New in PaperScorer: describe the test, get the questions

Writing a good multiple choice question takes longer than it looks. The stem is the easy part. The wrong answers are where the time goes, because a distractor that is obviously wrong teaches nothing, and one that is accidentally right ruins the item. Multiply that by forty questions and a unit test is an evening's work before a single sheet is printed.

PaperScorer can now do the first draft. Describe what you want to assess in a sentence, and it drafts the questions, writes the distractors, checks the answer key, and drops the set into your assessment ready for you to edit.

One line in, a draft test out

There is no prompt engineering to learn. You describe the assessment the way you would describe it to a colleague:

5th grade water cycle, focus on evaporation and condensation

Then you set the shape of the test: how many questions, and how many answer options each one gets, anywhere from two to ten with five as the default. PaperScorer sends that off and, a few seconds later, the questions appear in your assessment as ordinary editable items.

What comes back for each question:

• A stem written to the topic and level you described
• Plausible distractors, not obviously wrong filler
• A verified answer key, with the options shuffled so the correct answer is not always in the same slot
• A link to a standard, if you generated against standards
• Optional topic tags for grouping and filtering later

The last step is the one that matters. Nothing prints until you have read it. Generated questions land as drafts in the same editor you already use, so you can reword a stem, swap a distractor, retag, reorder, or delete anything that misses the mark.

Standards are optional, not assumed

If you teach to standards, add them before you generate and every question comes back linked to the one it addresses. You can either set a number of questions per standard, or give one total and let PaperScorer spread it across the standards you picked. Because the link is made at generation time, mastery reporting works from the first scan.

If you do not teach to standards, skip them. You get the same questions with no standard attached, and you can tag them yourself afterward like any hand-written item.

Auto-tagging is a separate switch. Turn it on and each question gets one to three short topic tags for filtering in your item bank. It is off by default and uses no extra tokens.

It does not have to be a school subject

The prompt is plain text, so the generator will write questions about anything you can describe. Teachers use it for coursework. Trainers, coaches, tutors, and quiz writers use it for everything else:

• Knife cuts and kitchen safety for new line cooks
• Forklift safety refresher for warehouse staff
• Reading rhythm notation, beginner level
• World capitals and national flags, hard mode

A non-academic set simply comes back without standards.

What it costs

One token generates one question. Tokens are a one-off purchase, not a subscription, and they sit in your balance until you spend them.

• 25 tokens for $3.00, which is $0.12 per question
• 50 tokens for $5.00, which is $0.10 per question
• 100 tokens for $9.00, which is $0.09 per question
• 250 tokens for $20.00, which is $0.08 per question
• 500 tokens for $35.00, which is $0.07 per question
• 1,000 tokens for $60.00, which is $0.06 per question, the best rate available

Generation is available to every PaperScorer user on every plan, including Free. There is no free token allotment. New accounts start at a zero balance and buy the package that fits.

What it does not do

The specifics, so you know what you are buying before you buy a token:

• Multiple choice only. True/false, numeric, matching, ordering, and written response are still built by hand, and you can mix them into the same assessment alongside generated questions.
• Managed assessments only. If you are scanning against an existing paper test where PaperScorer only holds the answer key, there is no question content for the generator to add to.
• Nothing publishes on its own. Every generated question is a draft until you review it.
• It does not know your classroom. A question that references Tuesday's discussion still has to come from you. Course-specific prompts are what make an assessment hard to shortcut, and that argument is laid out in the True Test of Knowledge approach.

AI writes the questions, paper keeps the answers honest

We have argued on this blog that generative AI has made digital take-home work impossible to trust, and that the fix is testing on paper, in person, under supervision. So it is fair to ask whether selling AI question writing contradicts that.

It does not, because they are different problems. The integrity problem is a student handing in work a model wrote and having it graded as theirs. Who drafted the question is a separate matter. A well-written item is well-written regardless of what typed the first version, and a supervised paper test is trustworthy regardless of how its questions were authored. AI helping you build the test does not weaken the test. If you want the longer version, read why paper tests are making a comeback.

On the data side: questions are generated by a commercial large language model accessed over an API, and your assessment content is not used to train it. If your district needs specifics for an AI review or a procurement questionnaire, get in touch and we will answer in writing.

Where it fits with the other ways to build a test

Two weeks ago we counted four ways to get questions into an assessment: upload a document, import from your LMS, generate with AI, or pull from your item bank. This is the one we had not written up properly.

They combine. Import last year's midterm, generate ten fresh questions on the unit you added this year, pull two from your item bank, and print the result as one assessment. Generated questions go into your item bank like any other managed item, so a good set written once is available for every test after it.