page breaks where numbering often slips.

Maths and chemistry

Formulas copied out of a PDF often arrive broken: superscripts flattened (x2 instead of x²), fractions split across lines. AI assistants are good at rebuilding them as LaTeX, which most modern exam software can render. Asking for LaTeX also makes the CSV safer, because everything stays plain text. Still, read every formula once. A misplaced bracket changes the answer.

Diagrams and images

A CSV can’t hold images. Let the AI flag those questions with NEEDS-IMAGE, import them as text, and attach the diagram afterwards in your question editor. Crop the diagram from the original paper or redraw it if the scan is poor.

Step 4: Check the output before importing

AI does the typing. You still own the accuracy. Paste the CSV into a spreadsheet (or save it as a .csv file and open it) and run through this checklist:

• Count. The number of rows matches the number of questions in the batch.
• Wording. Compare five random questions with the original, word for word.
• Answers. Check every correct answer against the source key, or at least a solid sample if the key is printed cleanly.
• Flags. Filter the notes column for NEEDS-ANSWER and NEEDS-IMAGE and deal with each one.
• Keys. No duplicates. A quick =COUNTIF(A:A, A2) helper column shows any repeats.
• Encoding. Save the final file as CSV UTF-8 so symbols survive.

Step 5: Import and review

Upload the file through your exam software’s import page and let it validate every row. Fix whatever it flags, re-upload, and import once everything passes. If you haven’t done a CSV import before, our step-by-step guide on how to import questions from a CSV file covers templates, encoding and the common errors in detail.

Treat imported questions as drafts. Preview a few from each batch the way students will see them, check that formulas render, and only then add them to live tests.

How this works in Academy & Exam OS

We built Academy & Exam OS so that AI-assisted question entry stays safe:

• The CSV import validates every row first and is all-or-nothing, so a bad batch never half-imports.
• Imported questions arrive as drafts. An owner or administrator approves them before they can appear in a test.
• Maths and chemistry are written in LaTeX and mhchem, and the question editor shows a live preview exactly as students will see it.
• If you connect an AI provider, owners and administrators can also generate new question drafts inside the platform. Staff review them before anything is saved to the question bank.

FAQ

Start with one paper

Pick a single chapter test, run one batch of 20 questions through the prompt, check it, and import it. Once you’ve done that end to end, converting the rest of the archive is just repetition, and an afternoon’s work can bring years of questions back into use.

More practical guides like this are on the way. Join the OsDot newsletter to get them first.