AI tools for teaching preparation
A companion working guide for Learning with AI: A Framework for Students, Instructors, and Universities. See Part II, “AI tools for teaching preparation.”
The book names the durable categories and leaves the practical detail here, where it can be kept current. Products change names, merge, and disappear. The categories persist, and so does the way you use them well. This note is meant to be updated as the tools change, which they will. Tool names, free-tier limits, and which tools lead all reflect mid-2026; check each tool’s own site before you rely on it.
Set the context before you ask. Before you ask for anything, tell the AI who you are, what the course is, and who the students are. Paste the syllabus, or the unit you are working on. Describe the class in general terms (a first-year non-majors section, mixed preparation, most without calculus) rather than uploading anything that identifies an individual student; that line is drawn in Before you upload anything, below. The same request returns generic filler on its own and something close to usable once the tool knows the setting. This single step does more for the quality of what you get back than any other.
Show it what good looks like, and push past the first draft. The tool’s first answer is a starting point, not the finished product. Paste an example of the quality or style you want and ask it to match that, not just your description of it. Then react to what comes back: ask for a stronger version, ask what it would change, ask for two or three options and choose. The output improves faster when you push on a draft than when you re-prompt from scratch.
The one rule that governs all of it. Every capability below produces a draft. The instructor who uses it is responsible for verifying the result, the same standard the book asks of students. None of these tools knows your course, your students, or what you meant to teach. They remove the friction of a first draft. They do not remove your judgment or your responsibility for what you hand out.
Which kind of tool to reach for. Two families cover most of this work. In 2026, the general assistants (ChatGPT, Claude, Google Gemini) are the workhorses. They handle almost every task below, they have free versions, and a clear prompt with your own material attached gets further than any menu of buttons. The teacher-specific tools (Diffit, MagicSchool, and others) wrap those same models in a fixed workflow, which saves setup time for a narrow task. Much of that ecosystem was built for K-12; a university instructor will often get more from a general assistant given good instructions. Where a specialist tool earns its place, it is named below.
One boundary. These uses share a low-risk profile because they act on your own materials, not on student assessment. Using AI to give feedback on student work or to grade is different in kind and carries real risk. The book treats those with the boundaries they require in the feedback-and-grading section. They are not repeated here.
Before you upload anything. Student work and student data are governed by your institution’s policy and by law (FERPA in the United States). Check what you are permitted to upload before you paste a roster, a graded exam, or anything that identifies a student into an external tool. Your own lecture notes and reading lists carry no such restriction.
Rewrite a reading at several levels
Good for. Meeting a mixed-preparation class where each student can reach an idea, and producing a plain-language version of a dense passage.
Using it well. Give the tool the actual passage and ask for two or three versions at named levels (“a version for a first-year non-major,” “a version that keeps the technical precision”). Read the simplest version against the original with one question in mind: did it drop a qualification that matters? Simplification smooths away exactly the caveat a careful reader needs, and it does this most often with technical or mathematical claims. A good result keeps the meaning intact at every level and changes only the reach, not the truth.
Getting the tools. Any general assistant does this from a pasted passage. Diffit is a specialist built for it; it rewrites a text or a web page at several reading levels and across languages, with a free tier.
Turn an outline into draft materials
Good for. Escaping the blank page: a slide sequence, a lesson plan, a set of worked examples, or a problem set with solutions.
Using it well. Feed it your outline and your emphasis, and it will draft the structure in seconds. The work that remains is the work that makes the material yours. Check every worked example and every solution by hand; generated mathematics is confident and often wrong. Ask whether the emphasis matches what you want students to carry away, because generated material drifts toward the generic. A good result is a scaffold you can teach from after your edits, not a finished handout you trust on sight.
Getting the tools. A general assistant drafts lesson plans, worked examples, and problem sets from your outline. For slides, Gamma builds a designed deck from a prompt or a document and exports to PowerPoint or PDF on a free tier; Google Slides with Gemini and Copilot in PowerPoint do the same inside those suites. Expect to fix the exported file; the design rarely survives the export cleanly.
Draft assessment questions and rubrics
Good for. Candidate quiz or exam questions drawn from your content, and a first-pass rubric for an assignment you describe.
Using it well. Give it your material and ask for more questions than you need, then cut. Check that each question tests the reasoning you care about and cannot be answered by pattern-matching or a quick search. For a rubric, check that the criteria reward the thinking, not the surface features that are easy to score. Treat the output as raw material for an instrument you build, not the instrument. A good result is a short list of questions you would actually put on the exam and a rubric a second reader would grade consistently. (Grading student work with AI is the different, higher-risk activity covered in the book’s grading section.)
Getting the tools. A general assistant does this well from your content. MagicSchool and similar teacher suites offer fixed quiz and rubric generators with free tiers if you want the workflow rather than the flexibility.
Convert materials for accessibility
Good for. An audio version of a reading for students who learn by listening or who have print disabilities, described alternatives for figures, and captions for recorded material.
Using it well. Generate the conversion, then check it where a small error changes the meaning: technical terms, equations, symbols, and anything a screen reader will voice literally. A described figure has to carry the information the figure carries, not just say that a figure is present. A good result is a version a student could rely on without seeing the original.
Getting the tools. NotebookLM turns your uploaded sources into an audio overview for free. General assistants draft figure descriptions and clean up auto-generated captions. Your institution’s disability-services office often has vetted tools and standards; ask before building your own.
Draft routine communication
Good for. A syllabus announcement, an FAQ built from recurring questions, a first pass at administrative correspondence.
Using it well. These are logistics, and a general draft handles logistics well. The line to hold: anything carrying a judgment about a specific student must be written or substantially rewritten by you. A recommendation letter or a comment on a student’s standing is your assessment of a person, and a generic draft cannot make it. A good result reads in your voice and says nothing about a student you did not decide to say.
Getting the tools. Any general assistant drafts this from a short description of what you need to say.
Turn a recorded lecture into notes
Good for. A searchable transcript and a summary of a session, for a student who missed it or as a study resource.
Using it well. Transcribe first, then summarize from the transcript. Check the transcript on technical vocabulary, where these tools fail most, and check that the summary kept the emphasis that made the session worth attending. Automatic summaries flatten what mattered. A good result is a transcript a student can search and a summary that would not mislead someone who was not there.
Getting the tools. Otter.ai captures a live lecture and produces a transcript on a free tier, though it handles STEM terms and strong accents poorly, so read its output for your own field with care. NotebookLM does not record, but it turns a transcript or a posted recording into a source-grounded study guide for free, and it stays inside the material rather than inventing.
Gather sources for course design
Good for. A survey of a topic or a set of candidate readings when you are building or updating a course.
Using it well. Ask for the survey, then verify every source before it goes near students. This is the capability most prone to fabricated citations and confident misdescription. Check that each reference is real and says what the summary claims, exactly as the book asks students to do with their own AI-sourced work. A good result is a reading list you have personally confirmed, not a list you trust because it looked plausible.
Getting the tools. General assistants with a current search or research mode do this; the verification burden is the same whichever you use.
Why this is a companion, not a book chapter
A printed list of tools would be wrong within a year, and the book’s argument is about judgment, not tools. This guide can be concrete where the book stays durable, and it can be revised. If you adapt it, or find a category worth adding, that is the intended use. Treat every AI output as a draft to verify, and follow your institution’s policy on data privacy and permitted use before you upload student work or student data to any tool.