Educational Technology and Change Journal

By Jim Shimabukuro (assisted by Gemini and Claude)
Editor

Introduction: The rise of generative Artificial Intelligence (AI) has forced a critical re-evaluation of what constitutes a “good” student-written essay. Traditional benchmarks like organization, focus, development, and formal correctness are now easily met by AI, rendering them insufficient as definitive markers of student learning and unique intellectual effort. The criteria that remain stubbornly human—originality of insight, genuine personal voice, and nuanced engagement with lived experience—are now paramount. I asked Gemini and Claude to examine articles that address these personal criteria to provide a glimpse into the future of writing pedagogy and the outer limits of AI’s current capabilities.

Gemini

In “AI-Proof Your Assignments: 5 Strategies to Prevent Cognitive Offloading in Higher Education” (Packback Co., 7 Oct. 2025), author Selina Bradley directly addresses the need to design assignments that make AI generation difficult by requiring a personal, authentic connection.

Authenticity is the enemy of automation. AI can’t fake lived experiences, personal reflections, or genuine connections. As Keith Hollowell, a Literature Professor at Virginia Commonwealth University put it, “I want to bring back humanity into their writing. And so much of their humanity is their student voice.”

This quote crystallizes the belief that a student’s genuine personal voice and their lived experience are the core, un-automatable qualities of a valuable essay. Hollowell’s focus on “humanity” in writing suggests that the essay’s worth is not in its formal perfection but in its capacity to reveal the unique self behind the words.

I believe AI can currently simulate a personal voice and may soon generate credible-sounding “personal reflections” by constructing narratives based on vast datasets of human experience. However, the requirement for a genuine personal voice and an authentic connection to the student’s actual lived experience remains a significant barrier. AI does not have a self, and this lack of internal reality will likely keep genuine authenticity beyond its reach for the foreseeable future. A clever, personalized prompt given by an instructor, followed by a human conversation or oral defense, would easily expose the lack of actual lived experience.

In “Authentic or Artificial? Identifying AI in Student Writing” (ETS Global, Feb. 2025), the author highlights the inability of AI to provide unique depth of thought, even when it is grammatically correct.

AI can summarize and report facts, but it cannot offer the personal insights or unique perspective that a student would naturally bring to their writing… If a paper feels like it could have been written by anyone or lacks reflection on the student’s experiences or learning, that is a red flag.

This excerpt clearly specifies two of the core human criteria: originality of insight and nuanced engagement with lived experience (through reflection). The call for a “unique perspective” that a machine, which draws on existing data, cannot provide, re-establishes the value of an essay that moves beyond mere information synthesis. A good essay, in this view, is one that clearly belongs to one particular student because of the way it processes information through their unique perspective.

AI is excellent at synthesizing and making connections based on its training data, which can often mimic originality. However, true originality of insight—the leap of thought that reconfigures known facts into a genuinely novel, non-obvious understanding—is still a high bar. As AI models become more sophisticated and their “world knowledge” expands, their ability to produce insightful connections will increase. Yet, the requirement for the insight to be tied to the student’s specific reflection on their learning process or life keeps it distinctly human. AI is rapidly approaching the ability to meet the general requirement for “insight,” but a well-designed assignment that demands a unique intellectual twist remains a temporary outer limit.

An academic article titled “On the essay in a time of GenAI,“ published in Educational Review in 2025, proposes a fundamental shift in how the essay is viewed, moving away from a standardized product. The authors argue that the essay develops crucial capacities.

Specifically, we attempt to show that such essays develop three crucial intellectual capacities that no other assessment format can cultivate as effectively: the ability to think productively within uncertainty, self-regulated learning through the recursive relationship between writer and text, and what Heidegger calls [the] ‘dwelling’ in a subject.

Although framed in academic terms, “self-regulated learning through the recursive relationship between writer and text” and “dwelling in a subject” point directly toward the kind of authentic, complex engagement that leads to genuine personal voice and nuanced engagement with lived experience. This suggests the “good” essay is defined not by its final polish but by the trace of the struggle and exploration within the text—the process of the writer thinking through a problem.

