Is It Safe to Use AI for College Applications?
The program that no longer existed
A parent is proofreading her son's "why this college" essay the night before it is due. It reads well. It is warm and specific, and it names a research program he says he cannot wait to join.
She looks the program up. The college closed it two years ago.
Her son had not invented it. He had asked a chatbot for help, and the chatbot supplied the program along with most of the paragraph around it. So there were two problems on one page: a fact that was wrong, and writing that was not his.
Students are already using it
This parent is far from alone. In a national survey by EAB of more than 5,000 high school students, 46% said they had used AI tools in their college search, up from 26% a year earlier.
For most families, then, the decision is about which AI their student uses and what it is allowed to do.
What goes wrong with a general chatbot
A general chatbot is built to be helpful with anything, which means it has no rules about college applications. Three risks follow from that.
It will write the essay. Ask, and it produces a finished draft in seconds. Colleges treat that as fraud. The Common App's fraud policy covers submitting the output of an AI tool as your own work, and Notre Dame, which follows that policy, says it can withdraw an offer of admission over it.
It also does not work. Researchers at Cornell compared 30,000 essays written by applicants with essays from eight AI models. The AI essays stayed generic however much detail about the applicant the models were given, and a classifier told the two apart with near-perfect accuracy.
It states things that are not true. The closed research program is a typical example. A paper by AI researchers explains why: these models are trained and scored in ways that reward a confident answer over admitting they do not know. For a college's programs, policies and deadlines, a confident wrong answer can cost an applicant.
It does not really know the student. A chatbot works mainly from what is typed into it. Some now remember details from past chats, but that is not a full record of the courses a student has taken, what they care about and where they are applying, so its advice can still suit almost anyone.
Seven safeguards to look for in any AI tool
An AI tool that is safe for college applications has rules about what it will and will not do. These seven cover the risks above. They apply to any tool your student might use, and the second column shows how Pursiva, an AI counselor built for admissions, handles each one.
| Safeguard to look for | How Pursiva does it |
|---|---|
| It does not write the essay | The Essay coach asks questions, points to what the student has written, and suggests corrections one at a time. It will not produce a draft, a paragraph or an opening line, even when asked |
| It does not write recommendations | Its Recommendations module helps the student choose whom to ask and what to share with them. It never writes a letter as if it were the teacher |
| It does not invent anything about the student | Every statement it makes about a student traces back to something the student entered |
| Its facts about colleges come from checked sources | College facts come from sourced records, not from the AI's memory |
| It does not predict scores or admission chances | It never predicts a test score or a chance of admission |
| It does not rely on the AI to remember deadlines | Deadlines and tasks are kept by the system itself, so they cannot be misremembered |
| It limits the personal information it shares with the AI model | It sends only what a task needs, and leaves out family financial details unless they are required |
With those safeguards in place, the essay from the opening could not have happened. The words would have been the student's, and the program would have been checked against a record.
More than safety: one story across the whole application
Safety is the floor. The larger difference is what a counselor can do once it knows the student.
Pursiva keeps one profile and one application theme (Pursiva calls it a Core Narrative) for each student, and every part of the product works from them. Advice on courses, activities, essays and the college list all points in the same direction. The result is an application that reads as one person, which is the hardest thing for a family to achieve by asking a chatbot one question at a time.
Choosing a tool
Whichever tool your student uses, the seven safeguards above are a fair test of it. A tool built for admissions should be able to say how it handles each one.
Sources
AI guidance for college admissions, with the safeguards built in.
$49.90 a month, every module included. Cancel anytime.
Get started