How does a school, nursery or course train the inTusell AI? A step-by-step guide
School AI training in six steps: institution type, knowledge base, tour and trial-lesson appointments, persona, response rules and staff approval. Requires no technical knowledge.
During admissions season, an education provider's inbox fills up with the same questions: "Do you have places in middle school?", "Can we come and see the campus?", "How much is the monthly fee?", "Can we take a trial lesson?". A private school, a nursery, a study center, a language school, or a private tutor — it makes no difference; the same twenty questions repeat a hundred times a day. While the front desk tries to keep up with WhatsApp, dozens of "price?" comments pile up under an Instagram post. The parent who writes "can we visit the school tomorrow?" at 9 PM waits until morning and goes to another institution. Every message answered late is a prospective family lost.
inTusell takes over this load. To do that, it first has to learn your institution, your grade levels, and your admissions routine. This article explains step by step how an education provider trains the inTusell AI from scratch. It is the education chapter of our sector-based "how to train your AI" series, and the main article of that series.
Quick answer
An education provider trains inTusell from the panel: pick your institution type at setup (private school, nursery-preschool, tutoring center, language school, general course provider), upload program/fee/admissions information to the knowledge base, set up the tour and trial-lesson appointment system, configure the persona and tone, write response rules such as "quote fees only from the written source," and teach it from past enrollment conversations. In schools and nurseries, WhatsApp, Instagram DM, Messenger and Telegram replies pass staff approval before reaching a parent; web chat and email are outside that queue. The AI gives concrete facts (fees, address, hours, conditions) only from the knowledge base you upload; it does not invent what it cannot find, promises no discounts or places, and gives no academic advice.
Why does training matter?
inTusell is not an off-the-shelf chatbot. It is a fully autonomous AI assistant that behaves like an experienced enrollment advisor. A good enrollment advisor cannot start on day one without knowing the institution either. They need to know which programs you offer, your capacity and fee routine, and which question to leave to the instructor. Training makes the AI do two things correctly: give the right information (program content, fees, dates) and give it the right way (your institution's language, tone, and enrollment discipline).
An untrained assistant either gives overly generic answers or guesses on a topic it doesn't know. During enrollment, guessing is expensive. A wrong date, a made-up fee, or a remark about the student's level misleads the parent and undermines trust at the same time. A well-trained assistant stops when it doesn't know and directs the person to the instructor or the enrollment advisor.
Who is it for?
This guide is for education providers whose inbox is dominated by enrollment and information requests:
- Private schools (preschool, primary, middle and high-school levels) and their admissions desks
- Nurseries and preschools
- Study centers, cram-school models, and private tutors
- Language schools, foreign-language academies, and exam-prep centers
- Art schools such as music, painting and ballet; coding bootcamps and certificate programs
- Teams that get dozens of tour, trial-lesson and information requests a day from Instagram and WhatsApp
You don't need a technical team. You complete every step below from sections in the panel, without writing code.
The first decision at setup: institution type
Education does not fit a single mold. A private school's admissions process and a language school's enrollment process are not the same thing: one is built around bringing the family to campus and showing them around, the other around level assessment and group placement. That is why you pick your institution type first:
| Institution type | Admissions flow | Typical goal |
|---|---|---|
| Private School (K12) | Tour-based 8-stage admissions funnel | Introductory meeting and campus tour |
| Nursery & Preschool | Tour-based 8-stage admissions funnel | Site visit and introductory meeting |
| Tutoring & Study Center | Broader education flow | Level meeting, trial lesson |
| Language School | Broader education flow | Level assessment, trial lesson |
| Course / Training Center (General) | Broader education flow | Trial lesson, enrollment meeting |
The choice configures four things at once: the AI's persona and tone, the stages of your admissions funnel, the custom fields the AI collects during a conversation, and the ready-made objection handling. For private schools and nurseries the funnel is tour-based and runs through eight stages: New Prospect, First Contact, Tour Booked, Tour Done, Offer Presented, Decision, Enrolled, Lost. A separate article covers working this funnel: the school admissions funnel: eight stages from inquiry to enrollment.
