What makes a writing voice recognizable? A humanities answer.

The tech industry has spent three years trying to answer the question "what is voice?" as if no one had ever asked it before. The answers have been thin. Some tools treat voice as vocabulary. Some treat it as tone. Some treat it as the output of a prompt that says "write in the style of [writer]." None of these feel adequate to the thing readers recognize when they say, of a piece they have never seen before, "this sounds like her."

The humanities have been at this question for a long time. Rhetoric, stylistics, literary criticism, and writing pedagogy have been building theory about voice since at least the first century BCE. What these traditions offer is not better technology. What they offer is a better question to ask. Once the question is right, the answer is not hard to sketch.

古What rhetoric has said for two thousand years.

Cicero, writing De Oratore in 55 BCE, identified what he called genera dicendi: kinds of speech. Three registers, each characterized not by the words a speaker used but by what the speaker chose to do at a given moment (inform, move, delight). The register was a structural choice about the shape of the speech, not a lexical one.

Quintilian, a century later, extended the argument in his Institutio Oratoria. A speaker's oratio (their style in the deep sense) arose from the combined pattern of what they emphasized, how they arranged their arguments, and which examples they habitually reached for. A speaker could be recognized across speeches by the consistency of this pattern, even when the topics were unrelated.

The tradition that runs from the Greeks through the Roman rhetoricians into the medieval artes dictaminis and forward into the modern writing-craft tradition shares a consistent claim. What distinguishes a speaker or writer is not what they say. It is what they reliably do at the structural level: what they open with, what they emphasize, how they move from claim to evidence, what they leave unsaid.

This is the same claim that surfaces in every serious theory of voice since, from Buffon's "le style, c'est l'homme même" in 1753 to contemporary stylistics. The claim is boring, in the sense that it has been made repeatedly for two millennia. The claim is also right.

詩What literary critics call the signature.

Twentieth-century literary critics, working mostly on fiction, gave this cluster of structural habits a name: the signature. A writer's signature was the recurring pattern of structural choices across their body of work, detectable by close readers, mostly invisible to casual ones.

Some signatures are unmistakable even to readers with no training. Hemingway's preference for short declarative sentences ending on a noun. Faulkner's sentences that run for half a page before landing. Didion's habit of inserting a bare fact (a date, a body count, the weather on a specific Tuesday) to turn an essay. These are structural signatures. The writer's attention to their own pattern, over years, is part of what makes them readable as themselves.

Most writers do not have signatures this visible. Most writers have signatures that are quieter, subtler, detectable only by the reader who has read enough of them to form an unconscious pattern. But every writer who has written seriously for years has a signature. The signature is the structural fingerprint of their attention.

Literary criticism has developed tools for describing signatures. The work of scholars like Leo Spitzer, who in the 1940s and 1950s built a discipline called "stylistics" around identifying a writer's characteristic structural moves. The more contemporary work of narratologists (Gérard Genette, Mieke Bal, James Phelan) tracing how a writer habitually handles time, perspective, reported speech. The work of rhetoricians like Kenneth Burke, whose "pentad" (act, scene, agent, agency, purpose) was itself an attempt to catalog the structural moves a writer relies on.

None of these traditions treat voice as word choice. All of them treat it as the structural habits, repeated across a body of work, that let a reader say "this is her."

声Six things that, together, make a voice.

When Onaji's Voice Profile models voice, it does so across six dimensions. These are not arbitrary; they are the smallest set of structural axes that, together, capture what the humanities traditions above have been describing. The six are:

One: opening habits. Where does the writer start? With the reader's situation, a specific event, a direct claim, a category statement, a question, a fragment? Writers have strong tendencies here. The first sentence of a post is often the most structurally diagnostic thing about it.

Two: source integration. When the writer refers to an outside fact, study, quote, or event, where does it appear relative to the writer's own claim, and how tightly is it attributed? Some writers quote directly and attribute in-line. Some paraphrase and bury attribution. Some work entirely from their own observation. The choice is structural and consistent across a writer's work.

Three: argumentative moves. When the writer has a counterargument to address, what do they do? Concede and pivot. Name and reframe. Drop into a specific. Ask a question the piece does not answer. Just assert. Each writer has a small handful of moves they reliably use. The combination is specific.

Four: rhythm and cadence. Sentence length patterns. Paragraph length patterns. Whether the writer mixes long and short, or stays within a narrow range. Whether short sentences land at paragraph ends or get scattered through. This is the one dimension most advice talks about; it is one of six, and not the most important.

Five: diction preferences. Word choice at the register level (formal, casual, technical, plain). Specific words the writer avoids and specific words they reach for. Whether they prefer Latinate or Anglo-Saxon roots. This is the other dimension advice usually addresses; also one of six.

Six: closing patterns. How does the writer land? On a short declarative, a specific observation, a question, a forward-looking statement, a return to the opener? Closers are, after openers, the most structurally diagnostic feature of a writer's work. A writer's closing pattern is often the most distinctly theirs.

Three of these (rhythm, diction, and to some extent source integration) are what most surface-level voice advice attends to. Three (opening habits, argumentative moves, closing patterns) are what the humanities traditions would call the deep layer. Both matter. The deep layer is what a generic AI tool cannot model without a persistent profile built from the writer's actual work.

模Why a model of voice has to be structural.

This is the consequential claim. A voice model that attends only to diction, tone, and rhythm (the surface three of the six dimensions) will produce drafts that sound like generic business writing with the writer's favorite words. Most lightweight AI tone tools do exactly this.

A voice model that attends to opening habits, argumentative moves, and closing patterns (the structural three) produces drafts that read as recognizably the writer's, even when the specific words are rough. The structural three carry the signature.

This is what the humanities traditions have been saying since Cicero. The reason the tech industry's answers to "what is voice?" have been thin is that those answers have treated voice as a style problem, which it partly is, rather than as a structural problem, which it mostly is. The humanities traditions got the structural claim right two thousand years before large language models existed. The traditions are still worth reading.

Onaji's Voice Profile exists because a serviceable model of voice has to be structural. Two of the six dimensions address surface; four address structure. The tool's drafts read as the writer's because the structural four get modeled first. The surface two come along with them.

The claim is old. The application is new. What has changed is that a writer now has a tool capable of doing, at drafting time, what previously only a year of deliberate revision could accomplish: producing first drafts that are in the writer's structural shape.