AI Exit Tickets: Prompts for Checks That Tell You Something

What is an AI exit ticket?

An exit ticket is a short check for understanding — one to three questions — that students complete in the last minutes of class. An AI exit ticket is one you draft with an AI tool from two inputs: today's objective, and the wrong answers you expect. Seconds to generate, minutes to check, no more guessing about what landed.

This is the deep version of a prompt that appears once in our guide to AI lesson planning. That article covers the whole lesson; this one covers the last five minutes, because that's where the next lesson gets decided.

Why an exit ticket beats "any questions?"

"Any questions?" samples exactly the wrong population: the students confident enough to speak and self-aware enough to know what they don't know. A confused student's most rational move at 2:58 p.m. is to stay quiet and hope.

An exit ticket asks everyone, in writing, before they leave. That puts it inside the practice with the strongest evidence base in classroom assessment: Black and Wiliam's 1998 review "Inside the Black Box" (Phi Delta Kappan) concluded that formative assessment — checking understanding as you go and adjusting instruction to what you find — reliably raises achievement, most of all for the students furthest behind. The exit ticket is the lowest-prep daily version of that practice.

But only if the ticket is diagnostic, and this is the design principle the whole prompt bank rests on: a good exit ticket makes wrong answers informative. A question where every wrong answer is just "wrong" tells you who to worry about; a question where each wrong answer maps to a specific misconception tells you what to do. That second kind is where AI earns its place: distractor-writing is the tedious craft skill of assessment design, and AI does it in seconds — provided you feed it the misconceptions. (It can't know those; more at the end.)

The 3-question shape

Most of the prompts below ask for three questions, and the number is doing work:

1. A floor question most students who were present and awake should get. Missed, it says the problem is upstream of today.
2. The objective itself, in a form slightly different from the one you practiced — same skill, new surface.
3. A misconception trap: a question a student gets wrong in a predictable way if they're carrying a specific wrong idea.

One question can't distinguish "they got it" from "the question was easy"; more than three or four is a quiz you won't have time to read tonight. Three gives you a gradient per kid in about ten seconds of looking.

A ticket written separately from the lesson also tends to drift from what the lesson actually taught. If you draft lessons with AI, draft the exit ticket in the same conversation so both work from the same objective — this is why Planning Partner drafts the exit ticket alongside the plan, worksheet, and answer key, so the ticket can't drift from the lesson it checks.

The prompt bank: 10 patterns by what you want to learn

Nine patterns here, a tenth in the next section (it needs responses to exist first). Swap the grades, topics, and wrong answers for your own; subjects vary on purpose — the pattern is the point.

Did they absorb the concept or just today's example?

Misconception check: "Write a 3-question multiple-choice exit ticket for 7th-grade math on adding negative integers. My students carry two wrong ideas: that adding always makes a number bigger, and that -8 is greater than -3 'because 8 is bigger.' Build distractors from those misconceptions, and give me a key that says which wrong idea each distractor detects."

Tomorrow morning you count answer choices, not just right and wrong. Nine kids on the "-8 is greater" distractor means the gap is number lines, not addition — a different re-teach entirely.

Transfer to a new context: "My 6th-grade science class learned photosynthesis today using a diagram of a maple tree. Write a 2-question short-answer exit ticket that applies the same process to organisms we never mentioned: algae in a pond, and grass that yellowed under a kiddie pool left out all summer."

This separates students who learned the concept from those who memorized the diagram. A kid who can label every arrow on the maple tree but can't explain the yellow grass learned a picture.

Generate an example: "Today's 5th-grade ELA mini-lesson was personification. Write an exit ticket asking students to (a) write one original example of personification about our cafeteria and (b) write one vivid sentence about the cafeteria that is NOT personification. Two-line teacher key on what counts."

Recognition is cheap; production isn't. The not-example is the diagnostic half — it catches the students who've decided every vivid sentence is personification.

Can they hold it longer than a class period?

Retrieval from last week: "Write a 3-question exit ticket for high-school U.S. history. Questions 1–2 pull from last week's material on causes of the Great Depression — pure recall, no notes. Question 3 asks students to connect one New Deal program from today's lesson to one of those causes."

The recall questions aren't filler — the act of retrieving is the intervention. In Roediger and Karpicke's 2006 experiments (Psychological Science), students tested on material retained substantially more a week later than students who restudied it the same number of times, even though restudying felt more effective in the moment. A weekly look-back question turns your exit ticket into free retention.