You choose the institution type during setup. If you need to change it later, write to our support team; there is no separate institution-type selector in the panel yet. What you can change from the panel is your industry (Settings > Company Settings > Industry). Click "Apply industry defaults" and a preview shows what will change; your manual customizations are preserved.
The structure of training: 5 core steps + 2 continuous steps
Splitting the training into two groups makes your job easier. The first 5 core steps get the assistant up and running; you do them once. The 2 continuous steps that follow — label review and opening channels — are the ongoing part you keep sharpening as you use the assistant. Here they all are, in order.
1. Upload the knowledge base
The foundation of training is the knowledge base (RAG). In the panel's Knowledge Base section, you upload all of your institution's written information. Each time the AI generates a response, it searches these sources automatically and uses only the information written there.
Supported source types:
| Source type | Typical content | Example |
|---|---|---|
| Program/curriculum descriptions, pre-information | "Summer term program.pdf" | |
| Excel / CSV | Course list, fees, capacity, lesson hours | "Groups and fees.xlsx" |
| Web URL | FAQ and program pages on your site | "/frequently-asked-questions" |
| Free text | Frequently asked individual questions | "Which documents are required for enrollment?" |
Every file you upload is chunked automatically and made searchable with pgvector. Upload a "summer term program" document, and when a parent asks "What time is the B2 group on weekends?", the AI finds the right line and responds.
The must-haves for an education provider: program and course content, levels and groups, lesson hours and term calendar, fees and discount terms (if you prefer to share them), capacity status, enrollment documents, instructor staff, and the most frequently asked questions. One important limit: the knowledge base is not for academic assessment. The AI uses what you put here to answer "which program, which time, how to enroll" questions. It leaves questions that need expertise, like "what level should my child be at?", to the instructor.
2. Set up the tour, trial-lesson and consultation appointment system
The knowledge base explains what you offer; the appointment system lets the assistant give the parent or student an actual time. For an education provider, the most valuable conversion is getting an interested person into a campus tour, a trial lesson or an enrollment meeting — that is where the enrollment most often closes. In the panel you define three things:
| Definition | Content | Example |
|---|---|---|
| Appointment types | Event + duration (+ break/buffer if any) | "Campus Tour 45 min", "Trial Lesson 45 min", "Enrollment Meeting 20 min" |
| Working hours | Opening/closing for each day of the week | Mon-Fri 09:00-18:00, Sat 09:00-14:00 |
| Exceptions | Holidays, public holidays, term breaks | "Mid-term break — closed" |
There are two appointment shapes, and they do different jobs.
A one-to-one appointment is a slot reserved for a single family — a campus tour or an enrollment meeting. The AI suggests times based on your working hours and busy slots, the system prevents two appointments at the same time, and if Google Calendar is connected the appointment is copied one-way to your calendar. That copy covers appointments the AI books and ones you create by hand in the panel. Bookings taken through your public appointment link, and reschedules a parent makes from that link, are not reflected on the calendar today — treat the panel record as the source of truth there.
A group session with capacity is a fixed-hour, multi-attendee event, such as an open-day meeting or a group trial lesson. You define something like "Saturday 14:00 introductory meeting, capacity 20", and the AI invents no arbitrary times. It shows upcoming sessions with their remaining places, registers the person into the session they choose, and re-checks capacity at the moment of registration. You manage group sessions from the panel, and calendar sync does not run for those records.
The real power of the appointment engine is in the reminders. The system sends automatic reminders 1 day and 2 hours before the appointment, which noticeably lowers the trial-lesson no-show rate — the biggest headache for education providers. The parent can reschedule or cancel from the /manage-appointment/{token} link, which frees the slot for someone else. A separate article gathers the details of the trial-lesson and enrollment flow: enrollment and trial-lesson automation for education providers. You can also see the education capabilities on the education solution page.
3. Set the persona and tone
You taught the AI what to say in the first two steps. Now you define how it says it. There are two settings in the panel:
- Assistant name (
ai_persona_name): The name it uses to introduce itself to the parent or student. Most institutions choose a warm name that fits their enrollment team. - Tone (
ai_tone): For education, the typical tone is warm, patient, and guiding. You can strike a reassuring line with parents and a friendlier one with students. Sales pressure stays low, and informing comes first.
