Answering Applicants Around the Clock: An AI Chatbot for Higher Education

Admissions teams answer the same questions all year. What are the entry requirements? When is the deadline? How much are the fees? Which documents do I need? Questions arrive by email, phone, web forms and messaging apps, often outside office hours and at peak times all at once.

Slow replies cost applicants, especially international students in other time zones. Staff spend their time pasting links instead of helping students with complex cases. Current students repeat the pattern with questions about registration, timetables and fees. An AI assistant can handle the routine questions. It must be accurate and accessible, and know when to pass a student to a person.

What this assistant does

The assistant uses retrieval-augmented generation (RAG): it searches your official content, then has a language model answer using only what it found. It:

• Answers general questions about programs, entry requirements, deadlines, fees and required documents, from your official pages.
• Links to the source, so applicants always land on the authoritative page.
• Shows application status after the student verifies their identity, through an integration with your student information system or admissions CRM.
• Works on several channels, starting with a website widget and adding messaging apps where your applicants are.
• Escalates to admissions staff with the conversation attached, for personal, complex or sensitive questions.

Designing the escalation path

Escalation is what makes students trust the assistant. Students should be able to ask for a person at any point. Some topics should always go to staff, such as appeals, disability accommodations, visa problems, financial hardship and wellbeing concerns.

When the assistant escalates, it creates a case in your admissions inbox or CRM with the conversation and a short summary. Outside office hours, it tells the student when to expect a reply. Staff should never have to ask a student to repeat their question.

How it works, step by step

1. Ingest documents. We crawl approved pages such as program listings, admissions requirements, fee schedules and deadline calendars, plus FAQs and handbooks. Each page is tagged with intake, program, student type and last update.
2. Split and index. Pages are split into sections and stored as embeddings, numeric representations of meaning, in a vector database. Exact facts such as dates and fees are also kept in structured form where possible.
3. Retrieve for each question. The assistant identifies the program, intake and student type the question is about, then searches only matching content.
4. Answer with citations. It replies in plain language with a link to the official page. Personal questions trigger identity verification or a handoff.
5. Log and improve. Unanswered questions, handoffs and feedback are reviewed by admissions staff. Gaps become new web content or FAQ entries.

Tools we use

• Backend: Python with FastAPI for crawling, retrieval and the chat service.
• Vector database: pgvector in Postgres or Qdrant.
• Language model: OpenAI or Anthropic Claude under business terms, or a private open-source model if your data rules call for it.
• Channels: an accessible website widget, and messaging channels such as WhatsApp where applicants expect them.
• Integrations: your student information system, admissions CRM or learning platform, through their APIs.

Check each vendor's current pricing, data terms and messaging-channel rules before you commit.

What you need to get started

• A list of the official pages and documents the assistant may use.
• Admissions staff who can supply real questions and correct answers.
• Escalation rules: which topics go to staff, and how quickly they reply.
• API access and a sign-in method if you want status lookups.
• Input from your privacy, accessibility and IT teams.

Typical scope and timeline

A first version is typically 2 to 4 weeks, depending on your content, channels and integrations. That is an estimate. A website assistant answering general questions is at the short end. Adding messaging channels and verified status lookups usually belongs in a second phase.

Timing matters too. Launching a few weeks before a busy application period gives staff time to review answers while volume is still manageable.

We have worked on education systems before. For Hart College, we built a student information system that made the enrolment workflow 5x faster. For a Panopto and Canvas lecture-capture automation, we cut administrative overhead by 80% and auto-published about 3,000 lectures a term. Neither project was an AI assistant, but both involved the student systems an assistant connects to.

How we keep answers accurate

• An evaluation set of real questions. Admissions staff supply common and tricky questions with correct answers. Every change is tested against them.
• Citations on every answer. Applicants and staff can check the official page in one click.
• Refusal when unsure. If the content does not answer the question, the assistant says so and offers to connect the student with staff.
• Intake-aware content. Deadlines and fees are tied to specific intakes, so last year's dates are not reused.
• Monitoring. Weekly reports show unanswered questions, handoffs and pages that are often cited.

Risks and how we handle them

Student privacy

General questions need no personal data. Status lookups require verified sign-in and show only that student's record. Logs are minimised and kept on a fixed schedule. Rules such as FERPA in the US, PIPEDA and provincial laws in Canada, and UK GDPR set obligations for student data. Review the design with your own counsel. Our article on RAG privacy by design covers the engineering patterns.

Accessibility

The widget must work with keyboards and screen readers, have readable contrast and use plain language. There should always be another way to reach a person.

International and multilingual applicants

Language models can reply in many languages, which helps applicants abroad. The official page remains the reference, so answers link to it. Fees, deadlines and visa-related details should always be checked against the source.

Wrong or outdated answers

A wrong deadline can cost a student a place. Intake tagging, citations and frequent re-crawls keep answers current. Admissions decisions and exceptions are always left to staff.

When not to build this

• Your website content is incomplete or contradictory. Fix the key pages first.
• Your question volume is small and staff already reply quickly.
• Your admissions CRM already includes an assistant that meets your needs and data rules.
• You cannot staff the escalation path. An assistant without a human fallback frustrates students.

How UnlockLive can help

We build student and applicant assistants with official-source answers, escalation to staff and the integrations behind status lookups. See our RAG development service and AI Workflow Automation for connecting student systems.

Related reads: WhatsApp AI assistants for booking, a support chatbot that cites its sources, and Telegram bots for business. To talk through your admissions questions and systems, book a free 30-minute call.