Higher Ed Perfected the Language of Prestige. AI Search Throws It Away.

Updated August 28, 2026: I clarified what the Blanding corpus measures and revised the AI-search argument to distinguish what the evidence shows from what it suggests.

A few months ago, I built an experiment called Blanding. Paste in a higher-ed web address, and it reads the homepage and several key pages, evaluates the writing and positioning, counts the clichés, and returns a score out of 100. When I published the results from the first 250 sites, the finding was blunt: higher education has a sameness problem, not a writing problem.

People kept submitting sites. By August, the dataset had grown to 887 web properties, including institutional sites, departments, programs, centers, and microsites. I expected a larger, more varied sample to soften the result.

It didn't.

The curve barely moved

The median Blanding score in a frozen August 26 snapshot of 887 higher-ed web properties was 48 out of 100. It was 50 when the dataset held 250 sites. Just over 75 percent scored below 55, and only six scored 80 or higher. Four of those six were focused departments, centers, or microsites, places that can clearly state who they are and what they do. The curve barely shifted as the sample more than tripled.

Methodology note: The corpus is self-selected, not nationally representative. Records reflect different audit dates and page counts, and scores may change when a site is re-audited.

That matters, but not because this distribution proves every site arrived here for the same reason. It cannot. The sample is self-selected, and the scoring model is designed to detect certain kinds of sameness. Still, the stability is consistent with something I have watched for decades: higher education’s incentives keep pulling very different institutions toward the same language.

The writing is only part of the question. I am more interested in why so much of it becomes generic in exactly the same direction.

Why so many schools arrive at the same language

In my experience, many schools sound alike because higher education gives them powerful incentives to sound like respected peers. A university homepage is rarely written for one seventeen-year-old. It often passes through the president, provost, deans, enrollment, advancement, faculty, athletics, legal, and communications. A concrete sentence is also a sentence someone can object to, and in a consensus process, the things people might challenge are often the first to go.

What survives is the language no one will fight about: world-class, academic excellence, a vibrant and diverse community, transformative experiences, preparing tomorrow’s leaders.

The language isn't false. That is part of what makes it so hard to cut. It is just too broad to create preference.

Underneath the committee is something harder to escape. Much of higher education still sells prestige. Prospective students, parents, donors, trustees, legislators, and faculty recruits may want very different things, but the signal of “elite” is broadly recognizable. That does not mean prestige is every institution’s real value proposition. Access, affordability, faith, geography, program strength, support, and outcomes may matter far more. Still, institutions reach for language that signals status broadly, and those words are, by definition, the words everyone else is already using.

Sociologists named this drift more than forty years ago. Paul DiMaggio and Walter Powell called it institutional isomorphism: organizations in a field grow more alike because they adopt the conventions and signals of the peers they believe are winning. Prestige hierarchies long predate modern rankings. Rankings quantified and amplified them, especially through peer-reputation measures. When institutions are judged partly by what peers think of them, the safest posture can be to resemble a school those peers already respect. The result isn’t always outright imitation.

It is gradual convergence.

The same declarations, the same photography, the same proof points, the same visual signals of seriousness and ambition.

How feedback turns distinction into beige

The sanding doesn’t only happen in the first draft. It can come from the way the institution asks for feedback. My friend Richard Banfield has written about what he calls the tyranny of feedback: the way accumulating reactions from more and more people gradually makes a piece of work less like itself. Every suggestion sounds reasonable on its own. Add this audience. Make sure this department feels represented. Soften that phrase. Put this priority back in. Good work needs input. It also needs clear decision rights. Contributors and approvers are not the same thing.

But hundreds of reasonable suggestions don’t add up to a stronger idea. They average into something no one dislikes and no one remembers. Surveys, focus groups, cabinet reviews, committee markups: each note can be defensible alone, and together they can be fatal to a point of view.

I'm a painter, and in the studio this is what makes mud: every color mixed together, nothing left out. A bland homepage is an overworked canvas. No one was willing to leave anything out, so everything had to be included, and everything had to carry the same weight. Nothing could dominate, because dominance implies a choice. Nothing could be omitted, because omission feels dangerous. The result isn't fullness. It is the absence of hierarchy.

No selective reduction, no constraint, no point of view. Just the belief that more is more accurate, and therefore safer. When every virtue is presented at equal volume, the differences get harder to see.

It is hard to form a strong preference among things you cannot distinguish.

AI changes the cost of sounding alike

For years, institutions could absorb more of the cost of this sameness. Traditional search usually returned a page of links. A student could browse several sites, read program pages and student stories, and slowly uncover the distinctions the homepage failed to express. Well-known schools also carried accumulated authority through links, press, rankings, and reputation, which helped keep them visible even when their language was generic.

That environment is changing. A growing share of students now bring conversational AI into the process. In a 2026 survey of more than 5,000 high school students, EAB found that 46 percent used tools such as ChatGPT during the college search, up from 26 percent in spring 2025. Eighteen percent said information surfaced through AI had led them to remove a college from consideration. AI is now part of the search. It has not replaced Google, college websites, counselors, social media, or campus visits.

