AI for Formative Assessment: A Practical Teacher Guide

Formative assessment is any check on learning that a teacher uses to decide what to do next: move on, reteach, regroup or give targeted practice. AI helps formative assessment by making it faster to write good questions, quicker to collect whole-class evidence and easier to spot patterns in student responses. It does not replace the decision itself, which stays with the teacher.

This guide covers where AI fits in formative assessment, five practical workflows, how to write questions that reveal misconceptions, how to use the evidence, and the guardrails that keep it trustworthy.

The short answer

• Start with the decision. Before writing any question, decide what you will do differently depending on the answers.
• Use AI to draft questions, not to judge them. AI is quick at drafting hinge questions, exit tickets and quizzes aligned to an objective; you check that they test what you taught.
• Collect whole-class evidence. Live quizzes, mini-whiteboards and short exit tickets give you information about everyone, not just the students who put their hands up.
• Look for patterns. Use AI to group wrong answers by misconception and suggest a reteach.
• Close the loop in the next lesson. Evidence that does not change teaching is just data collection.

What formative assessment is, and is not

The idea comes from a long line of research, most famously Black and Wiliam’s Inside the Black Box, and later work by Dylan Wiliam describing five strategies:

1. Clarifying and sharing learning intentions and success criteria.
2. Engineering effective classroom discussions and tasks that elicit evidence of learning.
3. Providing feedback that moves learners forward.
4. Activating students as learning resources for one another.
5. Activating students as owners of their own learning.

Formative assessment is not a type of test. The same quiz can be formative (used to decide what to teach tomorrow) or summative (used to record a grade). What makes it formative is how the evidence is used.

That matters for AI. A tool that generates a hundred quiz questions is not doing formative assessment. A teacher who uses five of those questions at a planned moment, reads the responses and adjusts the next lesson is.

Where AI helps, and where it does not

AI helps with

• Drafting questions quickly that target a specific objective.
• Writing plausible wrong answers that reveal common misconceptions.
• Producing variations so you can re-check the same idea later without repeating questions.
• Running whole-class checks through live quizzes that collect every student’s answer.
• Summarising patterns in responses, such as which distractor was chosen most often.
• Drafting feedback that you then edit and personalise.

Teachers stay responsible for

• Choosing what to check and when.
• Judging whether a question is valid, meaning it tests the intended learning and not reading speed or guesswork.
• Interpreting evidence in the context of the class.
• Deciding the response: what to reteach, to whom, and how.

Five AI formative-assessment workflows

1. Hinge questions mid-lesson

A hinge question is a single, quick question at a pivotal point in a lesson. Students’ answers tell you whether the class is ready to move on.

How to use AI

1. Give the objective and the concept you just explained.
2. Ask for three or four multiple-choice hinge questions, each with distractors tied to a named misconception.
3. Choose the best one. Check that a student who understands the concept would get it right quickly, and that each wrong answer points to a specific gap.

Prompt pattern

Write four multiple-choice hinge questions for [grade level] on [concept]. Each question should take under a minute. For each wrong answer, state the misconception it reveals.

Decision rule: decide in advance what proportion of correct answers means you move on, and what you will do if it is lower.

2. Exit tickets

An exit ticket is two or three questions at the end of a lesson that check the objective directly.

How to use AI

1. Paste the lesson objective and a short summary of what was taught.
2. Ask for one recall question, one application question and one explanation question.
3. Ask for a short marking guide so you can sort responses quickly into “got it”, “partly” and “not yet”.

Tip: sort exit tickets into three piles rather than marking them. The piles are your plan for the start of tomorrow’s lesson.

3. Live classroom quizzes

A live quiz gives you every student’s answer in real time. It is particularly useful as a retrieval starter or a mid-unit check.

How to use AI

1. Generate a short quiz from the lesson content or the unit’s materials.
2. Preview every question and remove anything ambiguous.
3. Run it live and watch which questions most students miss.
4. Stop after the quiz to discuss the hardest question while it is fresh.

In Duetoday, you can create a live quiz from your own content, a YouTube link or a web page, set question types, grade level, difficulty and hints, and students join at duetoday.ai/join with a code. The live classroom quizzes guide covers this in depth.

4. Misconception analysis

Once you have responses, AI can help you see the pattern.

How to use AI

1. Collect anonymised responses: the answers, not the names.
2. Ask the AI to group wrong answers by the misconception they suggest.
3. Ask for a short reteach activity for each group.

Prompt pattern

Here are anonymised student answers to the question “[question]”. Group the incorrect answers by the misconception they suggest. For the two most common misconceptions, suggest a five-minute reteach activity and one follow-up question to check it worked.

For turning that evidence into feedback quickly, see AI grading for teachers.

