Guide 8 of 10
Published by InGenius Prep · Figures checked 27 August 2026
What universities have said, what the tools are useful for, and where they fail.
Two things are true at once. Most selective universities have now published something about artificial intelligence and application essays, and most admissions offices have not finished deciding what they think. A Kaplan survey of 220 admissions officers, fielded in July and August 2025 and released that November, found 68 percent had no policy at all on students using generative AI to write essays. Thirty percent banned it. Two percent allowed it. That gap is the real story. This guide covers what specific universities have put in writing, what AI does well in an application process, and what it does badly.
The published policies are narrower and more specific than the general debate suggests. Four examples, quoted from the schools themselves:
Read those four together and a line appears. AI as a thinking aid is broadly tolerated and sometimes encouraged. AI as a text producer is prohibited. Almost every published policy sits on that line, and the schools that have written nothing are mostly silent rather than permissive: in the same Kaplan survey, 50 percent of officers reported an unfavorable view of applicants using AI, against 14 percent favorable.
The numbers also show where the tolerance is. For brainstorming, 27 percent of officers said their office allows it and 4 percent ban it. For feedback on drafts, 21 percent allow and 5 percent ban. For writing the essay, allowance collapses to 2 percent. Check each of your child's schools directly, on the admissions site, in the cycle you are applying. Policies published in 2023 have already been revised.
The useful applications are administrative and preparatory, not compositional.
Voice. This is the failure admissions officers name first. One respondent in the Kaplan survey described the tell plainly: the essay writing does not match the writing in other sections of the application, which prompts the reader to question how well the student actually writes. An application is one document with several authors' worth of evidence in it, and inconsistency is visible.
Specificity. Language models produce the average of what has been written before. Admissions reading rewards the opposite: the detail only one applicant could supply. A model can write a competent paragraph about learning resilience through cross country. It cannot know that the student ran the last mile of the district meet with a stress fracture and then quit the team, which is the essay.
Factual reliability. Models generate text that reads as confident whether or not it is correct. Course names, professor names, program requirements, and deadlines invented by a model have all landed in submitted applications. Anything school-specific must be checked against the school's own page.
OpenAI withdrew its own AI text classifier on July 20, 2023, citing its low rate of accuracy. Published figures for that tool were a 26 percent true positive rate against a 9 percent false positive rate. A 2023 study in the journal Patterns found that widely used detectors consistently misclassify writing by non-native English speakers as AI-generated while classifying native writing correctly.
Both facts cut in the same direction for a family. Detection is weak enough that no student should assume the tool will catch them, and unreliable enough that a student who wrote every word can still be flagged. The practical protection is a documented drafting history: dated files, revision history in a word processor, notes. Keep it.
Firms use software internally, and the reasonable version of that is narrow: tools working on the firm's own data and operations, such as scheduling, matching students to counselors, tracking requirements across a caseload, and maintaining internal research on program deadlines. The point of that automation is hours. Administrative time a counselor does not spend on records is time spent with the student. InGenius Prep, which publishes this site, holds the same line: those tools never touch an applicant's writing.
Three questions, and get the answers in writing:
A firm that cannot answer the first question with a flat no is telling you something. So is a firm that markets AI-assisted essay writing as a feature. Under Yale's published policy, that output is fraud.
Next guide
The firms that run the courses on your child's transcript, and the question to ask them.