Pre-Course Questions: AI Chat for Course Providers

AI assistant for course providers: what it answers from your course data before the start, where it hands over to the office and what stays in-house. Read on.

In plain language

Many people in work take further training. Most of it is provided by employers. So questions about costs and time off belong in the knowledge base. Before signing up, people ask about dates, open places, price, funding, requirements and format. An assistant answers this from your course data. That takes load off the office. On withdrawal it only states the rules and is not meant to judge the case. Many want a person when there is a problem, and the assistant shortens the way there.

September 25, 2026 14 min read By XICBOT

Key points

Participation in continuing education among people in work stood at 58 percent in 2022; over the four weeks before the survey, 5.8 percent of those in work aged 25 to 64 took part in 2025 (Federal Statistical Office). Both are correct and measure different things.
77 percent of vocational continuing education in 2022 was employer-provided training (Federal Statistical Office 2022). Questions about who pays and who grants time off are therefore not a side issue; they belong in the knowledge base as their own building block.
Six questions cover most enrolment conversations: date, remaining places, price, funding, prerequisites and format. Making them answerable relieves the office more noticeably than any extra page in the printed programme.
The right of withdrawal under Section 4 FernUSG applies only to distance-learning contracts under Section 3(2); contracts concluded online fall under the distance-selling rules of the Civil Code (Section 312g). The assistant states the position, not the individual case.
In a Bitkom survey, 62 percent of online shoppers wanted a quickly reachable human contact when something went wrong (Bitkom). The assistant shortens the path to that person; it does not replace the conversation.

Why everything piles up in the weeks before the course starts

Education work has a rhythm other industries do not share. A term programme is planned months ahead, then printed or published, and then a great deal happens at once over four to six weeks. Adult education centres, academies, language schools, driving schools and chamber-affiliated providers all see the same curve: between the programme appearing and the first day of class lies the window in which people decide. Anyone who gets no answer inside that window often does not decide at all – not against the provider, but against the effort.

Official statistics show how large the base population is. The share of people in work who had taken part in vocational continuing education over the preceding twelve months stood at 58 percent in 2022 (Federal Statistical Office 2022). Measured instead over a four-week window, it was 5.8 percent of people in work aged 25 to 64 in 2025 (Federal Statistical Office), and 10.4 percent among people in academic occupations (Federal Statistical Office). The annual figure describes how many take part at all. The monthly figure describes how thin the same group becomes once you narrow it to a single week. For the website that means: the people searching right now are a minority with a very tight window.

In practical terms, the pile-up before the course start cannot be solved with more staff. An extra person for six weeks a year is hard to find and hard to train. What can be solved is the share of enquiries answerable from existing data. That is exactly where a lead assistant comes in: it takes the question, answers it where a substantiated answer exists in-house, and otherwise collects the details the office needs for a single, conclusive reply.

The six questions that come before almost every enrolment

Anyone reviewing a term's worth of conversation logs finds surprisingly little variance. The wording differs, the concerns barely do. Six questions carry most of the volume, and they usually arrive in the same order – first date and place, then money, then eligibility, then format.

When does the next course start? It sounds trivial and is not: what is wanted is the next date that fits the location, the format and the level of prior knowledge. A list of all dates does not answer the question, it shifts the work onto the person asking.
Are there places left? The only one of the six whose answer can change by the hour. An answer taken from the printed programme is worthless here; it has to come from the booking system or not be given at all.
What does taking part cost? Usually tied to questions about reductions, instalments and material costs. A price without its conditions produces exactly the call-back it was meant to save.
Will my employer contribute? 77 percent of vocational continuing education in 2022 was employer-provided training (Federal Statistical Office 2022). Questions about cost coverage, time off and the invoice address are therefore the norm, not the exception.
Do I need prior knowledge? In language, IT and advancement courses this is the most common drop-off point. People who find the prerequisites only in the small print would rather not book than book wrongly.
In person or online? Since the changes of recent years, a decision in its own right rather than an afterthought. It covers venue, technology, time zone and whether switching between the two routes is possible.

These six questions are the work list for the knowledge base. They are also the test bench: an assistant that answers all six cleanly and hands over honestly on the seventh is more useful day to day than one that has something to say about everything. What the underlying content on the website has to look like before it can be quoted at all is described in the article on the website assistant.

