The Vocabulary Police Don't Read Enough Books

There’s a new genre of internet criticism that has emerged alongside the rise of generative AI, and it goes something like this: someone posts a piece of writing, and a commenter swoops in to announce, with supreme confidence, “This was written by AI. It uses the word ‘delve.’” Or “No human would ever say ‘tapestry of experiences.’” Or “Real writers don’t use ‘furthermore.’” Or, my personal favorite, “It’s not x, it is y.”

Gag.

This has become its own cultural tic. The AI word-spotter, armed with a mental checklist of supposedly machine-generated vocabulary, appointing themselves the arbiter of authentic human expression. The problem? The words and literary devices they’re flagging as artificial hallmarks of machine writing have roots stretching back centuries, appearing in some of the most celebrated texts in the Western canon. What the critics are actually revealing is not an AI’s limitations, but their own.

What Are “AI Words,” Exactly?

The list of supposedly telltale AI phrases has been catalogued extensively by AI detection platforms and content marketers. The most frequent flagged terms fall into five groups: formal transitions (moreover, furthermore, consequently), vague action verbs (leverage, utilize, facilitate), generic emphasis words (crucial, significant, comprehensive), hype phrases (revolutionary, transformative, game-changing), and hedging qualifiers (it can be argued, to some extent).

Atop nearly every list sits the word “delve.” Words like delve, pivotal, robust, and leverage appear so consistently in AI output that they now trigger detectors and signal generic writing to human readers alike; meaning even genuine human work is sometimes flagged as AI-generated, frustrating both students and professionals.

Researchers have noted the statistical reality behind this: words like delve (48 times more common in AI text), tapestry (35 times), and multifaceted (28 times), as well as phrases like it’s worth noting (31 times more common), have become strong AI signals that detectors use to track vocabulary frequency distributions.

The reason these words cluster in AI output has a clear explanation rooted in how language models are trained: AI overuses formal academic transitions that most people never write. These words appeared frequently in training data (academic papers, formal articles, business writing) and received positive reinforcement during the training process because they sound “polished.” The models learned that these words satisfy user expectations for quality.

So far, so reasonable. But here’s where the logic breaks down: the conclusion that because AI overuses these words, these words are inherently artificial or illegitimate is a profound non sequitur. And it requires either a very short memory or very little reading to believe it.

“Delve:” A Word With a Thousand-Year History

Let’s start with the most frequently cited AI tell. “Delve” is treated online as though ChatGPT invented it sometime around 2023. In reality, the word delve derives from the 9th-century Old English word delfan, which itself came from the Old High German word telban. These were all words for digging at a time when the word dig hadn’t been uttered yet. That original meaning of delve has given way to the more common connotation of searching or researching.

The same applies across much of the flagged vocabulary. “Tapestry” as a metaphor for complexity and interconnected experience has been a literary device for centuries and used to describe everything from the social fabric of Victorian England to the moral texture of Tolstoy’s Russia. “Nuanced” comes from the French nuance (meaning shade or tint) and entered English literary criticism in the 18th century. “Pivotal” is straightforwardly Victorian. “Realm” appears in Keats: “much have I travell’d in the realms of gold.”

None of these words arrived with the iPhone.

Formal Transitions: The Cornerstones of the Essay Form

The condemnation of words like furthermore, moreover, and consequently is perhaps the most telling indicator of how little the critics know about the history of written argument. These are not quirks of algorithmic prose. They are the structural ligaments of the essay: a form with a four-hundred-year pedigree.

Michel de Montaigne, who invented the personal essay in the 1570s, and Francis Bacon, who developed the form in English at the turn of the 17th century, are the fountainheads of expository English prose. Formal connective transitions, words that signal logical movement between ideas, are not stylistic tics; they are the architecture of reasoned argument. Writers from Samuel Johnson to George Orwell, from Virginia Woolf to James Baldwin, have used furthermore, moreover, consequently, and thus as tools of intellectual precision. To flag these as machine-generated is to be unfamiliar with the entire tradition of formal essay writing.

AI’s writing style tends to resemble highly structured high-school essays, which is worth noting, but high-school essays didn’t invent these transitions. They inherited them from centuries of academic and literary tradition, and taught them to students because they work.

“Multifaceted,” “Intricate,” “Holistic”: The Vocabulary of Serious Thought

The deeper flagged list includes words like “multifaceted,” “intricate,” “holistic,” “meticulous,” and “nuanced” all of which are treated as signs of soulless machine writing. What they actually are is the vocabulary of careful analysis. These words exist because simple language sometimes cannot carry complex meaning.

Henry James built his reputation on syntactic intricacy and layered qualification. Virginia Woolf’s novels employed stream-of-consciousness narration to dive deeper into the inner feelings and thoughts of her characters, and her exploration of gender roles, individuality, and the complexities of human consciousness helped establish her as a pivotal figure in the evolution of modernist literature. The word pivotal, used there in a scholarly description of Woolf - is that AI-generated? Or is it simply an accurate word?

The issue is not the words themselves. The issue is frequency and context. AI overuses them because it was trained on dense formal writing. But the solution to overuse is not to brand the words themselves as counterfeit. That is like banning metaphor because bad writers abuse it.

The Real Problem: A Generation That Stopped Reading

Here is the uncomfortable truth beneath this whole conversation: a significant portion of the people who confidently declare that a piece of writing “sounds like AI” have not read enough literature to know what sophisticated human writing actually looks like. This is a systemic failure that deserves honest reckoning.

