AI slide generators for teachers: what to check first
Before you pick an AI slide generator for teaching, check seven things: whether it cites the sources it used, where your files live, what file format comes out, whether it writes speaker notes, whether you can edit the result as text, what happens to the material you upload, and what it costs to leave. A tool that fails the first two is a demo, not a teaching tool.
This guide is a checklist, not a ranking. Vendors change plans and features monthly, and any comparison table is stale by the time it is read. The questions below stay the same, and you can answer them for any tool in fifteen minutes with one real chapter and one real paper.
Why is "make slides from a prompt" the wrong test?
Every generator produces a plausible ten-slide deck from a prompt like "introduction to operating systems". That is not what a lecture needs. A lecture is a claim-by-claim path through material the students are supposed to have read, and each slide has to survive the question "where is this from?". The right test is: give the tool the two readings for one session and see whether the deck it writes is traceable to them.
The seven checks
1. Does it cite what it used?
Upload a chapter and a paper, ask for a session outline, and look at the slides. Is each claim tied to a source and, ideally, a location (a section, a page, a figure)? If the deck reads well but points nowhere, you will have to reverse-engineer every slide before you can defend it in class. A tool that names the source on the slide or in the note saves that work.
2. Where do the files live?
Ask where the deck is stored and how you get it out. If the answer is "in your account", your course now lives on someone else's server, under someone else's export button. Prefer a tool that writes plain files into a folder you own, so a backup is a copy and a colleague can open it without an account.
3. What format comes out?
PowerPoint and Google Slides are fine as a final export. As a working format they are opaque: hard to diff, hard to search, hard to regenerate. Markdown-based slides (Marp, for example) are text: a change from last year is a line in a diff, and a search across the whole course is grep. Ask what the native format is, and whether PDF and PPTX are one command away.
4. Does it write speaker notes from the same sources?
A deck without notes is half the job. Check whether the tool writes notes that say what is not on the slide and name where it came from, rather than paraphrasing the slide text. The test is the same as for the deck: give it the sources and read what it wrote against them.
5. Can you edit the result as text?
You will change the deck. If the tool's editor is a drag-and-drop canvas, small edits are slow and bulk edits are impossible. If the deck is text, you can rename a term across forty slides in one find-and-replace, and you can let an assistant rewrite one slide without touching the rest.
6. What happens to the material you upload?
Chapters from a textbook, unpublished notes and student data all end up in the upload box. Read the data terms: is the material used for training, is it retained, who can see it? For tools that call a model provider on your behalf, ask which provider, and whether you can bring your own key so the data path is yours to choose.
7. What does it cost to leave?
Take a deck you made and try to reproduce it elsewhere. If that means re-uploading sources and rewriting from scratch, the tool has locked you in, whatever the pricing page says. If it means copying a folder, you are free. "Free during a beta" is fine as long as the files are yours when the beta ends.
A fifteen-minute test you can run today
- Pick one session of a course you already teach, with its two or three readings.
- Upload the readings, ask for an outline for that session, then a deck.
- Read the deck against the readings. Mark every slide that you cannot trace.
- Ask for speaker notes. Read them against the readings the same way.
- Export. Open the export in the tool you will present with.
- Find the files on disk. If you cannot, note that.
Count the untraceable slides. That number, and where the files ended up, tells you more than any feature list.
How Lecture Studio answers the seven
Lecture Studio is a Mac app written around the first two checks: you attach the readings to a course or a unit, the assistant searches and cites them, and the deck is a Marp Markdown file in a folder you choose. Notes, practice questions and an instructor guide come from the same sources. PDF and PPTX are produced by marp-cli from the same file; the app does not export them itself. The assistant runs on Oberik, so the app needs an Oberik project key and macOS 15 or later on Apple silicon. It is free during the beta, and the source is public, so the cost of leaving is a copy of the folder.
FAQ
Is an AI slide generator safe to use with copyrighted textbook chapters?
Check the tool's data terms before uploading. Prefer tools that let you bring your own model key or run locally, so the material takes the path you chose, and keep uploads to what you need for one session.
Do AI slide generators cite sources?
Some do, most do not. Test with real readings: ask for a deck and see whether each slide can be traced back to a place in a source. If not, plan the time to check every slide by hand.
What file format should lecture slides be in?
A text format such as Marp Markdown for working and versioning, exported to PDF or PPTX for presenting and sharing. Text is diffable and searchable; exports are for the projector.
Is Lecture Studio free?
It is free during the beta. It needs an Oberik project key, and Oberik's own pricing applies to the model calls.