Withdrawal, approval and what the law settles

Several legal questions are tied together when courses are booked through a website, and none of them belongs in a freely formulated answer. Distance-learning contracts have their own statutory regime. The German Distance Learning Protection Act points to the Civil Code for the right of withdrawal:

Under a distance-learning contract pursuant to Section 3(2), the participant has a right of withdrawal under Section 355 of the German Civil Code. (Own translation of the German original.)

Section 4 sentence 1, German Distance Learning Protection Act

For in-person courses booked online, the general distance-selling framework of the Civil Code applies – with one exception that regularly surprises course providers. Contracts excluded from the right of withdrawal include those for "the provision of further services in connection with leisure activities, where the contract provides for a specific date or period of performance" (Section 312g(2) no. 9 of the German Civil Code, own translation). Whether a particular course falls under that is a question for the individual case, and not one an assistant may decide. It may state the rule, name the source and hand over to administration. Where contact details are collected in the chat, consent has to be logged cleanly – how that works in dialogue is set out in the article on consent in chat.

WarningDistance-learning courses generally require approval

"Distance learning courses require approval." (Section 12(1) sentence 1 of the German Distance Learning Protection Act, own translation). Courses that by content and purpose serve solely leisure or entertainment purposes need no approval; their distribution merely has to be notified to the competent authority (Section 12(1) sentences 3 and 4). For a pure hobby and leisure programme, notification is therefore usually the applicable route, and approval the route for occupation-related courses; where the line runs in an individual case is for the authority to decide, not the assistant. Anyone offering their programme as a distance-learning course subject to approval operates inside a regulated framework, and the assistant may neither assert nor dispute that status. The knowledge base therefore holds either the verified statement including the approval number – or nothing at all. An invented or sloppily copied approval detail is a statement about your own legal standing and thus the most expensive sentence an automated conversation can produce.

Funding, cost coverage and who actually pays

In continuing education, the money question is rarely a pure price question. It is a question of who pays, who grants time off and who needs which piece of evidence. Since 77 percent of vocational continuing education in 2022 was employer-provided training (Federal Statistical Office 2022), the route to enrolment runs through a third party in many cases – the HR department, the employer, a chamber or a funding body. An assistant that only quotes the course fee answers the smallest of those questions.

Employer bookings

A company booking needs an invoice address, a cost centre and a confirmation of attendance. The assistant recognises this route from the question about invoicing and keeps it separate from the private route.
Price and extras

Course fee, examination fee, materials and reductions belong in one record. Fields maintained separately produce answers that are individually correct and jointly wrong.
Public funding

Federal, state and agency programmes change independently of the term rhythm. The assistant states the applicable rule with its date and points to the responsible body instead of holding out the prospect of approval.
Documents

Enrolment confirmation, invoice, attendance record and certificate are four different documents with four different points in time. Distinguishing them cleanly in chat saves the longest follow-up threads of the term.
Application deadlines

Funding applications have their own deadlines, which can fall before the enrolment deadline. That order is the most common cause of later cancellations and belongs explicitly in the answer.
Qualification and effect

A confirmation of attendance, a certificate and a regulated qualification are not the same thing. The assistant states what your own course awards and is not meant to judge what that qualification is worth on the labour market.

The difference between a good and a dangerous answer comes down to a single word here. "This course is eligible for funding" is a statement about a programme. "You will receive the funding" is a promise about an individual case the provider does not decide. That distinction has to sit in the assistant's instructions, not in the hope that the model will find it on its own. How tightly product and offer data need to be structured for that is shown in the article on product data your assistant can answer from.

Checking prerequisites before anyone books

The second most common cause of cancellations after enrolment is a prerequisite nobody checked beforehand. Language level, prior education, professional practice, technical equipment, valid documents: what is self-evident to the administration sits in the small print for the person enrolling. An assistant can run that check as a short dialogue, provided the criteria are stored in the knowledge base as testable conditions rather than as running text. How such a knowledge base is built in the first place is described in the overview of the knowledge base.

