How to add an AI tutor to your online course

Three ways to add an AI tutor to an online course

Most course creators who want to add an AI tutor to an online course are trying to fix one moment: the student hits a concept they cannot get past, has nobody to ask, and quietly stops. The fix you choose depends on what that moment needs. Sometimes it is a quick answer. Sometimes it is someone who explains the idea again, asks a question and waits for the answer.

You have three realistic options:

• The AI assistant your course platform already offers, a chat box inside the course player that answers from your lessons.
• A custom assistant you build yourself with a general AI product (a custom GPT, a Claude Project or a Gemini Gem), loaded with your material.
• A dedicated teaching tutor that turns each chapter into a lesson it teaches one to one, with checks, hints and a report.

The first two are answering tools: they respond to whatever the student types. The third is a teaching tool: it has a plan and asks the student questions. Both kinds are useful. They solve different problems, and they cost and report differently.

Option 1: the AI assistant your course platform may already offer

Start here, because it needs no migration and lives where your students already are. What exists depends on the platform and the plan. As of September 2026, according to each vendor’s own pages:

• Thinkific offers Thinker, an AI teaching assistant that appears as a button in the course player and gives “helpful, sourced responses” drawn from your published courses, plus PDFs you upload as background reference. After a quiz, a Post-Quiz Review mode explains missed questions and links back to the relevant lesson. It is available on Thinkific Plus plans, with a monthly allotment of credits (one credit per answer). Thinkific notes that only the text of uploaded PDFs is indexed; diagrams and photos in them are ignored.
• Kajabi offers a Teaching Assistant, one of its Expert Agents, that answers member questions from lessons, video transcripts, community posts and downloads, cites the lesson and points to a timestamp. Kajabi’s help center says it “only answers from your published materials,” that conversations land in your inbox, and that agents are priced per agent, per month, with 1,000 messages a month included.
• Teachable’s AI Hub lists creator-side tools: a course starter, a quiz generator and video transcription with translations. It does not list a student-facing tutor, so check with Teachable if that has changed.

Where this option is strong: setup is short, answers are grounded in your lessons, and nothing changes for the student. Where it stops: it is still a chat box. It helps the student who knows what to ask. The student who is confused but cannot name the confusion, often the one about to quit, gets nothing unless they start typing. Thinkific’s post-quiz review is a step toward teaching, but the student still has to choose to open it.

Option 2: a custom GPT, Claude Project or Gemini Gem

For a while, the standard advice was to “make a custom GPT for your course.” That advice has aged badly. Here is what the vendors’ own help pages say in September 2026:

• ChatGPT GPTs. OpenAI says personal ChatGPT accounts (Free, Go, Plus and Pro) “cannot create or publish new GPTs,” and that it is “planning to retire custom GPTs” in favor of plugins, with retirement planned for affected Enterprise workspaces on December 11, 2026 and other plans expected to follow the same timeline. Students must be signed in to ChatGPT to chat with a GPT, each conversation starts fresh, and GPT builders “cannot view” individual conversations.
• Claude Projects. Anthropic’s help center says project sharing is available on Team and Enterprise plans and shares the project with members of your organization. In practice, every student would need a seat in your workspace.
• Gemini Gems. Gems can be shared much like a Google Drive file, but Google’s help page warns that “any Gem instructions and files that you have uploaded to the Gem can be viewed by any user with access to the Gem.” Sharing the Gem means sharing your course files.

Put together, the trade-offs are clear. Your content can leak. You see nothing of what students ask or where they struggle. The assistant can mix your material with general knowledge unless you tightly restrict it. And the product may be retired or re-priced under you. A custom assistant is still a fine private study aid for you, or a stopgap for a small group of adults who already pay for the tool. It is a weak base for a paid product.

Option 3: a dedicated teaching tutor

A teaching tutor starts from the other end. Instead of waiting for questions, it takes each chapter and teaches it: explains one idea at a time, asks a question after each, grades the answer, gives a hint when the student is wrong, and comes back to weak concepts before moving on. The student can still interrupt with a doubt.

