AI Chatbot for Online Courses

Before checkout, after purchase, inside the lesson

Course businesses field the same questions at three stages. Before checkout, a prospect wants to know whether the course suits a beginner and what happens if it doesn’t work out. Minutes after purchase, a new student can’t find the login email. Three weeks in, a student is stuck on a concept in module four at eleven at night. An AI chatbot for online courses can take all three, but only if each gets its own rules.

Mixing them is the most common design mistake we see. A sales assistant that also tutors will happily explain your paid material to people who haven’t bought it. A tutor that also sells interrupts a lesson with an upgrade pitch. A support bot that doesn’t know which product a student owns can’t fix an access problem. So we build one assistant with three separate modes, chosen automatically by where it appears — a public sales page or the logged-in course area — and by who is using it: an anonymous visitor, or a student enrolled in a particular tier.

This page is for independent creators, training companies and membership sites that sell courses online. Accredited schools and districts face different obligations; our page on the AI chatbot for schools addresses those.

Course creators anywhere in the country work with EVOTECH IT LLC remotely. Our US-based team sits in the Houston area, the business has run for 20-plus years, Google reviewers rate us 5.0, and every engagement opens with a free phone or video call.

The questions that stall a checkout page

People hesitate on sales pages for predictable reasons, and most of them can be answered from facts you have already written down. A pre-sale assistant typically handles:

• Fit. “I’ve never used spreadsheets — is this too advanced?” The bot answers from the prerequisites you define and says honestly when the course isn’t a match. A refund avoided is worth more than a sale made.
• Time. Hours of video, a suggested weekly pace, and whether there are deadlines or the course is fully self-paced.
• Access. Lifetime access versus a fixed term, whether future updates are included, and whether downloads are allowed.
• Live elements. Cohort start dates, live-call times shown with the time zone spelled out, and whether replays are provided.
• Payment and refunds. The payment-plan options you actually offer, and your refund policy quoted exactly, conditions included.
• Certificates. What the certificate of completion is and what it isn’t. The bot never implies accreditation, continuing-education credit or professional licensure unless you genuinely hold it and have given us the specifics.
• Buying for a team. Group licenses and invoices for employer reimbursement usually go to you or to a booked call.

When a prospect is clearly ready to talk — a company wanting twenty seats, or a question your facts don’t cover — the bot offers a slot from your Calendly or similar booking link, or collects an email for follow-up.

After purchase: the support tickets that repeat every launch

Every course launch produces the same support tickets in roughly the same order. They are rarely hard; they are relentless, and a student who can’t get in during the first hour after paying is a refund request in the making.

• Bought with one email address, trying to log in with another
• Welcome email filtered into spam or a promotions tab
• Purchased the wrong tier, or can’t find a bonus that came with it
• Videos won’t play on a work network or an older browser
• Access paused after a failed payment-plan installment
• Wants the email address on the account changed
• Can’t sign in to the mobile app

For anything the bot can’t fix directly, it gathers what your support person will need so the first human reply can be the answer rather than a list of questions: the email used at checkout, the product name, the order or receipt number, device and browser, and a screenshot of any error message.

Account lookups done safely

With access to your course platform’s API, the assistant can check whether an enrollment exists for the email a student provides and trigger the platform’s own password-reset or login-link message. The safeguard is where that message goes: always to the address on file, never displayed in the chat. The bot also avoids confirming or denying that an account exists to someone who hasn’t shown they own it.

Course creators commonly sell through Teachable, Thinkific, Kajabi, Podia or LearnWorlds, or run LearnDash on a WordPress site, with Stripe or PayPal taking payments and a help desk such as Help Scout, Zendesk or Freshdesk holding tickets. Each platform exposes different capabilities, and some depend on your subscription level, so we check your account before promising any automation. The mechanics of those connections are described under AI integration.

A tutor that answers from your lessons, and says so when it can’t

The most valuable mode, and the riskiest, is a tutor that answers questions about your actual material. We build it from your lesson transcripts, slides, workbooks and the FAQ you’ve accumulated from past cohorts. Each answer points back to its source, along the lines of “this is covered in Module 3, Lesson 2, in the pricing worksheet,” so the student returns to your teaching instead of treating the bot as a substitute for it.

