Tuesday, 9 pm. A participant in your accounting course sits over the course notes after work and gets stuck on provisions.
The next classroom session is eleven days away. She asks her question anyway, just to a general chatbot that has never seen your materials. This is how people learn in professional education today: more than two thirds of German chamber of commerce graduates prepare for their exam while working (DIHK, Erfolgsstudie Weiterbildung 2023). Most questions come up between sessions.
An AI tutor is meant to close that gap. Whether it can do so in your course is not something a data sheet will tell you. Even AI systems that answer only from verified sources demonstrably make mistakes: in a Stanford University study, specialised legal research tools gave incorrect or invented answers in 17 to 33 percent of cases (Magesh et al., Journal of Empirical Legal Studies 2025). For a course that prepares people for an exam, that matters. So test it yourself, with your own materials. It takes about 30 minutes.
acemate offers an AI tutor for education providers itself. The checks below apply to every vendor, including us.
Key takeaways
- An AI tutor suits your course if it answers from your materials, names the source and admits when it does not know.
- Six tests with a real chapter show more in 30 minutes than any product demo.
- An LMS is not a tutor: Moodle's AI features first need an AI provider set up by your administrator, and then work only with the page that is open.
- Clarify the rights to course materials before uploading them. Freelance lecturers keep their copyright.
- The EU AI Act has required transparency since August 2026. Stricter rules for AI whose assessments feed into grades or qualifications apply from December 2027.
What an AI tutor should do in a course
An AI tutor is a learning companion based on generative AI. It answers participants' questions, explains content and creates exercises. For education providers, three properties matter: it works with your content, it shows where in your materials each answer comes from, and your lecturers decide how it responds.
It covers the weeks between classroom sessions. It does not replace teaching. In an mmb Institute survey, 84 percent of experts expect learners to keep wanting human lecturers, and 83 percent see the future in a tandem of lecturer and AI (mmb Trendmonitor 2025/26). That is where a tutor should come in.
The 30-minute test: six checks with your own materials
Take a chapter from a running course, ideally one that participants often ask about. Upload it to the vendor's platform and work through the six checks in order.

1. Sources: does it name the passage?
Ask a question whose answer appears only in your materials, for example about a case discussed in class. Then open the cited source. The page should contain exactly the statement the answer relies on. A general reference to "the course notes" is not enough.
2. Gaps: does it admit what it does not know?
Ask about something your chapter does not cover. A good tutor says that your materials contain nothing on it. A poor one answers anyway, fluently and from general model knowledge. This is the most important test, because a plausible but wrong answer is the hardest for participants to spot shortly before an exam.
3. Currency: does it follow your version?
Pick a rule that changed recently and that your materials already explain in its new form, such as an allowance or a deadline. Ask about it. The answer has to follow your materials. If the tutor gives the old value, it is drawing on what the language model learned in training.
4. Practice: does it spot a half-right answer?
Have it create exercises for your chapter and answer one of them deliberately only half right. Good feedback names what is correct, what is missing and where it is covered in your materials. For exam preparation, practice counts for more than asking questions, because in the exam participants have to apply what they know.
5. Control: do your lecturers stay in charge?
Check whether lecturers can set rules, such as "never give the solution to assignments directly", whether topics can be excluded or AI features paused, and whether they can see where the group is struggling. These insights should be aggregated and should not monitor individual participants.
6. Access: how do participants get in?
Walk through a participant's route: from the link in your LMS or the invitation email to the right chapter. Every additional login is a hurdle. What you want is an entry point that fits your existing setup.
"We already have an LMS": what Moodle and ILIAS can do today
This is the most common question, and a fair one. Since version 4.5, Moodle can summarise or explain the page that is open, once an administrator connects an AI provider and enables the features (Moodle documentation: AI subsystem). ILIAS offers AI through extensions. What these features do not do is answer questions across all of a course's materials with page references, create exercises with feedback or show lecturers where the group is struggling. What Moodle and ILIAS can do in detail, and how to connect an AI tutor to your LMS, is covered in AI tutor alongside Moodle or ILIAS: what it adds to your LMS. For your test, one question is enough here: can participants reach the tutor directly from the LMS?
