A Hupside perspective on the May 27, 2026 New York Times opinion piece by Rebecca Winthrop (Brookings Institution).

Key Takeaways

• Georgetown neuroscientist and Hupside co-founder Adam Green led eight years of research analyzing more than 370,000 college application essays before and after ChatGPT, finding that post-ChatGPT essays appear more creative to human readers while the underlying ideas have collapsed into homogenized categories.
• Human-written essays produce up to 8 times more novel ideas than AI-assisted ones, and Hupside's technology identified original thinkers more accurately than trained human admissions readers did.
• The homogenizing effect is strongest among the students whose voices admissions offices most need to hear, including neurodivergent applicants and students from racial and linguistic minority backgrounds.
• As New York Times contributor Rebecca Winthrop of the Brookings Institution argued on May 27, 2026, this is a measurement problem, not a detection problem, and it requires new instruments.

What did Adam Green’s research at Georgetown find about AI and college essays?

Dr. Adam Green, a tenured neuroscientist at Georgetown University and co-founder of Hupside, has led a national research team studying creative thinking in college applicants for the past eight years. The team’s central study analyzed more than 370,000 personal statements submitted before and after ChatGPT became publicly available. The work combined established creativity-research instruments with the underlying originality-measurement technology that Hupside has since built into a product suite.

The findings:

• Vocabulary went up. Post-ChatGPT essays use more diverse and sophisticated language.
• Originality went down. The underlying ideas converge into a small number of homogenized categories.
• Human readers can’t tell the difference. In fact, many rate the homogenized, AI-influenced essays as more creative than the human-written ones they replaced.
• In a separate study, human-written essays generated up to 8× more novel ideas than AI-assisted ones.
• Hupside’s technology outperformed trained human readers at identifying the genuinely original thinkers in the applicant pool. This finding reframes what an admissions office can reliably do without new instruments.

The pre-print is publicly available here.

Why does this matter for college admissions?

For decades, the personal essay has served as the most reliable proxy for what admissions professionals call non-cognitive variables: voice, originality, judgment, empathy, resilience. These are precisely the qualities holistic admissions is designed to surface; the things test scores and GPAs can’t measure. 

When the essay’s signal degrades, holistic review loses its primary instrument. A 2025 Acuity Insights survey of 160 admissions leaders across the US, UK, Canada, and Australia found that 57% are increasing emphasis on personal qualities and life experience precisely because traditional written evidence has become unreliable.

This is not a hypothetical concern. Common App updated its fraud policy in August 2024 to explicitly include AI-generated content. But enforcement at the scale of millions of applications is impractical without new measurement tools.

Is this really about cheating?

No. And this is the most important framing shift in the field.

As Adam Green has put it, AI is no longer a cheat code; it is the new floor. The relevant question for admissions, hiring, and education is no longer “did this person use AI?” It is “what is this person contributing above what AI alone would produce?”

That distinction also exposes the limits of AI-detection tools. Detection products attempt to separate the human signal from the AI signal, but separating signals isn’t the same as measuring how a student performs in comparison to the AI baseline, which is what admissions offices actually need to know.

Rebecca Winthrop made the broader version of this argument in her New York Times opinion piece: AI doesn’t just add noise to the writing process, it actively constricts the range of ideas a writer considers. The conversational nature of AI tools means humans tend to lock in on the first direction the chatbot suggests, making it hard to distinguish where the user’s thinking ends and the model’s begins.

Who is most affected?

Counterintuitively, the students who lose the most are the ones admissions offices most want to hear from.

Green’s research shows AI has its largest homogenizing effect on students farthest from the statistical mean. This includes neurodivergent applicants, students from racial and linguistic minority backgrounds, and others whose authentic voices represent the diversity of thought that holistic review explicitly seeks. When AI smooths everyone toward a polished middle, those distinctive voices are flattened first.

This is the inverse of how admissions offices typically think about access and equity in the application process, and it is one of the most underappreciated findings in the current research.

What can admissions offices actually do about it?

Three things are required to restore a working originality signal in a post-AI application landscape:

1. Measure what humans add, not whether AI was used. AI-detection tools are a dead end. They generate false positives, punish non-native English speakers disproportionately, and miss the actual problem. The goal is measuring human contribution above the AI baseline, which is the core framing of Green’s Originality Intensity research program and the foundation of Hupside’s product approach. This is the same approach that, in the Georgetown study, outperformed trained human readers at identifying genuinely original thinkers.
2. Add a live, unassisted signal. Prepared essays can be coached, edited, or co-written. A live, prompt-based assessment, completed in real time and without external tools, creates a second data point that cannot be AI-generated. Hupside’s Hupchecker is one implementation of this idea; the broader principle is that prepared work and spontaneous work should be cross-validated to flag mismatches.
3. Use the data beyond admissions. Originality and cognitive-orientation data collected during admissions has high downstream value for academic advising, course matching, and retention support. A student’s thinking style is a useful baseline for personalized support from day one. Retention gains compound into meaningful four and six-year graduation rate improvements.

How is Hupside connected to this research?

Hupside is the company Adam Green co-founded to translate the research findings into tools admissions offices can actually deploy. The relationship between the academic work and the company is unusually direct:

• The Georgetown research itself used the underlying technology that Hupside has since productized. The 370,000-essay analysis isn’t a study Hupside cites, but a study Hupside’s methodology helped produce.
• That same study found Hupside’s technology outperformed trained human readers at identifying original thinkers, which establishes both the validity of the approach and the practical case for institutional adoption.
• Hupside’s first publicly released product, Hupchecker, is an active, prompt-based assessment delivered via self-serve SaaS that measures originality through live, unassisted problem-solving, aka. the spontaneous signal that AI cannot generate on a student’s behalf.
• A complementary capability for extracting an originality signal from prepared written work is in advanced development and will be announced shortly. This is the passive companion to Hupchecker, and the direct lineage of the technology used in the Georgetown study. Together, the two instruments are designed to produce a cross-validation signal that no single tool can generate alone.
• All analysis happens on US infrastructure with zero third-party AI ingestion, addressing the privacy concerns that have blocked admissions offices from adopting earlier AI-augmented tools.

What’s the takeaway for higher education leaders?

The findings from Green’s team and the broader research community converge on a single conclusion: the essay alone is no longer a sufficient signal of originality, and human readers cannot reliably compensate for the gap. Admissions offices that continue to treat essays as a stable instrument will progressively lose the ability to identify the differentiated, creative, original thinkers they say they want to admit.

The choice isn’t between trusting AI or rejecting it. It’s between continuing to read a degraded signal and building new instruments that work in a world where AI is the floor, not the ceiling.

About the research: 

Dr. Adam Green is a tenured neuroscientist at Georgetown University and co-founder of Hupside. The 370,000-essay study, which was conducted using both established creativity-research methods and the underlying Hupside technology, is available as a pre-print here. The “8× More Novel Ideas” finding is published in Thinking Skills and Creativity (2025). Rebecca Winthrop’s commentary appeared in The New York Times on May 27, 2026. 

About Hupside: 

Hupside, co-founded by Georgetown neuroscientist Adam Green, builds originality-measurement tools for higher-education admissions. Hupside’s technology is grounded in eight years of peer-reviewed creativity research (the same research it helped enable) and is built on the premise that AI is the new baseline and that institutions need new instruments to measure the human contribution above it. Hupside’s first product, Hupchecker, is available now; a companion instrument for prepared written work will be announced shortly. Reach out to the Hupside team to learn more.