Rules that make a course tutor trustworthy:

1. It says when something isn’t in the course. If a question falls outside your material, the tutor says so and offers to pass it to you or the community. Whether it may add general knowledge beyond your lessons is your choice, and if you allow it, that part is clearly labeled.
2. It respects tiers. A student in the basic tier doesn’t receive explanations drawn from premium-tier modules.
3. It won’t do graded work. For quizzes, capstone projects or certification assessments, you decide where the line sits: concept explanations only, hints, or nothing at all.
4. It stays current. When you re-record a module, the index has to be refreshed or the tutor keeps quoting the old version. We build that refresh into your update routine.

One practical trade-off: video-heavy courses need accurate transcripts first. Automatic transcription tends to mangle jargon, brand names and acronyms, so we assemble a glossary of your terms and correct them before indexing.

The side benefit creators tend to value most is the list of unanswered questions. A weekly summary of what students asked that the material didn’t cover — or covered confusingly — is direct evidence of which lessons deserve a re-record.

Refunds, disputes and frustrated students always reach a person

Some conversations cost you far more when a bot mishandles them than they save when it handles them well. These always go to a person:

None of these is resolved by the bot, but each arrives in your help desk already summarized with the student’s order details attached, so your reply can be the first and the last one.

Cart-open weeks and students who study at 2 a.m.

During a launch, questions peak exactly when you’re busiest — hosting webinars, answering comments, watching sales come in. The assistant works from a launch fact sheet you update: cart-open and cart-close dates, bonus deadlines, payment-plan availability and early-bird rules. Deadlines are always given with a time zone; “midnight” on its own generates a surprising number of annoyed emails from students on the other coast or overseas.

When the cart closes, the bot changes jobs. Instead of answering buying questions, it explains when enrollment reopens and adds interested people to a waitlist in Kit, ActiveCampaign or whichever email platform you use.

Evergreen courses follow a different rhythm: students around the world working on their own schedule, often late at night. That is where the tutor and access support earn their keep, because a student who hits a wall at 2 a.m. and gets an answer keeps going, while one who waits until your morning often doesn’t return that week.

If your students gather in a community on Circle, Skool or Discord, the assistant can answer logistics questions there as well. We usually recommend limiting it to posts tagged for it, or ones left unanswered for a while, so it supports peer discussion instead of crowding it out.

Keeping the tutor from giving your course away

A tutor connected to your full course library is, in effect, a search engine for material you charge for. Treat it that way from the first design meeting.

• Keep the public assistant away from lesson content. The bot on your sales page knows your curriculum outline and your policies. It has no access to lesson transcripts at all.
• Explain, don’t reproduce. Inside the course, the tutor gives short explanations with citations and limits how much of any lesson it will quote word for word. A request like “paste the full transcript of lesson five” is declined.
• Tie usage to logged-in accounts, with per-account rate limits. Unusual patterns, such as hundreds of questions an hour, are flagged — which also helps you spot shared accounts.
• Collect little. The tutor needs to know who a student is and what they’re enrolled in, not their employer, age or phone number.
• Courses for children are a separate conversation. US federal rules restrict collecting personal information online from children under 13. If your audience includes kids, talk to counsel before any chatbot collects anything.
• Accessibility applies to the widget too. The chat window itself must work with a keyboard and a screen reader, and accessibility requests always go to a person.

What EVOTECH does, in order

1. Free consultation by phone or video: what you sell, where students get stuck, and which platform, payment processor and help desk you use.
2. Read your support history. We sort a recent batch of tickets or emails by type. A handful of issues usually account for most of the volume, and those go first.
3. Confirm the platform’s limits. We check what your course platform’s API and plan actually permit before designing any automation around it.
4. Write the policy sheet: refund terms, access rules, certificate wording, launch dates and the exact claims the bot is allowed to repeat.
5. Prepare the content: transcripts, glossary corrections and a module map for the tutor, if you want one.
6. Beta with real students. A small group of current students uses it first, and we tune answers from what they actually ask.
7. Launch in stages — usually support first, then the tutor, then the sales-page assistant — with a weekly report of unanswered questions.

Every project is quoted in writing with a fixed scope after the consultation. Background on our chatbot work is on the AI chatbots page. If you need software that acts beyond conversation, such as an agent that processes enrollment exceptions or builds a custom student dashboard, see AI agents and custom software.

Related services

Frequently asked questions