This is perhaps the most difficult challenge for current AI. An AI system, by its very nature, is designed to produce a product—a complete, coherent answer—not a record of productive uncertainty or a “recursive relationship” of self-discovery. The AI’s process is not a journey of self-regulated learning; it is a statistical calculation. While AI can simulate human confusion or thought-in-progress, it cannot experience the internal cognitive work that this criterion describes. Therefore, if the evaluation of a “good” essay shifts to the evidence of the author’s intellectual process and personal struggle, as this article suggests, this specific requirement will likely define the outer limits of AI’s capabilities for a very long time, as it ties the criteria for goodness to the act of learning itself.

---

Claude

The question you raise strikes at the heart of a profound tension in contemporary education. As AI systems become increasingly sophisticated at producing technically proficient prose, educators and scholars are being forced to articulate what, precisely, makes human writing valuable—and whether those qualities represent AI’s outer limits or merely its current boundaries.

Among the articles that address these personal criteria, several stand out for their clarity in defining what AI cannot yet replicate. In a 2024 article published in Frontiers in Education titled “Exploring the boundaries of authorship: a comparative analysis of AI-generated text and human academic writing in English literature,” researchers Forough Amirjalili, Samira Neysani, and Elham Nikbakht conducted a systematic comparison between student-authored and AI-generated academic essays. Their findings were unequivocal: “the STUDENT text exceled in accurate citations, varied voice usage, rhetorical engagement, and maintaining a nuanced authorial presence. GPT, while generating content, struggles with accuracy in quotes, factual errors, and lacks the depth and authenticity exhibited by the student’s essay.” This articulation is particularly valuable because it moves beyond vague assertions about “authenticity” to identify specific textual features—varied voice usage, rhetorical engagement, nuanced authorial presence—that distinguish human from machine writing. The authors conclude that AI-generated text “falls short in replicating the depth and individuality inherent in human academic writing.”

Can AI meet these requirements? In the short term, probably not. The researchers’ emphasis on “nuanced authorial presence” suggests something more than surface-level stylistic variation—it implies a consistent sensibility that emerges from sustained engagement with ideas and reflects genuine intellectual wrestling. Current AI systems, which generate text through statistical pattern matching rather than understanding, struggle to maintain this kind of coherent presence across extended passages. However, I’m skeptical that these limitations are permanent. As AI systems become more sophisticated at modeling long-range dependencies and maintaining consistent “personas,” the gap may narrow considerably.

A more philosophically rich account appears in a 2024 article by Alessandra Giugliano titled “How to Preserve Your Academic Voice While Using AI Writing Tools,” which features extensive interviews with Dr. Michael Meeuwis, Professor of English at the University of Warwick. Meeuwis describes writing as fundamentally meditative: “slowing down allows you to ‘figure out what you think’ before shaping it for an audience. This reflective space, he argues, is where originality begins.” He continues: “When students let AI take over entirely, they bypass the slow work that develops personal ‘taste’—the critical judgement needed to decide what to include, what to discard, and how to refine an argument.” This formulation is compelling because it locates the value of writing not in the product but in the process—in the cognitive work of deliberation, selection, and refinement that shapes both the text and the writer’s thinking.

This represents a more substantial challenge for AI. Meeuwis is essentially arguing that the value of student writing lies in its role as “a mental gym,” a site where intellectual muscles are developed through resistance and effort. If we accept this framing, then AI-generated writing isn’t merely inferior—it’s categorically different, serving no developmental function for the student. But does this mean AI will never be able to produce writing with genuine originality of insight? Not necessarily. It means that even if AI produces brilliant insights, using it to write student essays would be as pointless as using a robot to lift weights at the gym—it defeats the purpose of the exercise entirely. The limitation here is pedagogical and developmental rather than technological.