There is decent evidence that structure and grounding matter. In a 2025 randomized crossover trial in Scientific Reports, Kestin, Miller, Klales, Milbourne and Ponti had 194 Harvard introductory physics students learn one lesson from an AI tutor built on the course’s teaching approach and another in an active-learning class. Median learning gains with the AI tutor were more than double, with a median of 49 minutes on task against a 60-minute class. The authors kept the tutor accurate by writing step-by-step solutions into its prompts. That is two lessons in one course at one university, but the design lesson carries over: the tutor worked because it followed prepared material, not because it improvised.

With Tutezy, for example, each chapter of your course is compiled into a teaching plan (topics, concepts, board steps, narration, checks and hints) that you can preview before any student sees it. The lesson is then taught live on a whiteboard, in your cloned voice if you want, by voice or text. The details are on how it works.

How the three options compare

None of these is best in every row. The question is which row matters most for your course. For a deeper look at the chatbot versus tutor difference, see Tutezy vs ChatGPT. For how any of them should stay faithful to your material, read Will an AI tutor make things up?

Which option fits your course?

• Your students mostly ask “where is…” or “what did you mean by…”. A platform assistant is enough. Turn it on, seed it with your FAQ, and read the questions it gets each week. They tell you which lessons to re-record.
• You sell to self-directed adults and do not need data. A platform assistant still beats a custom GPT, because it stays inside your course and does not expose your files.
• Your course is sold on an outcome (a certification, a job skill, an exam), or completion drives your reviews and refunds. You need something that asks the student questions and tells you who is stuck. That is a teaching tutor.
• You want both. Nothing stops you running a platform assistant for logistics and a teaching tutor on the three or four chapters where students struggle most.

If completion is the real problem, pair this with the fixes in How to improve online course completion rates. An AI tutor helps most when units are short and each one ends with something to answer.

A five-step rollout plan

Whichever option you pick, do not switch it on across the whole course at once. Run it like a product test.

1. 1

   Pick the module with the worst drop-off

   Open your course analytics and find the lesson after which the most students stop. That is where a tutor earns its keep, and where you will see a change fastest.
2. 2

   Prepare the material

   Gather that module’s videos, slides, PDFs and quiz questions. Write down the three mistakes students make most often; they become hints and checks. See turning PDFs, slides and videos into interactive lessons.
3. 3

   Set it up and review every answer path

   Configure the assistant or compile the lessons, then go through them as a student would. Ask the awkward questions. Fix anything wrong before a student sees it.
4. 4

   Pilot with about 20 students

   Invite a mix of strong and struggling students from a current cohort. Tell them it is an AI and ask for blunt feedback. Twenty students is enough to find problems, not to prove an effect.
5. 5

   Measure, then decide

   Compare module completion with the previous cohort, then look at minutes used, the concepts students missed, support tickets and refund requests. Expand only if the numbers and the comments both improve.

How to package and price an AI tutor

The simplest model is a premium tier: the same course, sold “with a live AI mentor,” at a higher price. It gives students a clear choice, and it pays for the tutor from the buyers who value it most.

Here is the arithmetic at a per-minute rate of $0.03 (Tutezy’s base rate; $0.05 with an animated avatar shown):

• A student who uses the mentor fully: 8 modules × 20 minutes = 160 minutes × $0.03 = $4.80. With the avatar on throughout, $8.00.
• A heavy user: three times that, 480 minutes, costs $14.40.
• A cohort: 100 buyers take a $399 premium tier instead of the $299 standard one and average 120 minutes each. That is 12,000 minutes, or $360 in usage, against $10,000 in extra revenue.
• One-time costs: a voice clone is $20 per voice, and compiling a typical 40-slide course costs under $2.50.
• Your pilot: 20 students × 30 minutes = 600 minutes = $18.

Two cautions. First, cap session length per course so a single student cannot run up an open-ended bill. Second, name it honestly: “AI mentor trained on this course” is accurate; “personal mentoring” implies a human. The full rate card and more examples are on the pricing page, and other pricing models are compared in How much does an AI tutor cost?