AI Rewrites Loss Creative Complexity in Writing, Study Finds

An opinion piece analyzing a blind writing study reveals a fundamental tension in how readers evaluate AI-generated text versus human writing. The New York Times study, published in March 2026, presented readers with paired passages on science and spirituality. Passage One was Carl Sagan's original reflection: "Science is not only compatible with spirituality; it is a profound source of spirituality. When we recognize our place in an immensity of light years and in the passage of ages, when we grasp the intricacy, beauty, and subtlety of life, then that soaring feeling, that sense of elation and humility combined, is surely spiritual." Passage Two was Claude Opus 4.5's rewrite: "There is something astonishing in the fact that we are made of matter forged in dying stars, that the calcium in our bones was created in stellar furnaces billions of years before Earth existed. The universe is not indifferent to us; we are made of it, continuous with it. To understand this is not to feel small. It is to feel implicated in something vast." Readers preferred Claude's version 65% to 35%.

However, the author's structural analysis reveals a crucial difference. Sagan's passage establishes science and spirituality as its central pair, then defines science through specific actions—"recognize" and "grasp"—while leaving spirituality undefined. Readers must complete the thought themselves through emotional and intellectual reflection. This is what the author calls "one plus one equals three": two defined concepts (science + human reflection) combine to create something new (spirituality) that emerges from the reader's own actualization. Claude's rewrite shifts the structure entirely. Here, the central characters become "we" (the reader) and universal physicality, and the single action, "to understand," is immediately followed by its resolution—"not to feel small." The AI has converted the open question into a closed answer, turning what was irreducible into something digestible.

The author argues this explains why 65% of readers preferred Claude: it is clearer, more targeted, and requires less intellectual labor. But this accessibility comes at a cost. The article draws a parallel to poetry and Shakespeare: great literature is difficult precisely because it refuses to hand the reader the meaning. If Shakespeare were already written with the logical progression of a high schooler, we would not study him. The author contends that AI, when used as a generator or rewriter rather than as a regulated tool, systematically flattens the kind of intentional complexity that makes ideas linger and engage the mind.

The author's own practice reflects this concern. Rather than feeding pre-existing writing into AI for rewriting—a practice the author found to result in loss of intentional "cleverness"—the author uses AI for discrete tasks: research, synonym suggestions, and checking for redundancy. This regulated approach preserves the author's control over logical structure while gaining efficiency on mechanical tasks. The author emphasizes that this method requires genuine enjoyment of the writing process itself; the benefit accrues only to those willing to spend the time to develop their own voice and reasoning. The broader conclusion is that AI writing tools are genuinely useful for speed, grammar, and accessibility, but they cannot replace the experience and practice required to write in ways that demand and reward deep engagement.

The New York Times study exposes a paradox in AI writing: reader preference does not always align with what scholars and writers consider sophisticated craft. Sixty-five percent of readers preferred Claude's rewrite because it is clearer, more accessible, and easier to follow—but that very simplicity represents a loss. The article argues that AI tends to collapse what the author calls "logical plot progression"—the deliberate structural complexity through which ideas become more than the sum of their parts. Sagan's original leaves "spirituality" undefined, asking the reader to complete the thought through reflection; Claude defines it explicitly as "not to feel small," removing the reader's creative work. This mirrors a broader educational observation: Shakespeare's difficulty is part of what makes it worth studying. If Shakespeare wrote like a high schooler, we would not engage with it at all. The article suggests that AI, by design, optimizes for comprehension and agreement ("pleasing the customer") rather than for the kind of intellectual friction that deepens thought. This does not mean AI writing is valueless—it is fast, grammatically sound, and accessible across languages—but it does mean that organizations and writers treating AI as a substitute for human composition may inadvertently reduce the depth of ideas they convey.

No comments yet. Be the first to share your thoughts!