A chatbot for colleges should make the next step clearer for a prospective student. That could mean finding an evening course, understanding an application requirement, checking where an open day takes place, or learning whom to contact about an unusual situation. The strongest starting point is the information your college already publishes, organised around the questions people actually ask.
A polished answer is not enough on its own. The assistant needs to distinguish campuses, course versions, entry routes and academic years. It should recognise when a student needs a person, and it should avoid turning a general eligibility rule into a promise of admission. This guide describes a practical public website assistant, rather than an admissions decision system or a replacement for your student portal.
Start with one student journey
Choose a specific audience for the first version. A college offering adult learning, vocational qualifications and full-time study may have several quite different application processes. Starting with adult evening courses, for example, creates a manageable set of documents and questions. Starting with everything often exposes contradictory information before anyone has agreed who owns it.
Write down the five decisions that audience needs to make. Which subject fits their goal? Can they attend the teaching schedule? What prior experience is expected? What does the published fee include? Where do they apply? These decisions provide a useful structure for knowledge and testing. They also help you avoid a bot that talks at length about the institution without helping someone choose a course.
The opening message can be simple: “Hi, I can help with courses, application information and visiting the college. What would you like to know?” Suggested questions should mirror real journeys, such as “Do you offer evening classes?” or “How can I visit the campus?”
Build a course information pack
Collect approved course descriptions, entry guidance, fee explanations, term dates, campus directions and application instructions. Use the version intended for public distribution. Internal assessment notes, applicant records and staff-only discussions do not belong in a public website knowledge collection.
Every course entry should identify the course name, campus, study mode, relevant academic year and information owner. If similar qualifications exist at two campuses, make the distinction explicit in the text. A generic paragraph saying “classes run on Tuesday” becomes ambiguous when detached from the page around it. A sentence naming the course and campus remains understandable when retrieved as an individual passage.
Include the meaning of specialist terms. Prospective students may not understand the difference between a certificate, a diploma and a foundation programme. Your assistant can explain your own published descriptions in accessible language, then link to the full course information. Avoid asking it to invent equivalencies that the college has not approved.
Keep dates and entry rules in context
Time-sensitive information needs especially careful organisation. An application deadline for one intake should not be reused for another. Label the year in headings and paragraphs, and remove or clearly mark superseded prospectuses. A collection of three nearly identical PDFs can look comprehensive while making reliable retrieval harder.
Entry requirements often involve alternatives. A course might accept a named qualification or relevant experience reviewed by the admissions team. The bot should preserve both routes. It should not simplify the policy into “you need this qualification” when the document describes an alternative, or promise acceptance because someone mentions experience.
Test hypothetical questions: “If I have worked in the industry but do not have the qualification, can I apply?” A useful reply explains the published route, makes clear that the college evaluates individual applications, and points to the correct contact or application page. The answer should advance the conversation without making a decision it cannot support.
Design for different kinds of visitors
A parent, an employer and an applicant may ask the same short question for different reasons. “How much does it cost?” could refer to tuition, equipment, transport or a short course. Teach the assistant to ask a focused clarifying question when the missing detail changes the answer.
Avoid interrogating every visitor before offering basic help. If someone asks where the library is, answer from the campus information or ask which campus they mean. There is no need to collect an email address to provide public directions. Contact details belong in an optional request when a person wants a response from your team.
Use familiar language and short paragraphs. A student reading on a phone should be able to spot the course name, the relevant condition and the next action. Longer replies can be appropriate for comparing study options, but the assistant should lead with the main distinction rather than reproducing a prospectus page.
Create the college assistant in Chaat
Start with the guided creator or manual setup. Describe the audience, the information it can answer and the boundaries of its role. For example: “Help prospective adult learners understand our evening courses. Ask which course they mean when necessary. Explain published requirements, and direct individual admission decisions to our admissions team.”
Add a public website URL, upload relevant documents, or paste a carefully prepared FAQ. Website extraction becomes a removable knowledge source. Review what was extracted: linked portals, embedded timetables and downloadable material may need separate attention. Do not assume that one URL guarantees complete coverage of every institutional system.
Set the bot name, greeting, suggested questions and reply length. Match the colours and photo to the college identity without sacrificing contrast. Decide whether to enable the contact form, and configure request notifications if the team wants email alerts. A named staff owner should know where requests arrive and who checks them.
Test conversations rather than isolated questions
Build a small acceptance set using actual enquiry themes. Include an exact course title, a shortened course name, a spelling mistake and a follow-up that changes the campus. Test “Can I study this after work?” after asking about a full-time programme. The bot should use the conversation context while noticing that the visitor’s needs may not match the original option.
Add questions whose answers are absent. Ask about a future intake that has not been announced, a fee exemption not covered by the material, and the status of an individual application. A good assistant explains the limit and offers the relevant next step. It does not convert missing information into a confident answer.
Chaat’s Test view can show sources for review. Compare the response with the underlying document, especially dates, qualification names and exceptions. Public shared and embedded chats do not display those testing source labels. The reviewer still needs to confirm that any links included in a reply lead to the appropriate public page.
Launch where the question happens
Place the widget where prospective students need context: course pages, open-day information and admissions guidance. Check the launcher on a narrow phone screen alongside cookie controls and accessibility tools. Visitors should still be able to use the page’s main application button.
A shareable chat link can support staff reviewing the assistant before the website launch. Explain its scope to reviewers and ask them to try their usual enquiries. A staff member who handles phone calls may contribute more useful test questions than someone who only reviews the institution’s polished marketing language.
Keep the normal admissions route visible. The chat is an additional way to find information, not a compulsory gateway. Someone who prefers reading a full course page or contacting the college directly should still be able to do so.
Give the content an owner
Assign responsibility for course facts, fees, dates and contact information. Before each intake, review the sources and retest a short set of important questions. Changes to a PDF should trigger a knowledge update; changing the original file elsewhere does not automatically prove the bot has the latest version.
Review conversations for patterns rather than just volume. Repeated questions about evening attendance may reveal unclear course descriptions. Repeated confusion between two campuses may indicate a naming problem. Improve the source material and the website as well as the bot’s instructions.
Useful measures include correct answers on a reviewed sample, successful navigation to relevant pages and the clarity of contact requests. A lower number of emails is not automatically a success if students are giving up. Ask reviewers whether the conversation left them knowing what to do next.
Questions colleges should resolve before launch
Can the assistant check application status? A public knowledge chatbot cannot infer a private application record. Provide the appropriate portal or contact route unless a separately implemented, authorised integration supplies that information.
Should it answer questions about student support? It can explain published services and how to reach them. Individual circumstances, accommodations and urgent welfare issues require appropriate human channels. Agree the wording with the teams responsible rather than treating these as ordinary course marketing questions.
Does every department need a separate bot? Start with the smallest useful scope. Separate assistants may help when information ownership and audiences differ substantially. A single assistant can work where sources are clearly labelled and maintained together. The decision should follow the information architecture, not a desire to deploy as many widgets as possible.