At education providers, the audience usually splits in two: for younger age groups the parent decides, and in adult courses the student decides. The AI picks this up from the language of the message. It speaks to the parent about the child's development and to the adult about career and goals.
A note on languages: the AI responds in whatever language the person writes in. When messages reach a foreign-language school in English, Arabic, or another language, you do nothing extra — the assistant detects the language automatically and responds in it while keeping the tone.
4. Define the response rules
The persona defines who the assistant is; the response rules define how it behaves. At an education provider, this step passes your enrollment discipline on to the AI. With the AI Manager Chat, you add rules by writing in plain language, just as you would brief a teammate.
Typical rules for education providers:
- "First find out the grade level and what matters most to the parent, then direct them to the right member of staff."
- "When asked about the net fee, quote the current fee only from the written source in the knowledge base. If there is no written source, say it will be shared in writing at the meeting and direct them to an appointment." (You can change this to match your institution policy.)
- "Always invite everyone interested to a campus tour, an introductory meeting or a trial lesson."
- "Do not give academic advice, level assessment, or exam-result interpretation; leave that to the instructor."
The guardrails a school setup ships with
In the private school and nursery institution types, some rules are on by default without you asking for them. These are not optional tone preferences; they cover four topics that carry institutional-reputation and regulatory risk.
Let us state the mechanism plainly: these rules are taught to the AI as non-overridable instructions and are never trimmed, even when the prompt budget is tight. That is why the AI does not produce these sentences. This is not a deterministic filter that cuts or rewrites the reply before it is sent, so look for the assurance in the right place:
| Heading | What the AI does |
|---|---|
| Discounts | It relays only the discount conditions defined in your knowledge base (early registration, sibling, payment in full, and so on) exactly as written. It produces no figure or commitment outside those conditions; a request beyond them is noted and handed to a staff member. |
| Places and queue | It does not produce sentences like "I've reserved your place", "only three places left", "we'll open a place for you". If you are full it says so honestly, explains the waiting list, and promises no date or queue position. |
| Results claims | It gives no exam-success, ranking or outcome guarantee. It relays a success figure written verbatim in your knowledge base only as written; it does not round it, update it or generalize from it. |
| Children's data | It never asks on its own for a student's name, identity details, or health/special-needs information. Grade and class level are enough. If a parent shares something sensitive, the AI does not record the detail; it only notes that there is a specific topic to discuss and suggests the matter is better covered face to face with a staff member. |
The technical counterpart of that last row: the AI never auto-extracts fields such as the student's name and birth year from chat. Only your staff can record them, with the parent's confirmation. Fields like source ("how did you hear about us") and referrer can be auto-extracted, though the referrer is a third person's name and therefore also asks for the parent's confirmation.
The most common mistake in education is an AI producing a confident answer on a topic it doesn't know — "will my child pass this exam?", for example. inTusell does not treat such topics as the AI's job. It says honestly that it doesn't know and hands off to the instructor. Nor does it make up a fee or a date that isn't in the knowledge base.
If you want to compare two different approaches, use the A/B test feature. You can put "give the fee first" next to "invite to a trial lesson first" at 50%-50% traffic and measure which one converts to more enrollments.
5. Teach from past conversations
This is the step that moves training from good to specific to your institution. In the panel, you upload audio recordings of your past enrollment conversations (MP3, MP4, WAV, M4A) and mark each one Won or Lost — did the parent come to the trial lesson and enroll, or did you lose them?
The system uses these recordings in two ways:
| Recording type | What the AI learns |
|---|---|
| Won conversations | Enrollment advising: asking the right question, directing to a trial lesson at the right moment |
| All conversations | Program, fee, and FAQ information (fed into RAG) |
From these, the AI learns how your best enrollment advisor put an undecided parent at ease and moved someone who said "let me think about it" toward a trial lesson. The KVKK side is protected: uploaded recordings get PII (personal data) masking and explicit consent.
This step is not mandatory, but don't skip it. The knowledge base teaches the AI what it knows; past conversations teach it how you convert to enrollment. Put the two together and the assistant really does resemble an experienced enrollment advisor.
