There is an old riddle that asks whether a tree falling in a forest makes a sound if no one is there to hear it. Based on our conversations with admissions executives, I began to wonder if higher education now has a version of that question.
When a family researches a college inside ChatGPT, Claude, or any other LLM, little of that research ever reaches the institution. So, did the institution take part in their decision?
The question is timely and urgent. Last week, admissions professionals from across the country were in Minneapolis for the NACAC Conference, where AI was firmly on the agenda. However, much of the conversation about AI in admissions focused on what an office can do with it internally, such as reviewing applications and drafting outreach. As a recovering CMO, I was struck by a big gap on the demand side of the equation: what are prospective students and their families already doing with AI, and what does that mean for how an institution is represented in conversations it cannot see? Reading applications faster is a worthy goal, but it improves the back end of the funnel while the front end, where families decide which schools deserve an application at all, slips out of view. An institution cannot cut its way to growth. Efficiency protects the budget; enrollment is what funds it. Admittedly, that is hard for any of us in mission driven roles to hear.
The college search often begins inside an AI answer
In EAB's 2026 Parent Survey of 2,515 parents of high school students, 57 percent said they had used AI tools such as ChatGPT to evaluate and compare colleges for their children, and 34 percent said AI surfaced schools they had not previously considered. Further, EAB's head of research described AI as a new gatekeeper in the college search, for parents as much as for teens.
Students are moving in the same direction. The Chronicle of Higher Education reported that 46 percent of students researching colleges used AI tools to do so in fall 2025, up from 26 percent the prior spring - a 77 percent increase in a single semester. About one in five had removed a college from their list because of an AI-generated response.
These figures describe discovery as much as evaluation. When AI introduces a family to a school they had not considered, it is also shaping which schools that family hears about first, and that shaping happens before you had any shot at a campus visit, an email, or a counselor conversation.
This raises a practical question worth pausing on: what actions can you think of that you have taken to explore how your institution shows up in the major engines? For many teams, the answer so far is a handful of informal searches, which is an understandable place to be for a channel that is new. However, in many of our conversations with admissions teams, limited time and limited familiarity with these tools are leaving a serious and growing blind spot.
This doesn’t have to be the case. The rest of this piece is about what a more deliberate approach can look like.
Why your dashboard can look “fine”... while applicant research dramatically shifts elsewhere
A deliberate approach starts with two facts about how these answers work.
The first is that people trust what AI tells them. Pew Research Center tracked real Google searches and found that when an AI summary appeared, users clicked a traditional result about half as often, in 8 percent of visits compared with 15 percent without one. They opened the sources cited inside the summary just 1 percent of the time. For many families, the AI answer is where the research ends.
Admissions offices already know a version of this. For years, a share of applicants have arrived as “stealth applicants,” students who never filled out an inquiry form or opened an email before their application landed. AI search extends that pattern to the research phase itself. A family can form a real opinion about an institution’s cost, fit, and outcomes and leave no trace in the CRM or in web analytics. The analytics are not wrong. They were built to measure a later moment than the one where much of the decision now happens.
Who narrates the value of college now?
The second fact is that AI decides for itself which sources are authoritative. When a parent asks whether a program is the right fit for their child, or whether the degree is worth the cost, a model builds its reply from whatever it can find and read: rankings aggregators, forums, third-party salary sites, and the institution’s own pages. It does not reliably favor what is current. A 2018 “best of” list can be repeated with the same confidence as 2026 data, particularly if the older source is easier to read and summarize.
That matters more while public confidence in higher education is under pressure. Lumina and Gallup found that 38 percent of Americans have a great deal or quite a lot of confidence in higher education, down from 42 percent in 2025 - but encouragingly 93 percent of bachelor’s degree students said their investment was worthwhile.
When the public and the people closest to the experience disagree this sharply, whichever sources the model draws on end up narrating the difference. The more complete and current and LLM accessible the institution’s own information is, the more likely it is to shape that reply - and to be part of the conversations that follow at home.