In the past decade, the number of students passionate about literature, history, and the arts has dwindled to new lows. The number of history majors has decreased by 45% since 2007, and English has plummeted by half since the mid-1990s. A mere 7% of Harvard University’s 2022 freshman class expressed an intent to pursue the humanities, a significant drop from 20% in 2012 and nearly 30% during the 1970s.

The shift away from humanities has led to a diminishing emphasis on critical thinking, communication, cultural literacy, and analytical skills. Without a strong foundation in these skills, future generations struggle to engage in critical thought and appreciate diverse perspectives.

The humanities, academic disciplines concerned with languages, literature, history, philosophy, and the arts, have been in steep decline for a long time, and recently the rate of decline has accelerated dramatically, with plummeting enrollment numbers now forcing universities to close whole departments and severely limit humanities course offerings.

A social move away from reading may be part of the explanation for the humanities’ decline. The disciplines that are dropping are the most book-focused, and the amount of time spent reading has been going down for a considerable amount of time.

What this produces, culturally, is a generation that has been excellently trained in coding, data science, engineering, and quantitative reasoning, and that has had very little sustained exposure to the tradition of English letters. These are capable, intelligent people, but they are being asked to make literary judgments they are not equipped to make.

When someone who has never read Middlemarch or The Varieties of Religious Experience or Baldwin’s collected essays encounters language that is formal, layered, and precise, they have no frame of reference. It reads as inhuman to them because the only writing they regularly consume are tweets, documentation, requirements, and SEO-optimized articles.

The fault isn’t theirs at all.

The fault belongs to the educational and economic forces that told them literature was optional and TikTok and python are essential.

I have seen Ivy league professors (mostly AI hustlers, honestly) who are guilty of the above, so pedigree has no influence. This is illiteracy on a mass scale and we can all thank the brutal push for “STEM in everything” for the outcome.

The Class Dimension Nobody Wants to Discuss

There is another layer to this that the conversation almost always ignores, and it is about access.

Not everyone grows up in a home with bookshelves. Not everyone attends a school with a strong English literature program, or has a teacher who makes them read Fitzgerald and Faulkner. Not everyone has the leisure time, as a child or an adult, to sit and absorb the kind of formal, elevated prose that builds a literary vocabulary. These are not failures of intelligence or effort. They are the direct consequences of economic inequality and educational underfunding.

When someone who is just beginning to learn the craft of writing, and who has perhaps found in AI tools a patient teacher that will engage with their drafts at midnight without judgment, produces a piece of writing that uses words like delve or furthermore or intricate, they may genuinely be learning. They may be discovering a vocabulary and a set of structural tools that formal education never handed them. To mock that process, to perform public “gotcha-ism” over their word choices, is nothing but gatekeeping dressed up as sophistication.

The irony is particularly sharp: the people calling out these words as fake are often themselves demonstrating unfamiliarity with the tradition that produced them. They are, in a sense, equally distant from the canon they claim to be defending…just on different sides of it.

What Legitimate AI Detection Actually Looks Like

None of this is to say that AI-generated text is indistinguishable from human writing, or that concerns about AI in academic and professional contexts are invalid. They are entirely valid. But the reliable markers of machine writing are not individual word choices. They are structural and statistical.

AI detectors analyze patterns in writing that deviate from typical human composition. At their core, they use machine learning models trained on vast datasets of human-written and AI-generated text. By examining metrics like perplexity, measuring how predictable the text is, and burstiness, which assesses variation in sentence length and complexity, these tools flag phrases that appear unnaturally uniform or formulaic.

In other words: it’s the rhythm, not the words. Human writing is varied, idiosyncratic, and inconsistent in ways that reflect a person’s thought process, personality, and moment-to-moment decisions. It has what linguists call burstiness, sentences that lurch and sprawl, then snap short. It has digressions, contradictions, and the little imprecisions of a mind working in real time. Machine writing, even when it uses perfectly legitimate words, has a flatness, a relentless smooth competence that lacks the texture of presence.

Someone who has read widely and carefully knows what that texture feels like. They don’t need to look for the word delve.

A Closing Word on Humility

The words being flagged as AI tells are not AI’s invention. They are the shared inheritance of the English language, accumulated across a thousand years of writing. The transitions are ancient, the metaphors are classical, and the vocabulary of precision and nuance was built by human hands, over centuries, precisely because language needed it.

What is actually being revealed when someone triumphantly identifies tapestry or multifaceted as machine fingerprints is a gap, not in the writing, but in the reader. A gap that is understandable, traceable to real structural problems in how we educate and what we value, and completely forgivable.

What is less forgivable is the confidence. The flat certainty that a thousand years of literary vocabulary is a machine’s tell. The willingness to call out, shame, and dismiss writers, some of them beginners, some of them underprivileged, some of them doing the hard work of learning, without the knowledge to back it up.

Read more books. The words will start to seem familiar.

The words most commonly flagged as AI-generated - delve, tapestry, nuanced, pivotal, furthermore, multifaceted, intricate, realm, leverage, transformative - appear in the works of Homer, Shakespeare, Milton, Montaigne, Bacon, Johnson, Keats, Austen, Dickens, George Eliot, Henry James, Virginia Woolf, James Baldwin, and virtually every significant writer in the English tradition. They were not invented by a language model. They were inherited by one.