What education and training teams use it for
The jobs that pay off are the ones where you already know what good looks like and the time goes into producing it: twenty quiz questions, a rubric with four levels, the same passage at three reading levels. Each job below lists the templates and tools that do it.
Quizzes and question banks from your material
Paste the chapter, the lab handout or the safety procedure and ask for a set number of questions across recall and application, with an answer key that names the section each answer comes from. Plausible wrong answers are the hard part of a multiple-choice item, so ask for distractors drawn from common mistakes you list. Then check the key yourself; a source line beside every answer makes that a ten-minute job rather than an hour.
Write quiz questions with an answer keyPrompt GraderPrompt Tester
Rubrics that match the assignment
Give the prompt the assignment sheet and the standards it addresses, and ask for criteria with observable descriptors at each level, not “excellent” and “good.” When three teachers share a course, save one rubric prompt and compare versions in the diff before anyone changes the weighting, so every section is graded against the same scale.
Build a grading rubric from an assignmentVersion CompareMy Prompts
Lesson and session plans
A plan prompt works when it carries the constraints a real class has: 50 minutes, 28 students, one set of lab kits, a fire drill on Thursday. For workplace sessions the constraints are different (adults pulled off the floor for an hour, a mix of shifts, a sign-off at the end) and the session template is built for those. Ask for timings on every block and one check for understanding before the end.
Plan a training session for adults at workBuild an agenda from what it must decidePrompt Grader
Reading levels, translations and family emails
Differentiating one passage for three groups is slow by hand and quick with a prompt that keeps the key terms fixed and adds no facts. Family emails about a missed assignment or a behavior concern need a calm, specific draft and a teacher’s final read. For families who read another language, translate from the approved English version and have a fluent colleague compare two or three models blind in Model Compare before you settle on one.
Rewrite expert text in plain languageTranslate for local readersWrite a difficult emailModel Compare
E-learning storyboards and training video
Instructional designers spend days turning a 40-page procedure into screens, narration and knowledge checks. A storyboard prompt does the first cut screen by screen from the source, flags steps the source leaves unclear, and leaves the subject-matter expert a short list of questions instead of a full rewrite. A video prompt can then describe each task shot for a generator or a camera operator.
Turn source material into an e-learning storyboardTraining video shot of a workplace taskModel Compare
Onboarding and compliance training at work
L&D teams rebuild the same onboarding every quarter. Draft the 90-day plan, the first-week guide and the written procedure from your own notes, then turn the procedure into a short quiz so completion means something. Keep the policy text pasted in rather than summarized from memory; a training module that misstates the leave policy creates more work than it saves.
Plan a new hire’s first 90 daysWrite a first-week guideTurn a process into an SOP
Course evaluations and survey comments
End-of-course surveys produce hundreds of comments that nobody reads past the first page. A themes prompt groups them, counts how many comments support each theme and quotes two or three verbatim, so the department meeting starts from evidence. In the Prompt Tester, an AI judge check can flag a quote that does not appear in the comments you pasted, and you read the ones it flags.
A worked example: a department quiz prompt
Take a hypothetical high school biology department of four teachers who each write a weekly quiz for the same course. They want one prompt that every teacher trusts, so a quiz on cell transport looks the same in every section and the answer keys are right.
- 1Start from the quiz templatePaste the chapter and set the number of questions, the question types and the learning objectives.
- 2Grade the promptThe free estimate checks scope and form, including what to do when the chapter does not cover an objective. No key needed.
- 3Test it on eight chaptersEach chapter is one test case. Checks count 15 numbered questions and a source line beside every answer.
- 4Compare models blindRun the hardest chapter on three models and let a colleague vote without seeing which is which.
- 5Price a semesterEnter the department’s weekly volume and compare what each model would cost.
- 6Share the department versionSave it with its history and send colleagues a share link, so every section uses the same tested prompt.
The eight chapters are the test cases, and the useful ones are awkward: the chapter that is mostly diagrams, the one with a lab procedure in the middle, the one that covers an objective in a single paragraph. A “Matches a pattern” check with a count of exactly 15 catches a quiz that quietly drops to 12 questions. A “Contains text” check for “Source:” catches an answer key that stops citing the chapter halfway through.
Running the suite needs an API key. Google AI Studio, OpenRouter and Groq give free ones, which is enough for a department to test on chapter text that holds no student information.