The most trusted voice is still the institution's own
One finding in the EAB research deserves more attention than it usually gets. Seventy-three percent of parents named colleges' own websites as their top source of information, ahead of direct communications and in-person events. In EAB's discussion of the survey, its head of research added that parents want to hear directly from colleges and are interested in communication written for them. This is a testament to the trust admissions professionals have spent many years cultivating.
Families, in other words, are not looking for a substitute for the institution's voice. They are looking for that voice, in the places where they now ask their questions. The opportunity for enrollment and marketing teams is distribution: carrying the institution's own approved answers, in its own words, into this new territory at the speed families are moving.
Answer engines favor content that resolves a specific question clearly and currently, in a form a model can read, trust, and cite. New conventions are emerging to support that, including the llms.txt file, which gives AI systems a structured summary of who an institution is and where its authoritative content lives. Adoption is early, as it was for many web standards that later became routine, and the institutions that build this foundation now will be ready as the engines mature.
An audit any admissions team can run this month
A useful first pass can be done by hand in an afternoon, and it tends to clarify where the effort should go next.
- Gather the questions families actually ask. Pull the twenty-five or so questions that counselors, the inquiry inbox, and campus tour guides hear most often. Aid, fit, deadlines, outcomes, and cost almost always lead the list.
- Ask the engines the way a family would. Use a private or logged-out window for each one, so your own chat history and saved preferences do not shape the answers. Many AI tools personalize responses to the person asking, and a staff member’s account can return a different picture than a family would see. Put each question to ChatGPT, Claude, Perplexity, etc.,in a parent’s voice and in a student’s voice. For example: families rarely search for “undergraduate admissions requirements.” They ask whether a B student with a part-time job has a realistic chance.
- Open every citation and tally the sources. Note how often each engine draws on the institution's own pages versus aggregators, forums, or rankings sites. Tracked engine by engine, that ratio becomes a simple measure of how much of the institution's story it is telling itself.
- Grade the answers for accuracy and currency. Check tuition, deadlines, program names, requirements, and outcomes data against what is true today.
- Confirm the engines can read the pages at all. Review crawler access, indexability and metadata, structured data, content structure, social and Open Graph tags, and how pages render. Content that appears only after scripts load, or that lives in a PDF, can be harder for AI systems to use. Most of these items are straightforward work for a web team.
- Give each top question one clear, current, citable home. Update or consolidate the pages behind wrong or thin answers. A dedicated page that answers the top questions directly, paired with an llms.txt file pointing to the institution's authoritative content, gives the engines an obvious place to look.
- Run it again on a schedule. Models update and pages change, so a one-time audit ages about as gracefully as last year's viewbook.
Readers of my last Campus Notes piece will recognize step six. The content work an AI guide surfaces from the inside is the same work the open web surfaces from the outside.
AEO is not your full-time job. Being present for the family conversation is.
Conversations with our admissions clients shaped how we built the AEO tool into skilly. The feature was easy for us to add because the knowledge it draws on is the same knowledge their AI admissions guide already holds - allowing us to design toward surfacing the hard parts to our admissions clients in a practical way that is easy to activate and consume.
We know from our work that the hardest part for admissions teams is knowing which questions matter most, leading us to design the feature to start from the questions prospective students actually ask a skilly guide and identify the top twenty-five. They told us checking every engine by hand did not scale, so it regularly queries the major models and tracks how often each one cites the institution's own pages. They told us the technical review was hard to translate into a web team's task list, so it runs a diagnostic across crawler access, indexability, structured data, content structure, social tags, and rendering. And they told us fixes went stale, so it maintains an llms.txt file and a top questions page that stays current as the guide learns.
The real reframe
The question is no longer whether families will ask AI about an institution. Nearly 60 percent of parents already have.
The more useful question is whether the answer they receive is built from the institution's own current, approved information, and whether the institution can see that answer at all. One version can be reviewed, corrected, and measured over time. The other still happens, just out of earshot.
Which brings me back to the forest. The tree will fall either way. What an institution can influence is whether anyone on campus is there to hear it, and whether the voice the family hears is the institution's own. Be intentional about placing yourself in that room.