5. Feedback that moves learning forward

Feedback is formative when students act on it. AI can draft comments, but the teacher decides what matters.

How to use AI

1. Draft or select success criteria for the task, ideally as a short rubric.
2. Use AI to draft feedback against those criteria: one strength, one specific next step.
3. Edit each comment so it sounds like you and refers to the student’s actual work.
4. Build in time for students to act on the feedback in class.

Guardrail: do not paste student work containing names or personal details into AI tools unless your school’s policy allows it.

Writing questions that actually reveal thinking

AI makes it easy to produce lots of questions. The skill is choosing ones that reveal what students think.

Good formative questions

• Have one clearly correct answer for multiple choice, or a clear success criterion for open questions.
• Use plausible distractors that a student with a specific misconception would choose.
• Test the objective, not vocabulary or reading speed.
• Are quick to answer so you can collect evidence without losing the lesson.

Questions to avoid

• “Trick” questions that test attention rather than understanding.
• Questions where the correct answer is the longest or most detailed option.
• “All of the above” and “none of the above,” which often reward test-taking strategies.
• Questions that can be answered from general knowledge without the lesson.

A quick check for AI-generated questions: for each question, ask “if a student gets this wrong, do I know why?” If the answer is no, rewrite or replace it.

Using the evidence

Evidence only matters if it changes what happens next. A simple routine:

1. Before the lesson, decide what you will do for each likely outcome. For example: most correct, move on; split, run a quick reteach with a new example; most wrong, reteach with a different explanation.
2. During the lesson, collect the evidence at the planned moment.
3. After the lesson, note the two or three biggest gaps.
4. Next lesson, start with a short retrieval question on those gaps.

AI can help at step 4 by generating fresh questions on the same concept, so students are not just remembering the previous answer.

Retrieval practice as formative assessment

Low-stakes retrieval practice (short quizzes on content from previous lessons and weeks) does two jobs at once: it strengthens memory and tells you what has stuck. AI makes it easy to generate a steady supply of retrieval questions across a unit.

• Mix old and new content in starters rather than only testing last lesson.
• Keep it low stakes: no grades, quick feedback.
• Track which topics keep coming up wrong and plan for them.

The live classroom quizzes guide shows how to run these checks with the whole class.

Guardrails for using AI in assessment

• Check every question before students see it. AI can produce incorrect answers, ambiguous wording or questions that do not match what you taught.
• Protect student data. Use anonymised responses for analysis unless your school has approved otherwise.
• Keep high-stakes judgements human. AI drafts can support marking, but grades that matter should be reviewed by a teacher.
• Watch for bias in names, contexts and examples.
• Be transparent with students about how quizzes and feedback are produced.

Common mistakes

Assessing too much. More data is not better. One well-chosen hinge question can be more useful than a twenty-question quiz.

No planned response. If you have not decided what you will do with the answers, the check is unlikely to change your teaching.

Testing recall when the objective is reasoning. If the objective is to explain or justify, include questions that require it.

Treating AI output as finished. Generated questions need the same scrutiny as questions from any other source.

How to do this in Duetoday

1. Plan the lesson and the check together using the Lesson Plan Generator; see the complete guide to AI lesson planning.
2. Build a quiz or assessment from the same content with the Assessment Builder, or create a live quiz.
3. Run a live quiz in class. Students join at duetoday.ai/join with the join code; see how to run a live classroom quiz.
4. Draft a rubric with the Rubric Generator for extended answers.
5. Share practice materials with students through a study room so they can keep practising after the lesson.

Duetoday has no free plan; every account starts with a 7-day free trial (card required, cancel before day 7 to pay nothing), then $15 a month or $99 a year.

Frequently asked questions

Is an AI-generated quiz formative assessment?

Only if you use the results to make a decision about teaching. The quiz is a tool; formative assessment is the process of collecting evidence and responding to it.

How often should I run formative checks?

Most lessons benefit from at least one planned check during the lesson and one at the end. The key is that each check is quick and leads to a decision.

Can AI mark open-ended answers?

AI can draft feedback and suggest a score against criteria you provide, which can save time. For anything that counts towards a grade, review the output yourself.

What is the difference between a hinge question and an exit ticket?

A hinge question comes at a pivotal point during the lesson and decides whether you move on. An exit ticket comes at the end and checks the lesson’s objective to inform the next lesson.

Should students know quizzes are low stakes?

Yes. Students answer more honestly when they know a quiz is for checking learning rather than for a grade, which makes the evidence more useful.

Further reading

• AI lesson planning: the complete guide for teachers
• Live classroom quizzes: a complete guide for teachers
• AI grading for teachers: a practical guide
• AI differentiation strategies for teachers
• Education Endowment Foundation: Teaching and Learning Toolkit