What it costs
The chart prices a hypothetical department of 15 teachers who produce about 120 drafts a school day between them (quizzes, rubrics, differentiated passages, family emails) over 20 school days a month. Each call carries about 4,000 tokens in, mostly the pasted source material, and 1,200 out, since a quiz with a key is long. A quarter of the input is a fixed instruction block that can be read from cache. That is about 9.6 million input tokens and 2.9 million output tokens a month.
A hypothetical department: 120 drafts a school day for 20 school days. Monthly cost of 2,400 calls with about 4,000 tokens in and 1,200 out each, 25% of the input read from cache. List prices checked October 20, 2026; your bill may differ.
Output is a large share of this bill because quizzes, rubrics and storyboards are long. Asking for exactly what you will use (15 questions, not “some questions”) cuts cost and review time together. A free tier can cover testing, but keep student information off it. Try your own numbers in the cost calculator.
A sample prompt
A prompt for one passage at three reading levels. The double-brace words are blanks you fill in.
You are helping a {{grade_level}} {{subject}} teacher prepare one
reading passage at three levels for the same lesson.
Source passage (the only source of facts):
"""
{{passage}}
"""
Write three versions:
1. On grade level: keep the length and structure, fix only what
is unclear.
2. Two grade levels below: shorter sentences, same facts, same
key terms.
3. For English learners at {{ell_level}}: short sentences, plus a
glossary of up to 6 key terms with plain definitions.
Rules:
- Keep these terms in every version, spelled the same way:
{{key_terms}}
- Add no facts, examples or numbers that are not in the source.
- Under each version, list any sentence from the source you left
out, and why.
The last rule is the one that saves time: it shows you what the simpler versions dropped, so you can decide whether a struggling reader still gets the idea the lesson is about.
Guardrails
Plain-language guidance, not legal advice. Your school, district or employer sets the rules, and US federal law is only part of them; many states add their own student privacy laws.
Student records stay in approved tools
FERPA protects education records, which include grades, assessments, attendance, discipline and special education files such as IEPs and 504 plans. A school can share them with a vendor without consent only when the vendor meets the conditions of the school official exception, including the school’s direct control over how the data is used, as the Student Privacy Policy Office explains in its guidance for vendors. In practice: use the AI tools your district has approved, and keep student names, ID numbers, grades and accommodations out of personal accounts. Anonymized work (“Student A”) is usually enough for feedback drafting.
Younger students are not your sign-ups to make
COPPA covers online services that collect personal information from children under 13. The FTC’s COPPA FAQ says a school can consent on parents’ behalf only for services used for the school’s educational purpose, not for any other commercial use. That is a decision for the district, not for one teacher signing a class up to a new AI app.
The teacher grades
A model can draft a rubric, suggest feedback wording and point out where an essay misses a criterion. Reading the work and assigning the grade stays with the teacher. Treat any tool that claims to detect AI-written work as a reason to talk with the student, not as proof.
Answer keys get checked
Models produce confident wrong answers, called hallucinations, most often when the source does not quite cover the question. Keep prompts tied to pasted material, ask for a source line beside every answer, and check the key before the quiz goes out.
Workplace training data is personal data
Assessment scores and completion records for employees are personal data too. The risk checklist covers how to handle them in AI tools.
Key takeaways
- Start from your own material: the chapter, the standard or the procedure goes into the prompt, and the prompt adds nothing that is not in it.
- Ask for exact counts and a source line beside every answer, then test for both in the Prompt Tester.
- One shared, tested prompt per course keeps every section consistent.
- Student records go only into district-approved tools; anonymized work is enough for most drafting.
- The model drafts; the teacher reads the work and assigns the grade.
Related templates, guides and posts
Questions
Can I paste student work into an AI tool for feedback suggestions?
Only into a tool your school or district has approved for student data, and even then remove names and anything else that identifies the student. Feedback drafting rarely needs more than the work itself and the rubric. The teacher still reads the work and decides the grade.
Will AI-written quiz questions be accurate?
They are as accurate as the material you paste and the checks you run. Tie the prompt to your source, ask for a source line beside every answer, test the prompt on several chapters in the Prompt Tester, and read the key before students see it.
Do teachers need an API key to use these tools?
Not for the templates, the Grader’s free estimate, the diff, the library or the cost calculator. Running tests in the Prompt Tester or comparing models in Model Compare needs a key. Google AI Studio, OpenRouter and Groq offer free ones, but free tiers may use what you send, so keep student information off them.