The default in schools and nurseries: staff approval
Private school and nursery setups differ from the other education types on one point: in these two institution types the AI reply lands in the approval queue as a draft before it reaches a parent. Your staff read the draft, correct it if needed, and send it. The reason for that default is simple. In a conversation with a child's parent, the institutional cost of a single wrong sentence — an invented discount, a careless promise of a place — is far higher than a few seconds of delay.
The default is a starting point, not a lock. You can turn approval off from Panel > AI > Advanced Settings and switch to instant replies. Most schools spend the first weeks in approval mode, see that they send most drafts unchanged, and then loosen the mode on frequently asked topics. For the other education sub-types — tutoring center, language school, general course provider — the default is already off, and you can turn it on if you wish.
You lose no speed while approval is on: the message lands in the single queue immediately, you can see who asked what, and the draft is waiting. A separate article covers that balance and how to measure it: first-response time in school admissions.
6. Correct with label review (continuous step)
The first five steps get the AI up and running. What follows is the continuous part that perfects it over time. Every AI response lands in a label review queue, where you can do three things: approve, reject, or correct.
The AI learns from these corrections. Say it described a group's hours incompletely, or called a program by the wrong name. You correct it, and from then on it uses the right wording in similar cases. Over time, patterns specific to your institution build up: your program names, your standard informational sentences, the way you invite people to a trial lesson.
For the first two weeks — especially during enrollment season — we recommend spending 10-15 minutes a day in this queue. The correction rate drops quickly during that period, because the AI learns the mistakes it makes most often. Review is the live part of training: the system gets smarter as you use it.
7. Open the channels (continuous step)
Once training is ready, you put the assistant in front of parents and students. inTusell gathers your channels in a single inbox: Instagram DM, WhatsApp, Facebook Messenger, Telegram and web chat. Instagram comment handling is a separate module that has to be enabled on your account. Email is not a channel you connect; when someone replies to an email inTusell sent, the reply lands back in the same conversation.
How you open each channel:
- Meta channels (Instagram, Messenger): connect with one-click OAuth.
- WhatsApp: connects in about 1 minute by scanning a QR code from your phone — no Meta Business approval required. The official Cloud API option is also available.
Channel gates let you control each channel separately. The Instagram DM gate (instagram_dm_enabled) is separate from the Instagram comments gate (instagram_comments_enabled). You can automate only DMs and WhatsApp first, keep the comments under promotional posts manual, and open the comments gate as well when you're ready.
Handing off to a human has three working modes: ai_only (the AI answers everything), human_only (everything goes to the live team) and hybrid (the AI normally answers and escalates to the team when needed). Most institutions start with hybrid. The AI handles trial-lesson and information traffic, and conversations that need academic advising go to the instructor. When a new term opens, proactive follow-up can also take a renewed offer over WhatsApp to suitable parents already in follow-up. For the details of Instagram and WhatsApp automation, read Instagram and WhatsApp automation for education providers.
How long does training take?
A working setup takes half a day:
- Uploading the first files to the knowledge base: 1-2 hours (shorter if your program and fee lists are ready)
- Tour / trial-lesson appointment types, group sessions and working hours: 20-30 minutes
- Persona, tone, and first response rules: 30 minutes
- Connecting channels: 1-5 minutes per channel
But there is no moment when training is done. Over the first two weeks, the responses you approve and correct in the label review queue sharpen the assistant to fit your institution. Uploading past conversation recordings and running A/B tests keep improving it after that. Setup is fast; mastery is continuous.
What it isn't
Putting inTusell in the right category matters, because the wrong expectation leads to the wrong setup.
- It is not an instructor or an academic advisor. It does not assess level, does not interpret exam results, and does not give academic advice. It directs these topics to the instructor.
- It is not a school management system or an LMS. It holds no grades, attendance registers, official student records, or lesson content. It is a pre-enrollment, parent-communication and tour/trial-lesson coordination layer. Because it holds no attendance data, it also produces no "the student did not come to class" notification.
- It is not a timetabling tool. It sends reminders for the appointment or group session you opened in the system — a campus tour, an introductory meeting, a trial lesson — not for a lesson slot.
- It is not a tool that promises success. It never says things like "you will definitely pass" or "guaranteed result". It informs the person and directs them to a trial lesson.
