ChatGPT
Claude
Perplexity
Gemini
GrokAgentic AI acts; a generative chatbot answers
Agentic AI is software that pursues a goal and takes actions in your systems: booking an applicant onto an open day, updating a CRM record, chasing an incomplete application. A generative chatbot writes a reply and stops.
For an admissions office the distinction matters because it moves the risk. A wrong answer can be corrected in the next message. A wrong action (a booking in the wrong slot, an overwritten record, a reminder sent to the wrong person) leaves a trace in your systems and with the applicant.
The label is also heavily overused. Many products sold as "agentic" are chatbots with a new name. The simplest test is one question: does the tool write to your systems, or only read from them and reply? Our piece on the definition of an AI admissions agent sets out that boundary, and the guide to AI agents for student recruitment places it in the full applicant journey.
Three kinds of tool, compared
Rule-based chatbots, generative chatbots and agents differ in what they can do after the reply.
| Criterion | Rule-based chatbot | Generative chatbot | AI agent |
|---|---|---|---|
| Source of the answer | Hand-written decision tree | Language model, sometimes grounded in your pages | Language model grounded in your content and data |
| Possible action | Send a link | Send a link, draft text | Book, record, remind, hand over |
| CRM access | None | Often none | Read and write within a defined scope |
| Main risk | Off-script question | Invented answer | Wrong action repeated at scale |
| Expected control | Script maintenance | Sources and guardrails | Action log, limited permissions, human escalation |
The last row is the one that changes your operating model. An agent needs access governance, not only review of its wording.
Five admissions workflows it can take on
Five tasks appear in almost every admissions team, and an agent can handle them without touching the decision itself.
Qualifying the first contact. A visitor asks about a course, an intake or a campus. The agent asks the two or three missing questions (current qualifications, intended start, full-time or part-time) and passes on a usable record rather than a bare email address.
Booking open days. The agent offers a slot, records the booking and sends the confirmation. This is where the gap with a chatbot that points to a form is most visible.
Chasing incomplete applications. A missing document, or an applicant who goes quiet after a first reply, is a silent loss. The agent sends the planned reminder instead of waiting for an adviser to check a list.
Keeping the CRM current. Course of interest, questions asked, stage reached: the agent writes them to the record, so nobody re-keys them.
Handing over to a person. It passes the full conversation to an adviser when a request falls outside its scope. A poor handover cancels the gain from the other four, so test it before go-live and track it with the KPIs for an admissions AI agent.
What stays with a person
The admission decision, appeals and complex personal circumstances stay human, for reasons of accountability as much as quality.
European law frames this in several places, and it matters to UK institutions recruiting EU students or working with EU partners. Annex III of the EU AI Act lists, as high-risk, systems intended to determine access or admission to educational institutions. For high-risk systems, Article 14 requires that a person can disregard, override or reverse the output.
An agent that answers questions and books appointments does not have the purpose of deciding who is admitted. An agent that ranks, filters or scores applications moves toward that purpose and changes category. The legal classification depends on the intended purpose of the system, so ask your counsel before deployment rather than after.
Article 50 of the AI Act requires providers of systems that interact directly with people to design them so those people know they are dealing with an AI, unless that is obvious from the context. Say so in the first message.
What this means in the UK
UK applicants use UCAS for undergraduate applications, and the offer and Clearing process runs through that service, not through your website. An agent should explain, route and book. It should never imply that it can alter or promise a UCAS outcome.
Where it helps is around the edges: explaining entry requirements and foundation-year routes, answering Clearing enquiries quickly when staff are stretched, booking campus visits, and handling postgraduate and international enquiries that sit outside UCAS. UK data protection law is overseen by the ICO, which publishes guidance on AI and data protection on ico.org.uk, including a set of questions on solely automated decision-making. Treat that guidance as your starting point for documenting purpose, lawful basis and retention. If you recruit under-18s, be stricter about what the agent collects.
Five questions to ask before you choose
Put these to any supplier, including us.
- What can it write, and where? Ask for the list of possible actions and editable fields.
- Who can undo an action? A readable log and a rollback procedure are essential.
- Where is data hosted, and is it used to train a model? The answer belongs in the contract.
- How is human escalation tested? Ask for a trial on real cases, not a scripted demo.
- How much of the claimed performance is verifiable? Be wary of any result figure without a published method. No credible public benchmark currently sets a normal success rate for an admissions agent.
Where Skolbot fits
Skolbot deploys AI agents connected to the institution's CRM: the website agent answers from the institution's own content and passes the conversation and its qualification into the CRM. The CRM remains the source of truth, and the admissions team keeps control of the cases that need it. For the wider picture, read our guide to AI chatbots for student recruitment.