A student may still type “best engineering colleges” into Google and open ten tabs. But a growing number also ask: what are some smaller colleges with strong engineering programs, close faculty relationships, access to nature, and a collaborative rather than competitive culture? To answer, the system has to find evidence that an institution fits.

Now read your homepage as evidence. “World-class academic excellence” may signal prestige to a human reader, but it offers almost nothing that distinguishes one institution from another. No specific fact matches the student’s question. Nothing concrete to extract. Nothing worth repeating back.

This doesn’t mean reputation has stopped mattering. A search-connected AI system may still encounter a well-known institution’s reputation through rankings, Wikipedia, news, program pages, and decades of accumulated authority. Prestige still travels. Generic language does very little work as a differentiating signal. When a system needs to explain why one school fits a particular question, it may route around the cliché and fill the gaps with information from deeper pages and outside sources. The institution may still enter the conversation. But it has less control over the story being told.

Generic language has always made an institution harder to remember. AI adds another cost: it may throw the cliché language away and go looking elsewhere for something specific enough to use.

AI does not fill every gap cleanly

Belmont University offers a useful example. In February, a reporter at Inside Higher Ed asked ChatGPT to recommend Christian colleges in the South, then narrowed the request to a student interested in music business. Music business is Belmont’s largest undergraduate major. Belmont did not appear in that run.

Belmont’s marketing team said it got better results in its experiments. When the reporter tried Claude, it asked follow-up questions about major, institution size, and Christian denomination. After the reporter supplied answers matching Belmont’s profile, Claude included Belmont among several options.

Belmont’s marketing team had also identified a possible terminology gap: its materials used “Christian” and “Christ-centered” interchangeably. The team wondered whether that inconsistency made the university less likely to appear when a student searched using only one of those terms. Maybe that contributed to ChatGPT’s omission. Maybe it didn’t. One query cannot establish the cause, especially when the university’s own tests produced different results.

That is what makes this difficult. AI recommendations can vary from one system, prompt, or moment to the next. If a school does not appear, there may be no abandoned form, no falling conversion rate, no clean analytics trail. And if it does appear, the answer may be built from sources the institution does not control.

Your homepage can't be an internal consensus document

At many institutions, the homepage is managed by making sure everyone inside is reasonably satisfied. It is reviewed, negotiated, expanded, and approved. That is one reason so many homepages become collections of statements no one can object to.

But the homepage is also one of the most visible, most linked, most authoritative pages the institution controls. It should not try to contain everything. It should establish a clear enough sense of the place that a prospective student, or a system trying to help one, can begin to answer three questions: What is this school? What is especially true here? Who is likely to thrive here?

The homepage doesn’t determine an institution’s entire AI identity. Depending on the product and the query, an answer may draw from program pages, rankings, news coverage, Wikipedia, and other third-party sources. That makes specificity and consistency across the site more important, not less. The pages a school controls should tell a coherent, specific, defensible story. Otherwise, a search-connected AI system may fill the gaps with whatever it can find.

Specificity has to live in two places

Specificity has to live deep in the site: on the page about the lab devoted to one unusual line of research, the program with a requirement few competitors have, the faculty doing something genuinely distinctive, the traditions and places that make the institution itself. It also has to live at the top, in language clear and consistent enough that a student can remember it and an AI system has something specific to use.

Consider Swarthmore. In the July snapshot, it scored 49 on Blanding, with a strategy subscore of 45. Its Swarthmore Forward strategic-plan site describes a future in which the community thrives, values are lived, and impact reaches beyond campus. It is admirable. It is also language that could appear on almost any college site in the country.

But Swarthmore also runs a century-old, Oxford-inspired Honors Program. More than 100 external examiners participate each year, evaluating seniors’ work, including through oral examinations. One description is a universally agreeable aspiration. The other tells you something specific about Swarthmore and the kind of student who may thrive there. Specificity isn’t a cleverer tagline. It is uncovering the facts that make the institution genuinely different.

Specificity is now infrastructure

The work is to make decisions institutions often postpone because they are hard.

What are we, specifically, that no one else can credibly claim? Who are we especially right for? What happens here that would not happen the same way somewhere else? What are we willing to emphasize, and what are we willing to leave out? Those answers have to live everywhere: in academic pages, faculty stories, outcomes, admissions language, metadata, and the homepage itself. Concrete enough for a student to remember. Consistent enough for a machine to recognize. True enough for the institution to defend.

This isn’t only a higher education problem. Every field has its “world-class.” Technology has “AI-powered solutions,” law has “trusted advisors,” healthcare has “patient-centered care,” agencies have “full-service.” Reach for the language that makes you sound like everyone else, and neither a person nor an AI system has much to use as evidence that you are different.

AI may still find what makes you different. But if it has to leave your homepage and piece the story together from program pages, rankings, news, Wikipedia, and other sources, you have given up some control over what gets found, repeated, and believed.

The institutions aren’t bland. Their most distinctive facts are buried. Search-connected AI may go looking for them. The question is whether it finds the story you meant to tell.

You can run your institution’s website through Blanding in about thirty seconds. The score isn’t the answer. Treat it as triage, not a verdict. It does not measure institutional quality, enrollment outcomes, or whether an AI tool will recommend you. It points to places where the language may be generic, proof may be thin, and real decisions may still be waiting.