ChatGPT
Claude
Perplexity
Gemini
GrokSummer melt is an American term, but Irish colleges see the same pattern
"Summer melt" comes from US research on students who intend to go to college, often with a deposit paid, and never arrive in the autumn. Irish admissions teams rarely use the phrase, but they know the pattern: an applicant accepts a CAO offer, then does not register, defers at the last minute, or leaves in the first weeks.
The published trials on how to reduce it are American. We say that up front because none of the numbers below measure Irish students, and we have not found a published Irish trial of an AI assistant in this window. What follows is what the US evidence supports, and how to adapt it to the CAO cycle and registration.
A note on method: no paid keyword-volume tool was available for this article, so it was scoped from public search results. Every figure below comes from a source we opened. Nothing comes from Skolbot or its customers.
What the US trials found
Castleman and Page estimated US summer melt "ranging from 8 to 40 percent" depending on the population. The wide range is the point: it varies by who the students are, and you cannot import it as an Irish rate.
| Study | Design | Published result |
|---|---|---|
| Castleman and Page, "Summer Nudging", 2015 | Automated, personalised texts or near-age peer mentors over the summer | Gains among students with less access to advice during the school year |
| What Works Clearinghouse review of the same study | 6,196 students, ten texts over July and August | Most comparisons not statistically significant; clearest gain in two Massachusetts sites; none in Boston |
| Page and Gehlbach, AERA Open, 2017 | Conversational AI texting would-be first years at Georgia State University | Committed students were "3.3 percentage points more likely to enroll on time" |
Both interventions were short, tied to a task the student could complete, and timed to a deadline. Neither tried to change a student's mind about the institution. The Georgia State assistant also predates generative AI, so treat it as proof of a pattern, not a forecast. Our reading of the wider research on admissions chatbots, for Irish institutions covers the other studies and their limits.
How the US setup differs from Ireland
| United States | Ireland | |
|---|---|---|
| Commitment | Deposit, typically by May 1 | Acceptance of a CAO offer, within days of the offer |
| Timing | Decisions in spring, three to four months before term | Leaving Certificate results and Round One offers in late August, weeks before term |
| Safety net | Waitlist, late admissions | Later CAO rounds and available places |
| Funding | Federal aid forms, institutional awards | SUSI student grant, often decided after the offer |
The consequence is that the Irish melt window is short and dense. A US college has a whole summer to nudge; an Irish college has a few weeks between offer acceptance and registration, often while the student is also hunting for accommodation. An agent that only sends generic reminders misses the moments when an Irish applicant actually decides.
Where an AI agent helps an admissions team
An agent can answer a student's question late on a Sunday: "When do I register?", "Where do I upload my documents?", "What happens if my SUSI grant is not confirmed yet?" It can also send reminders tied to the student's own checklist. Those two jobs map onto the Georgia State design.
An agent does not decide grant eligibility, approve a deferral, or talk a family through a financial crisis. Those go to a person. Our definition of an AI admissions agent sets out what an agent answers and what it hands off.
| Job | Agent | Staff member |
|---|---|---|
| Reminder before online registration closes | Yes | Not needed |
| "Which documents do I need for registration?" | Yes, from approved content | Backup |
| Questions about a SUSI decision | Gives the general process, then hands off | Yes |
| A student thinking of deferring or dropping out | Flags and routes quickly | Yes |
| Confirming that a fee or registration step is complete | Only if connected to the student record | Backup |
The last row is a capability question. An agent that cannot see the student record can only give general answers. Ask any vendor, including us, which systems the agent reads and writes, and test it before signing.
Who is most at risk in an Irish cycle
The US trials found their clearest gains among students with the least access to advice. That is a useful lens for an Irish team. Applicants who are the first in their family to go to college, who are moving away from home, or who accepted a lower preference while hoping for a later-round offer often have no one to ask the small practical question that decides whether they turn up.
The Higher Education Authority publishes data on access and progression that can help you decide which groups to watch. Your own registration data will tell you more: look at who accepted an offer last year and never registered.
A summer plan to test
This is an illustrative workflow, not a measured result. It follows the pattern the trials share.
- Build the checklist per student. Start from the steps your own data show hold up registration: accepting the offer, fees or grant status, documents, accommodation, orientation.
- Time messages to the CAO calendar. Results day, Round One acceptance and the registration window matter more than a fixed weekly schedule.
- Make every message answerable. A student who replies should get a useful answer, not a no-reply notice.
- Route emotion to people. Messages about money, accommodation or doubt should reach a staff member the same day.
- Prioritise the students with the least backing. The trial gains concentrated among students with fewer other sources of help.
- Clean the contact data first. A message sent to a wrong or missing number reaches no one, and the dashboard will not tell you.
Measure with a control group
Hold back a random sample of accepted applicants from the new outreach and compare the two groups on one outcome, such as completed registration or attendance in the first weeks. Year-on-year comparisons mislead because Leaving Certificate grades, CAO points and the supply of places change from year to year.
Data protection
Messaging applicants means processing personal data under EU and Irish data protection law. The Data Protection Commission publishes guidance, including on electronic direct marketing. Confirm your lawful basis and consent wording for text and chat-app messages with your data protection officer. Tell applicants when they are talking to an automated assistant, and keep admissions decisions with people.
A reminder about a step the student must complete is a different thing from promotion. Your officer should confirm where each message falls before a campaign goes live.
Where this fits in your recruitment plan
Summer melt is the last stretch of a longer funnel. The guide to AI agents for student recruitment shows where a post-offer agent sits in the full journey, from first enquiry to registration.