- It is not an intermediary that replaces the institution. The institution always provides the education and handles the enrollment; inTusell is not a party to the payment.
In short: it is not a bot that gives canned answers. It is an enrollment assistant that represents your institution as well as you train it — and leaves the academic decision to the instructor.
Frequently asked questions
What does choosing an institution type at setup change?
At setup you pick private school, nursery-preschool, tutoring/study center, language school or general course provider. The choice changes the AI's persona and tone, the stages of your admissions funnel, the custom fields the AI collects, and the objection handling that ships with it. For private schools and nurseries the funnel is tour-based: New Prospect, First Contact, Tour Booked, Tour Done, Offer Presented, Decision, Enrolled, Lost. Tutoring centers, language schools and general course providers get a seven-stage trial-lesson funnel: New Interest, Informed, Demo / Trial Lesson, Awaiting Enrollment, Awaiting Payment, Enrolled, Dropped. The institution type is chosen during setup; to change it later you need to write to support.
Do AI replies go straight to parents in a school setup?
In private school and nursery setups, no — not on WhatsApp, Instagram DM, Messenger or Telegram. The AI reply lands in an approval queue as a draft, and your staff read it and send it. Web chat and email do not enter that queue and are answered directly. That is the default for these two sub-types, because a single wrong sentence in a conversation with a child's parent carries a high institutional cost. You can turn approval off from Panel > AI > Advanced Settings and switch to instant replies; for the other education sub-types the default is already off.
How long does education AI training take?
You can complete a working setup within half a day: uploading program, fee and capacity information to the knowledge base, setting trial-lesson appointment types and working hours, configuring persona and tone, and writing a few response rules. The real refinement builds up over the first weeks as you approve and correct responses in the label review queue. Training is not a one-time job; it is continuous.
Does the AI quote the enrollment fee or course price?
It follows your institution policy. If you upload net prices and discount terms to the knowledge base, it shares them. If you do not, it invents no price and directs the person to a trial lesson or a consultation instead. Most institutions save the campaign price for the consultation and tune the AI accordingly.
Do I need technical knowledge to train the AI?
No. You do all the training from the panel; no code, API key, or developer is required. You upload PDF and Excel files to the knowledge base, enter trial-lesson appointment types and working hours, and write response rules in plain language. You connect channels with one-click OAuth or by scanning a QR code.
Does the AI create the trial-lesson appointment on its own?
Yes. The AI agrees a suitable day with the parent or student and creates a trial-lesson or introductory-meeting appointment with your working hours and full slots taken into account. The system prevents the same time from being double-booked. Automatic reminders go out before the appointment, which lowers the no-show rate.
Is parent and student data safe under KVKK?
Yes. inTusell is end-to-end encrypted and KVKK-compliant. When the student is a minor, the parent contact information is sensitive data, so the system works with PII masking and an explicit-consent record and processes personal data only for enrollment and communication purposes. The AI produces no academic advice or assessment.
Does the AI decide what level the student is at?
No. The AI does not assess level, does not give academic advice, and does not interpret exam results. Those are the instructor's job. The AI gives information and directs the person to a trial lesson or a level-assessment meeting, and it hands any topic that needs expertise off to the instructor or consultant.
Next step
You've trained your assistant; next comes using it in daily operations. The next article in the series explains how to run the inTusell you trained during a real admissions season: incoming tour and trial-lesson requests, calendar, reminders, fewer no-shows, and handing off to the instructor. Continue straight from there with how an education provider uses inTusell.
The other three articles in the school cluster go deeper into each of the steps above: the school admissions funnel: eight stages from inquiry to enrollment, first-response time in school admissions and consent in parent communication. You can see the full capability set on the schools and education solution page.
The same 5-step logic works in every sector, so take a look at the neighboring main articles too: AI training for law firms, AI training for tour agencies, and AI training for clinics. If you'd like to see it live, use Get a demo and we'll open your panel together, or write to hello@intusell.com. The pricing page covers package details, the support center answers common questions, and the all articles list holds the other guides.
You read the blog — now see it live.
Test inTusell live with your own sector scenario in a 20-minute demo.