Clarify the rights to your materials before uploading
Many course materials are written by freelance lecturers. Under German copyright law, anyone who is not an employee keeps the copyright in their materials. The rule for employees in Section 43 UrhG does not apply to freelancers. What you may do with the materials depends on the contract, and where it is unclear, only as far as the purpose of the contract requires (Section 31(5) UrhG). In 2014 the Higher Regional Court of Frankfurt ruled that seminar materials can be protected by copyright and that a provider may not simply pass them on to new lecturers (case 11 U 106/13, summary in German).
For an AI tutor, this means two things. Check whether your freelance contracts cover digital use of the materials, and extend them if in doubt. Ask the vendor to confirm in writing that your content is not used to train AI models and that participants cannot download the original files if your lecturers do not want that. For a binding assessment of your contracts, consult a lawyer.
Data protection and the EU AI Act: what applies when
For data protection, three questions cover most of it: is there a data processing agreement under Article 28 GDPR? Are content and usage data processed in the EU? And does the vendor confirm that your content is not used to train models?
Three dates matter under the EU AI Act (EU AI Act Service Desk):
| From | Rule | What it means for you |
|---|---|---|
| 2 February 2025 | AI literacy (Art. 4) | You take measures that support the AI literacy of lecturers and staff, such as an introduction to the tutor. Since 27 July 2026, no specific level is required. |
| 2 August 2026 | Transparency (Art. 50) | Participants must be able to tell that they are talking to an AI. |
| 2 December 2027 | High-risk uses (Annex III point 3) | Applies when AI decides on admission, assesses learning outcomes for grades or qualifications, determines the level of education or monitors exams. |
The wording of Annex III point 3(b) covers AI that evaluates learning outcomes, explicitly including when those outcomes steer the learning process. The European Commission narrows this in its draft guidelines on high-risk classification. High-risk systems are those whose assessments feed into a final decision, such as grades, qualifications or admissions. Adaptive learning systems that adjust difficulty to performance, tutoring systems with personalised feedback and learning analytics for lecturers are not to be classified as high-risk, as long as they support learning on an ongoing basis (draft Commission guidelines, May 2026, paragraphs 222 to 224). A tutor that explains, practises and gives feedback therefore does not fall under the high-risk cases. That changes once its assessments feed into grades, certificates or admissions. The guidelines are still a draft, so ask the vendor to explain in writing how it classifies its system. This assessment is not legal advice. For more on the AI literacy duty and the changes made by the Digital Omnibus, see our article AI in continuing education.
A pilot that shows something
Start with a course that involves a lot of self-study and ends in a clear exam. Decide beforehand how you will measure success, in your own numbers: how many questions are asked between sessions, how many participants still use the tutor after four weeks, how much time your lecturers spend on follow-up questions and how the pass rate compares with the previous year. For reference: in the German chambers' further training exams in 2025, around 69 percent passed (DIHK further training statistics 2025).
Involve the course's lecturers from the start and let them introduce the tutor in class. And agree with the vendor in advance what happens after the pilot: which term, which costs, and whether the pilot counts towards a licence.
These questions help in a first conversation:
- Which sources does the tutor answer from, and how does it show the passage?
- What happens when a question is not covered by our materials?
- Which rules can our lecturers set for each course?
- How do you protect our lecturers' materials against download and reuse?
- Where is data processed, and is there a data processing agreement?
- How do you classify your system under the EU AI Act?
- How do our participants get from the LMS to the tutor?
- Who sets up the first course and trains our lecturers?
Where acemate stands
We offer an AI tutor for education providers ourselves and have written the checks above so that you can apply them to any vendor, including us. Our AI tutor answers from your materials with document and page, your lecturers set rules and see in the analytics where the group needs support. Personal learning paths show each learner per topic what is already secure, and exercises and review follow their individual level. The platform can be your LMS, with course management, roles and SCORM, or fit into the LMS you already have through single sign-on and direct links, under your own brand if you wish (white label). Data is processed in the EU, there is a data processing agreement, and your content is not used to train AI models (AI control and data protection).
The DIHK Bildungs-gGmbH runs the platform as IHK AILA nationwide for all 79 German chambers of commerce, and the Studienwerk der Steuerberater is testing it in three courses alongside live teaching.
If you would like to run the test on one of your courses, we will set up access with your own materials. An overview and the option to book a call are on the page for academies and professional education.



