How to Check AI-Generated Medical Questions Against Your Lecture

Open the exact page attached to the answer. Find the sentence, diagram label, or table cell that supports the tested fact. A link to a relevant-looking lecture is not enough. Check whether the page number refers to the PDF page or a printed slide number.

• Can you find the supporting passage?
• Does it cover the same population, condition, and context?
• Does the explanation add a claim that the passage does not contain?

Use the public bone-and-joint reference PDF, page 3 of 3, or open the page image. The example below is an edited teaching example based on our existing public demonstration, not a fresh model run or an independent clinical benchmark. It checks agreement with the sample rather than diagnosing a patient.

Read the exact source row

The small page references printed below the sample table refer to the longer teaching material used to prepare it. For this demonstration, the downloadable file has three pages and the table is on PDF page 3. Do not search this short PDF for a printed reference such as page 12.

Answer before reading the reasoning

1. Septic arthritis
2. Cellulitis
3. Osteomyelitis

Check the answer and every alternative

The explanation should identify the affected compartment. No treatment, investigation result or diagnostic threshold needs to be invented to answer this source-based question. If a generated explanation adds such a claim, that claim needs its own support.

Catch a draft that sounds plausible but fails the source check

The draft confuses proximity with the infected compartment. The table supports septic arthritis for the joint space. Correcting the label is not enough: the explanation must also refer to the joint-space entry. This example checks agreement with the table, rather than diagnosing a patient.

Record the check and practise again

• Source: bone-and-joint reference PDF, page 3, Site row.
• Tested fact: the infected compartment, rather than a treatment decision.
• Answer support: each option was checked against the same table.
• Revision action: keep only the checked version and answer it again without looking at the row.

For an interactive example, try the existing SBA demonstration. For your own lecture, choose the relevant pages in the SBA workflow, generate a draft and apply the same checks. Once you have resolved doubtful items, practise saved questions in Custom Test and use Weakness Review to revisit missed topics. Generated wording can differ from this edited example.

For a single best answer question, the wording must make one option the best choice in the stated context. Plausible distractors should be distinguishable using the information supplied. If two answers remain defensible, flag the question instead of memorising the tool's preferred label.

• Identify what the lead-in actually asks.
• Explain why the selected answer fits that question.
• Check whether each distractor's explanation is supported.
• Watch for an invented age, investigation, or qualifier that changes the answer.

If a source is missing, ambiguous, or conflicts with another reference, keep the item out of your trusted revision set until you resolve it. Ask your educator about course-specific ambiguity and use appropriate current clinical references for medical claims. Do not use a generated practice item to make a patient-care decision.

Record what was wrong and the source used to correct it. Then answer the corrected question from memory. Retrieval practice and spaced review can support learning, but repeating an unchecked answer can reinforce the wrong association.

1. Open the cited page and identify the exact supporting passage.
2. Separate the sourced fact from any fictional case framing.
3. Confirm that one answer fits the question's context.
4. Check the distractors and explanation, not only the answer letter.
5. Flag unresolved claims and record corrections.
6. Practise the corrected item, then schedule another review.

Try the process on the PDF-to-questions example, then use your own lecture when you are ready. Recall Engine accounts are currently available to medical students in Egypt; this guide and the public sample are available worldwide.