Prompting tips for multiple choice questions (MCQ)

Best practices for MCQ generation prompting

The first few sentences of the prompt are prioritized by Item Assist. Therefore, we recommend the following:

The topic of the exam is added to the first sentence.
The instructions for key length to be similar to distractor length should be one of the first few sentences of the prompt.
The lowest levels of the blueprint are identified in the first sentence, leaving the highest level (the domain) out, to ensure content is better aligned to the selected blueprint level.
The Exemplar MCQ placeholder is added within the sentence that includes the blueprint topic. If your prompt does not include a blueprint topic, add the Exemplar placeholder near the beginning of the prompt, within the description of the Task.
All other metadata and additional author prompts are placed near the beginning of the prompt. The prompt includes the recommended additional author prompts (Additional Instructions and Exemplar MCQ) and the recommended metadata (Cognitive Level and Question Style).

Sample prompt text

The prompt text below ↓ is a sample of a content generation prompt. It is not the default prompt, which is data-based and tested for efficacy with the current LLM model. (Click here to learn more about the default prompt.)

This sample prompt is not tested for efficacy and will not work as well as the default prompt. It is provided as an example of data-based best practices for MCQ (multiple choice question) generation.

The reference to "claims adjuster certification" in the first sentence below is an example that can be substituted with your own relevant exam topic.

Task: Generate {{$itemcount}} assessment items that sample meaningful, representative knowledge and skills central to "{{$Level 2}}: {{$Level 3}}" for a claims adjuster certification exam {{$Exemplar MCQ}}.

You are an expert high‑stakes assessment developer.

For each item, you must follow typical professional assessment style requirements, including:

Write a key that is of comparable length to the distractors.

Use precise, unambiguous language that a well‑prepared test‑taker at this level can understand on a first read. Minimize construct‑irrelevant reading load, especially for items focused on non‑language skills.

Explicitly avoid trivia and overly narrow “gotcha” content.

Provide one clearly correct key.

Write any equations in MathML.

Write plausible distractors of comparable style to the key.

Ensure distractors reflect common misconceptions or typical errors.

Avoid writing distractors with random or obviously incorrect statements.

Avoid options such as "All of the above” or "None of the above."

Avoid grammatical or logical cues that reveal the correct answer (e.g., key is longer or more specific, stem–option grammatical mismatches, absolute terms only in distractors).

As a final step, self‑review the items for clarity, fairness, alignment, and internal consistency, and revise any items that do not meet high‑stakes assessment standards.