First the hard condition: is there a formal access requirement without which participation is ruled out? That question comes first because it is allowed to end the conversation.
Then the level: in language and IT courses, the level decides which course is the right one. The assistant guides people to an assessment instead of certifying a level itself.
Then the practice: work experience, training records and recognitions are evidence, not self-assessments. The assistant asks for the document, not for the feeling.
Then the equipment: for online formats, camera, microphone, bandwidth and a second screen are not side issues. Vagueness here produces drop-outs in the first session.
Finally the borderline case: where two conditions contradict each other or a detail stays unclear, the automated part ends. What follows is a handover with the full transcript, not a well-meaning estimate.

In person, online or hybrid: the format question

The format question is often treated as a detail and is in truth a decision axis of its own. It depends on travel time, caring duties, working-time models and technical equipment – things the provider does not know and should not guess. What it does have to know is its own data situation: which detail about which format comes from which source.

Format	What people ask beforehand	Where the answer comes from
In person	Venue, room, travel, parking, step-free access to the building, start time on day one	Location record and course description, maintained and versioned separately per location
Online	Technical requirements, time zone, whether sessions are recorded, how access is delivered	Participation notes on the course record, one field per statement instead of one catch-all paragraph
Hybrid	Whether switching between the two routes is allowed, until when, and whether it changes the fee	Terms of participation, linked to the enrolment status and to the cancellation rule

In regions with an international audience there is one more axis: the language of the answer. Anyone teaching in several languages should be able to give information in several languages, and from the same data – otherwise the versions drift apart and the enrolment terms say different things in two languages. How that works without separate editorial tracks is described on the page about the multilingual assistant.

Dates and remaining places from your own system

Of the six questions, exactly one cannot be answered from text: the one about the free place. Its value changes while the conversation is running. An assistant answering that from a knowledge base regularly says something untrue – not because the model errs, but because the text is from yesterday. The clean solution is a tool call against the booking system, connected for this as your own interface: the assistant queries availability, receives a number and passes it on with a timestamp. How such calls are connected and bounded is described on the page about tool control.

The arithmetic behind it

A course with twelve places, three dates and two locations has six possible combinations people can ask about. With eighty courses in a term programme, that is several hundred states, changing daily, that nobody maintains by hand. Which is precisely why the availability question belongs to the system and not to the text. How a request is taken cleanly in dialogue and handed to your institution is shown in the article on appointment requests via chat.

The knowledge base of a course provider

A knowledge base for a course provider consists of three layers that differ in shelf life. The first layer is stable: house rules, enrolment procedure, payment routes, cancellation rules, step-free access to the rooms. It changes rarely and may be phrased broadly. The second layer changes with the term: course descriptions, prices, tutors, dates, prerequisites. It has to be re-imported in full at every programme change, not topped up. The third layer is live: remaining places, cancellations, short-notice room changes. It does not belong in the knowledge base at all but in a tool call.

The most common mistake sits on the border between layers one and two. If the cancellation rule lives in the running text of a course description, it is deleted with the course and then missing everywhere. If the price lives on the general page, it survives the change of term and is then wrong. Both mistakes produce answers that sound plausible and are no longer true. Drawing that border cleanly once saves a week of rework every term; what such a maintenance process looks like is described in the article on keeping the knowledge base up to date.

For context on how widespread these systems have become: 41 percent of companies with 20 or more employees use AI, and a further 48 percent are planning or discussing its use (Bitkom). For education providers that mostly means the person asking is typically already familiar with a chat window from other contexts. The expectation is no longer that the chat exists, but that it knows something.

Enrolment as a dialogue instead of a five-step form

A classic enrolment form asks for everything that might ever be needed, from everyone. A dialogue asks only for what is needed for this course and this route. The difference shows in the result: someone who only has to give a name, a contact route and a course choice for an evening class drops out less often than someone working through fields for cost centre, provider number and invoice address that stay empty for them. Pre-qualifying in conversation produces a complete enquiry without sending everyone down the same path – that is the core idea of the lead assistant.

What matters is that the dialogue also works for people who do not operate it with a mouse. An enrolment route that is hard to use with visible focus, keyboard and screen reader excludes exactly the group that often needs continuing education most. The requirements and the typical gaps are summarised in the article on the accessible AI assistant.

