ChatGPT
Claude
Perplexity
Gemini
GrokAn AI agent for university communications answers from approved content; it does not publish
An AI agent for university communications is an assistant that answers applicants and parents on several channels, using only content the institution has approved. It does not replace the press office, the editorial calendar or the brand guidelines.
The confusion comes from the word "communications". A communications team does two very different jobs: publishing (news, social media, campaigns, press) and answering ("Is this course available with a placement year?", "What does the course cost?"). The AI agent belongs to the second job. Treating it as a publishing tool means asking it to improvise on subjects where a mistake costs the university credibility.
This article sets out the split, then the conditions for giving the same answer on the website, by messaging and by email. For the wider recruitment picture, see our guide to AI agents for student recruitment.
Two jobs, two sets of rules
Institutional publishing tolerates creativity and is approved before release. An admissions answer tolerates very little approximation and must be traceable, almost word for word, to an official university source.
| Institutional publishing | Controlled admissions answer | |
|---|---|---|
| Subject | News, campaigns, social media, press | Courses, entry routes, published fees, bookings |
| Source | Editorial calendar and brand messages | An approved knowledge base with a named owner |
| Approval | Before release, by the communications lead | When the knowledge base changes, then transcript review |
| Margin for error | A weak phrase can be corrected | A wrong answer misleads an applicant |
| Role of AI | Drafting support under human control | Answers inside the approved scope, handover outside it |
A poorly scoped agent blurs this line. It treats a dated blog post as if it were policy, or turns a campaign slogan into a promise. The fix is not technical: decide which content counts as an answer source and which does not.
Reliability comes from the source, not the channel
An answer is reliable when it comes from an identified, current source and the agent knows when to stop. The channel changes the shape of the message, not its truth.
Four practical requirements:
- One knowledge base. Every channel draws on the same content. Two bases mean two versions of your entry requirements.
- An owner and a review date per item. Admissions owns the substance; communications owns tone and consistency.
- A rule for silence. When the base has no answer, the agent says so and offers a person. Our piece on chatbot hallucinations and guardrails covers the controls.
- A trace. For each answer you should be able to find the source used, so errors are fixed at the root.
To feed the agent properly, see how to train a chatbot on your institution's data.
Several channels, one voice
Each channel has its own use and constraints, and the agent should adapt the form without changing the substance.
| Channel | Typical use | Watch point |
|---|---|---|
| University website | Course questions while the applicant researches | A greeting that says it is an AI |
| Messaging apps | Follow-up, booking reminders | Prior consent, restrained frequency |
| Reply to an enquiry, confirmation | Consistency with what the website said | |
| Social media | Public questions and private messages | Do not answer an application query in public |
Not every vendor has every channel live. Ask what is in production today and what is on the roadmap; a channel that is announced but not delivered should not appear in your plans. For messaging apps, check consent and the platform's own rules at the source before anything is sent.
Multilingual answers follow the same logic: an answer in Mandarin or Spanish for an international applicant must come from the same source as the English one. See our article on the multilingual chatbot.
Who does what: communications, admissions, data protection
The agent holds up only if three functions share the ground clearly.
- Admissions approves the answer content, sets the handover criteria, takes over conversations that need a person.
- Communications keeps tone and brand consistent, aligns campaign messages with the base, stops editorial copy being mistaken for a rule.
- Data protection lead checks purpose, lawful basis, retention and the information given to applicants.
A simple routine is enough: when a published fact changes (fees, application dates, course arrangements), update the base before the announcement, not after. Review the conversations where the agent handed over or hesitated; they show where the base is incomplete. For monitoring, use the KPIs for an admissions AI agent.
The agent supports the team rather than replacing it. Our comparison of AI chatbots and human advisers shows where to draw the line.
The UK framework
Three topics frame a UK deployment: personal data, transparency about AI, and whether the system could count as high risk.
Personal data. The ICO sets out expectations under UK data protection law: a defined purpose, a lawful basis, data minimisation, retention limits and clear information. An agent that talks to sixth-formers, some under 18, needs a documented assessment, and the vendor should state where data is hosted and whether it is used to train a model.
Transparency. Article 50 of the EU AI Act provides that people should know when they are interacting with an AI. It applies to EU deployments; a UK university recruiting in the EU should take note, and telling every applicant at the first message is good practice anyway.
High risk. Annex III of the EU AI Act covers systems used to determine access or admission to education. An agent that informs and books appointments is not designed for that; one that ranks or screens applications comes close. Have counsel confirm the classification.
On the admissions side, UCAS remains the route for most undergraduate applications, and the agent should explain and point to it without predicting outcomes. During Clearing, volume peaks and answers go stale quickly, which is exactly when the knowledge base needs a named owner. Quality expectations come from the Office for Students in England, so the agent must never state claims about teaching quality that the university cannot evidence.
Five questions to ask before choosing
- Which base feeds each channel? The right answer is one, and you control it.
- What does the agent do when it does not know? Ask for a demo on a question missing from the base.
- Can you trace the source of any given answer?
- Which channels are live today? Separate production from roadmap.
- Where is data hosted, and is it used for training? This is a contractual point.
Where Skolbot fits
Skolbot Chat is the agent on the institution's website: it answers from the content the institution supplies and approves, qualifies the enquiry and passes the conversation to the CRM, where the admissions team stays in charge. A messaging extension is offered as an option; details to be confirmed during a demo.