Confidence pairing: "Write a 3-question multiple-choice exit ticket on balancing simple chemical equations, with distractors built from real balancing errors — especially changing subscripts instead of coefficients. After each question add: 'How sure are you? Sure / Mostly / Guessing.'"

This sorts responses into four piles instead of two: right-but-guessing is fragile knowledge that won't survive the weekend, and wrong-but-sure is the pile you plan tomorrow around. Cognitive scientist Pooja Agarwal's retrievalpractice.org names the second return specifically: retrieval paired with self-assessment builds metacognition — students learn to notice what they do and don't know.

Can they work with it, not just answer it?

Find the mistake: "Here's my exit-ticket problem for 8th-grade math: solve 3(x − 4) = 18. Write a worked solution that makes one specific common error — distributing the 3 to the x but not the 4 — and arrives at a confident wrong answer. The ticket asks students to find the broken step and fix it. Include the corrected solution for my key."

A student who can run the procedure but can't spot a broken version of it is one distracted Tuesday from making the same error. The "I can't find a mistake" responses are your list for tomorrow's small group.

One-sentence synthesis: "We read chapter 9 of To Kill a Mockingbird today. Write an exit ticket asking students to state what this chapter reveals about Atticus in exactly one sentence containing the word 'because.' Give me a model answer at two levels — proficient and developing — so I can sort fast."

The forced "because" converts a summary into a claim with a reason — the actual ELA skill. Sentences that stall before the "because" show who has plot but not inference.

Claim vs. evidence: "Write a 2-part exit ticket for U.S. history. Part 1 shows this claim: 'The colonists at the Boston Massacre were innocent victims.' Part 2 asks whether Paul Revere's engraving supports, complicates, or contradicts it — two sentences, referencing a detail from the image. I'll attach the engraving to the ticket, so leave space for it."

Responses tell you who reads a source as information and who reads it as an argument. "Supports, because it shows the British shooting" and "complicates, because Revere made it as propaganda" are two different students; only the second is doing history.

Is tomorrow's lesson standing on solid ground?

Tomorrow's prerequisite: "Tomorrow I teach adding fractions with unlike denominators to 5th grade. Write a 2-question exit ticket for TODAY that checks the two skills tomorrow leans on: generating equivalent fractions and finding a common multiple. Four minutes, max."

An exit ticket doesn't have to look backward. This one tells you tonight, while you can still adjust, whether tomorrow's lesson has a floor or needs ten minutes of repair built into its opening.

What to do with the stack

The stack is only worth collecting if it changes something before first period. Sort, don't grade — three piles, under ten minutes:

• Most of the class got it (say, 80%+): move on. The handful who didn't become tomorrow's small group or get a scaffolded version of the next task — a differentiation move with its own guide.
• The class split: don't re-teach the lesson; re-teach the error. The most-picked wrong answer becomes tomorrow's warm-up, draftable in one prompt:

The morning after: "Exit-ticket results: 11 of 26 students said -8 is greater than -3. Write a 5-minute warm-up that starts from that answer and makes the error visible — put both numbers in a context (temperature, debt) where the wrong ranking obviously fails. Attribute the wrong answer to 'a student from another class.'"

• Most of the class missed it: re-teach — with a different representation, not the same explanation louder. If they missed it in symbols, come back in a picture or a context. The ticket told you the first route didn't work; repeating it is the one move the data rules out.

One boundary: sorting slips into piles isn't grading. When the responses are paragraphs that need scores or substantive comments, that's a different job with different rules — see AI grading and feedback.

Where AI gets this wrong

Two limits, both the kind that quietly ruin an unread ticket.

AI distractors can be too clever. Ask for "plausible wrong answers" and a model sometimes produces an option that's wrong for a defensible reason — or two options a careful reader can justify — so your data measures test-wiseness, not the misconception. The related failure is reading-level creep: a math ticket whose question stems quietly turn it into a reading test. The check for both is the same and it isn't optional: answer every question yourself, as a student, before you print. Roughly two minutes per ticket.

And AI doesn't know your class's actual misconceptions. It knows the literature's greatest hits — the classic fraction errors, the classic force-and-motion errors — which is a fine default when you're new to a topic. But "my third period thinks X" is data only you have, collected from boardwork, questions, and yesterday's tickets. Every prompt above works better the more specific that part gets, and no version of the tool supplies it for you.

Which is the practical reason to keep the sorted stacks: last week's wrong answers are exactly what you paste into next week's misconception-check prompt.