Several locations, several term programmes

Education providers rarely work in one place. An adult education centre runs branch venues, an academy several cities, a driving school several branches with different vehicles and testing centres. For the assistant that is not a detail but a structural question: every statement has to know which location it applies to. How to solve that without a separate assistant per branch is described in the article on one assistant for many locations.

Every course record carries the location as its own field, not as part of the title. A location in the title is readable for people and worthless for a query.
Opening hours, travel details and step-free access belong to the location, not to the course. Otherwise they sit in the stock eighty times and go wrong eighty times individually.
Prices and reductions can differ per provider. Where they do, the deviation belongs in the record and not in a footnote on the general page.
The assistant asks for the location early where the question depends on it, and remembers the answer for the rest of the conversation.
For online courses the location falls away and the time zone takes its place. Both cases need their own rule, not a shared exception.
Before every programme change the stock is replaced in full and then checked against a list of test questions that contains the old state.

Through all of this: enrolment data is personal data, often linked to someone's employment situation. Where it is processed, under whose responsibility and with what retention period is not a technical footnote but part of the information a person enrolling may expect. The framework for that is set out on the page about privacy and hosting.

What the assistant must not decide

Four statements have no place in an automated conversation: promising a place, promising funding, judging an access requirement in a borderline case, and any information about the legal effect of a qualification. All four look harmless in chat and are commitments with consequences. The assistant may prepare them by recording the details in full, and it may say who makes them. When and how such a handover to a person is designed is described in the article on handover to human agents.

Whether that boundary holds in day-to-day use cannot be asserted, only measured. In a Bitkom survey, 62 percent of online shoppers wanted a quickly reachable human contact when something went wrong (Bitkom) – in our experience, willingness to talk to a system typically ends where things get uncomfortable. Those are precisely the points the analysis of recorded conversations reveals: which question came back without a substantiated answer, where someone broke off, which phrasing misled. How to evaluate that systematically is described on the page about conversation analytics. After the first term, that list is the set of working instructions for the second. Which tasks an assistant takes on in other industries is shown in the overview of industries; we are glad to go through your own course programme with you in an initial conversation.

Frequent questions

Can an AI assistant say bindingly whether a course still has free places?

Only if it retrieves the number from the booking system at the moment of the question. An answer from a text corpus reflects the state of the last import. The sensible approach is a tool call that returns current availability, and phrasing that names the time of the query. If the call fails, the assistant says so and hands over rather than estimating a number.

How many enrolment questions can actually be answered automatically?

That depends almost entirely on the data situation, not on the model. Date, price, prerequisites and format are answerable as soon as they are maintained as separate fields. Remaining places need a system connection. Funding decisions, borderline cases on prerequisites and anything that would amount to a commitment stay with the office. Storing the six most common questions cleanly covers most of the volume.

May the assistant say that a course is funded?

It may state which programmes exist and which conditions apply there, with a date and a pointer to the responsible body. It may not hold out the prospect that an application will be approved. That is the difference between information about a programme and a commitment in an individual case. The distinction belongs explicitly in the assistant's instructions.

Is there a right of withdrawal for a course booked online?

Section 4 of the German Distance Learning Protection Act gives participants a right of withdrawal under Section 355 of the German Civil Code, but only for a distance-learning contract pursuant to Section 3(2) of that Act – a contract concluded neither off premises nor as a distance contract. For a distance-learning course booked online, as for in-person offerings booked online, the general distance-selling framework of the Civil Code applies, from which services connected with leisure activities are excluded where the contract provides for a specific date or period (Section 312g(2) no. 9 of the German Civil Code). Which classification applies to a specific offering is a legal question of the individual case and does not belong in an automated answer.

How often does a course provider's knowledge base need maintaining?

In three rhythms. House rules change rarely and are reviewed when there is a reason. Course data is replaced in full at every programme change, not topped up, because otherwise the old state survives as a plausible answer. Remaining places and short-notice cancellations do not belong in the knowledge base at all but in a query against the system.

What happens to the data someone enters in the enrolment dialogue?

It is personal data and needs the same basis as an enrolment form: a clear purpose, documented consent where consent is required, a stated retention period, and information about where processing takes place. An assistant that collects contact details shows these points before collecting them and records in the transcript that they were shown.

Ask it yourself, Assistant of this website

Automated assistant

Settings