What Is an AI Tutor? How Course-Grounded AI Tutoring Works
An AI tutor answers questions about one specific course in real time, by video call, voice call or chat. Here is how course-grounded tutoring works.
Last updated 8 min read
Key takeaways
A book is read start to finish, at the reader's own pace, with no checkpoint beyond a chapter break and nobody checking whether any of it actually landed. A course adds three things a book does not have by default: a syllabus — an explicit map of what is covered and in what order — lessons with a defined scope rather than a continuous narrative, and a way to check understanding, such as a question or a test, rather than trusting that reading equals learning.
None of that exists in a manuscript automatically; it has to be built on top of it. That is true whether you build it by hand over several weeks or generate a first draft of it from the file in minutes — the three additions are the same either way, and a reader can judge a course the same way they would judge a book: by opening it. The open catalogue lets anyone read a syllabus before they commit anything to it, on Cavua or anywhere else that publishes one.
Before touching production — recording anything, writing final prose, generating a single lesson — decide three things using nothing but the manuscript's own table of contents. First, the teaching order: a book's narrative order and a course's teaching order are not always the same thing, and it is worth noting anywhere they should differ, and why. Second, the lesson breakdown within each chapter, small enough that one lesson covers one idea rather than three. Third, what a reader must be able to do or answer at the end of each chapter to show they actually understood it, not just reached the end of it.
This ordering work is the part most often skipped, because a manuscript already looks structured — it has chapters, headings, maybe even numbered sections. That structure was built for reading, though, not for teaching, and the two pull in different directions often enough that it is worth deciding on purpose rather than inheriting it by default.
In practice, turning a manuscript into a course — with or without an automated tool doing the first pass — comes down to the same sequence:
Tools in this category generally do three things. They read the document and infer a structure from it — headings, chapter breaks, a table of contents where one exists. They draft a syllabus and a lesson breakdown from that structure. And they draft lesson text and test questions from your actual material, rather than writing a generic explanation of the topic from scratch.
What none of them remove is the requirement to check the result against your own book. A drafting tool works from what is actually on the page, and any structuring choice it makes — where a chapter splits, what counts as the key point of a section — is a reasonable thing for you to question and override. The quality of the source document matters too: a clean, text-based PDF gives a tool far more to work with than a scanned image of a page, where even reading the words correctly is the first problem to solve.
On Cavua, an author uploads a book or a PDF and the structuring work above runs in four visible phases: reading the document, designing the syllabus, writing the lessons, and finishing up. A progress bar shows which of the four it is in, rather than leaving you with an indeterminate spinner. What comes back is chapters in a defensible order, lessons written from your own material, and tests for the chapters and the lessons — all of it editable. Nothing reaches a reader until you have gone through it and approved it.
Every course you publish also carries the same tutor your readers get: a 24/7 tutor, answering out of your book rather than someone else's. Producing a course spends upload tokens, which you buy, so the cost tracks what you actually publish; an author subscription lowers the rate you pay per upload, and publishing pays royalties plus referral income on two separate streams — authors you bring to the platform, and students you bring. The current rates are on the pricing page, and the full picture of how the author side works is on the for authors page.
A few habits distinguish a course a reader will actually finish from a manuscript with some questions stapled to the end of it. The tests should map to what the lesson said it would teach — not trivia about whether you read it, but a check on the specific idea the lesson was built around. Lessons should stay short enough that someone can plausibly finish one in a single sitting, rather than matching whatever length the original chapter happened to be. And there should be a real answer to “what happens when a reader gets stuck on a paragraph” — which is exactly where a tutor bound to the material, rather than a static page, earns its place in the course.
An AI tutor answers questions about one specific course in real time, by video call, voice call or chat. Here is how course-grounded tutoring works.
Classic books are often dense, old or awkwardly translated. Here is how to read Plato, Kant, Homer or Shakespeare closely, and where an AI tutor helps.
No. A course draws its lessons from your existing material — you are restructuring it and adding checks for understanding, not rewriting the substance. You should still read and approve anything a tool drafts from it before it reaches a reader, since a drafting choice is always worth a second look.
It depends on the length and structure of the manuscript, and how much you want to change before publishing — there is no honest single number to give you. What a tool like Cavua's can tell you concretely is which of its four phases a specific upload is in at any moment — reading the document, designing the syllabus, writing the lessons, finishing up — rather than leaving you with an indeterminate wait.
You should be able to, and on Cavua you edit and approve every chapter, lesson and test before anything reaches a student. Nothing about an automated first pass should be treated as final.
Cavua reads a book or a PDF directly. Whatever the format, a clean, text-based file matters more than which one you use — a scanned image of a page gives any tool far less to work with than an export with real, selectable text.
Royalties on what the course earns, plus referral income on two separate streams — authors you bring to the platform, and students you bring — each priced independently. Producing the course costs upload tokens, which you buy, and an author subscription lowers the rate paid per upload.
Upload it and see what four phases of structuring do to a table of contents you already know by heart.