Every few weeks a new AI study tool launches with the same promise: learn anything, faster, with less effort. After three years of writing about education technology, I have developed a healthy suspicion of that promise. So this month I did something a bit tedious: I picked seven of the most talked-about AI learning tools and used each one for real coursework — a statistics refresher, a philosophy reading list, and a Spanish vocabulary grind — for thirty days.
The rules were simple. Each tool had to earn its place by measurably helping me understand or remember something. Feel-good productivity did not count. At the end of the month, I kept four.
How I scored the tools
Three criteria, weighted equally. First, learning gain: could I explain the material a week later without notes? Second, friction: how many steps between “I have a question” and “I got a useful answer”? Third, honesty: did the tool admit uncertainty, or did it confidently invent things?
The four that survived
NotebookLM became my reading companion. Uploading a stack of PDFs and asking questions grounded strictly in those sources turns out to be the killer feature — hallucinations almost disappear when the model can only cite your own material. The audio overview feature, which generates a podcast-style discussion of your documents, is a genuinely new way to preview dense material before reading it properly.
Anki with AI-generated cards solved my oldest problem: making flashcards is more work than reviewing them. I now paste a chapter into a local language model, ask for cloze deletions in a strict CSV format, and import the result. Card quality is about 80% of what I would write by hand, at maybe 5% of the effort. For memorization-heavy subjects that trade is a steal.
Perplexity replaced Google for factual questions during study sessions. The cited answers keep me honest — when every claim links to a source, I notice quickly when a search goes sideways. It is not a tutor, but it is the fastest reference desk I have ever used.
Notion AI earned its place as the glue. My course notes live there anyway, and the ability to summarize a messy page of lecture notes into a clean outline, without switching apps, keeps me actually doing it.
The three that did not
Khanmigo is charming and pedagogically careful — it coaches instead of answering — but it is built for K-12 curricula. As an adult learner outside that world, I ran out of things to ask it within two weeks. If you are a parent, your mileage will differ.
Quizlet’s AI tutor felt like a paywall with a chatbot attached. The free tier teased features it would not let me finish, and the answers were shallower than what I get from a general-purpose model with a good prompt.
RemNote is not bad — it is just too much. It wants to be my note-taking system, my flashcard system, and my knowledge graph all at once. The setup cost ate the week I had budgeted for learning statistics.
The uncomfortable takeaway
The tools that survived are not “AI tutors.” They are ordinary, excellent tools — notes, flashcards, search — with AI thoughtfully embedded. The dedicated AI tutors mostly re-skin a chatbot and add a subscription. Save your money: a general model with the right prompt beats most of them (more on that in my next post on prompt-based learning methods).
Where to start
If you only adopt one, make it NotebookLM for anything involving reading, or AI-generated Anki cards for anything involving memorization. Both are free or nearly free, and both respect your time. Start with a single course or book, run it for two weeks, and judge by one metric only: can you still explain it on Friday?
Read next: The Feynman Prompt: How I Use ChatGPT to Learn Twice as Fast (5 Templates Inside)