The most useful thing a college chatbot can say is “I don’t know”.
A student who gets a confident wrong answer about a fee or a last date is worse off than one who gets no answer at all. We think honesty about gaps is the feature that matters most.
When colleges ask us about a WhatsApp assistant for students, the first question is usually about coverage: how many questions will it answer? It is a fair question, but we think it is the second most important one. The first is: what happens when it does not know?
A wrong answer costs more than a missing one
Think about what students actually ask during admissions. Which documents to bring. The last date for a form. The fee for a course. Whether a subject combination is allowed. Say a hotel management college runs a short bakery and confectionery certificate alongside its degree. A parent asks on WhatsApp at 9 pm whether the bakery course fee includes the ingredients kit for practical classes.
If the assistant has no document that says so, there are three things it can do:
Response
What the parent does next
What it costs the college
Guesses “yes, it is included”
Plans on it; is surprised at the counter
A dispute, and a parent who no longer trusts the WhatsApp number
Gives a vague, friendly non-answer
Asks again, or calls the office
The phone call the assistant was meant to save
Says it does not have that information and passes the question to the office
Waits for a reply from a person
One question in the office queue, with the full chat attached
Only the third one is honest, and it is the only one that keeps the official number worth trusting. A college’s WhatsApp number carries the institution’s name. Anything it says, a student will treat as policy.
Why general chatbots guess
Language models are built to produce a fluent reply. Asked about a fee they have never seen, they will often produce a plausible one. That is fine for drafting an email; it is not fine for a fee, a date or an eligibility rule. The fix is not a cleverer model. It is a rule about sources: the assistant may only answer from passages it found in the college’s own documents, and if it finds no passage, it must say so.
That is how we built CampusDesk. It answers only from the brochures, circulars and timetables the college uploads. Each answer cites the document and page it came from. The model is told to say “I don’t know” rather than guess, and anything it cannot find goes to the office queue with the whole conversation, where staff can take it over from the dashboard.
Citations make the gap visible
A citation does two jobs. For the student, it shows where the answer came from, so “page 6 of the 2026 prospectus” carries more weight than an unsupported sentence. For the office, it makes every answer checkable. If a student says the assistant told them something wrong, staff can see the exact passage it used. Either the document was wrong, or an old circular was still live, and both are fixable.
Without citations, a wrong answer is a mystery. With them, it is a document problem with an address.
Design the hand-off before the answers
Most of the effort in a student helpdesk goes into answers. We would spend at least as much on the hand-off, because that is where trust is won or lost. A few things we would decide before going live:
Who owns the office queue in admissions week, and who owns it after hours.
How quickly a handed-off question should get a reply from a person, and what the assistant tells the student meanwhile.
Which topics should always go to a person, even when a document exists, such as fee waivers or individual results.
How staff close the loop: if the same question is handed off repeatedly, the answer belongs in a document.
The share of questions answered without a person is worth watching, but it is easy to push up the wrong way. A system that guesses will always look like it answers more. We would rather a college tracked three things together: how many questions were answered with a citation, how many were handed off, and how many answers were later found to be wrong. The last number should be close to zero, and the handed-off questions should shrink over time because the documents improve, not because the assistant got bolder.
If you are evaluating any student assistant, ask the vendor to show it a question your documents do not cover. What it does next tells you more than any demo of questions it can answer. For more on preparing documents, see the CampusDesk guides.
Questions
Should a college chatbot ever guess an answer?
We think not for fees, dates, eligibility or rules. If no passage in the college’s own documents covers the question, the assistant should say so and pass the question to a person.
How does CampusDesk handle a question it cannot answer?
It answers only from uploaded documents. If it finds no passage, it says it does not know and hands the question to the office queue with the full chat, where staff can take over from the dashboard.
Why do citations matter in a student assistant?
They show students where an answer came from and let the office check any answer against the exact document and page it used.