A third perspective comes from Nancy Sommers, whose work on revision and student writing has been foundational in composition studies. In a 2015 article in Teaching English in the Two-Year College titled “Living Composition,” Sommers offers this striking metaphor: “Voice is that elusive category we talk about with students—’find your voice,’ we urge, as if they left it somewhere, in a dresser drawer, perhaps, as if they could purchase it on Amazon. But there is no lost and found drawer for voice, no way to shop for it, or stumble upon it. It is something you have to write your way into, something that takes practice and play, attempt after attempt, as you arrange the alphabet into comfortable shapes and sounds, listening for your own idiosyncratic take on the world.” This passage beautifully captures what might be called the “earned” quality of authentic voice—the sense that it emerges through sustained practice and experimentation rather than being adopted wholesale or generated on demand.

Can AI develop this kind of earned voice? This is where things become genuinely murky. Sommers’s emphasis on “practice and play, attempt after attempt” suggests that voice is not a static feature but an emergent property of a long developmental process. Current AI systems don’t “develop” in this sense—each generation is independent, drawing on patterns in training data but not building cumulatively through practice. However, more sophisticated AI architectures that incorporate memory, learning across sessions, and iterative refinement might eventually approximate something like this developmental arc. Whether such a system would possess genuine voice or merely simulate it convincingly is a philosophical question that may ultimately prove unanswerable—or irrelevant.

A fourth article worth considering appeared in July 2025 in The Hechinger Report by Olivia Sanchez, titled “Students try using AI to write scholarship essays—with little luck.” The article quotes Will Geiger, a former college admissions officer who estimates he read about 10,000 application essays before ChatGPT’s release. Geiger reports that after 2022, essays began to feel “less and less like they had been written by 17- or 18-year-olds,” noting more “hyperorganized five-paragraph essays; more essays that were formatted as a letter to someone; and certain examples and words being used over and over again.” The article quotes one expert’s concern: “If we get to a point where colleges cannot confidently determine [its] authenticity, then they may abandon it entirely.” This highlights a crucial point about lived experience: AI can describe universal human experiences, but it struggles to render the specific, granular, idiosyncratic details that signal genuine personal history. A student writing about overcoming adversity will include sensory details, contradictions, tangents, and the kind of “messy” specificity that AI, trained on countless similar narratives, tends to smooth into generic platitudes.

This may represent AI’s most durable limitation. The specificity of lived experience—the way a particular morning smelled, the exact words someone said, the illogical but deeply felt associations that shape how we interpret events—is genuinely difficult to fabricate convincingly at scale. AI can generate “personal” narratives, but they tend toward the archetypal rather than the particular. However, as AI systems gain access to more personal data and develop better models of individual users, even this barrier may erode. An AI trained on years of someone’s writing, photos, and communications might eventually generate highly specific “personal” narratives that are indistinguishable from authentic memory.

Do these personal criteria define the outer limits of AI’s capabilities? In my assessment, they represent significant current limitations but not necessarily permanent boundaries. The meditative, developmental dimension that Meeuwis emphasizes may prove most durable—not because AI can’t produce original insights, but because the pedagogical value of writing lies in the struggle itself. The nuanced authorial presence that Amirjalili and colleagues identify is a harder question; current AI falls short, but future systems may close that gap. Sommers’s “earned” voice through iterative practice presents philosophical puzzles about whether simulation and genuine development can be meaningfully distinguished. And the specificity of lived experience, while currently a reliable marker of human authorship, may not remain so as AI systems become more personalized.

What seems increasingly clear is that the question “Can AI write as well as humans?” is giving way to more fundamental questions about what we value in student writing and why. If we value writing primarily as evidence of students’ ability to produce clear, organized, well-supported arguments, then AI is already approaching parity. If we value it as a site of cognitive development, intellectual struggle, and the forging of authentic voice through sustained practice, then AI may never be an adequate substitute—not because it lacks the capability, but because outsourcing the work defeats its purpose. The criteria you’ve identified—originality of insight, genuine personal voice, nuanced engagement with lived experience—may not so much define AI’s outer limits as clarify what we actually want from student writing in the first place.

[End]

Filed under: Uncategorized |