AI Voice Clichés

The workflow is: human provides direction, raw material, and editorial judgment. AI drafts, organizes, and formats. The final voice should be the author's — grounded in real experience, specific to the context, unmistakably human in rhythm and stance.

The problem isn't AI involvement. The problem is when AI-typical patterns leak into the final text and signal to readers that nobody real is behind the words. A ghost writer who draws attention to itself has failed at its job.

The patterns below are drawn from Wikipedia's "Signs of AI writing" guide and validated against our own published output. They are signals, not bans. When they cluster, the ghost writer is showing through and the text needs an editing pass from the author.

Negation Parallelism — "It's not X. It's Y."

The structure "That's not a feature. That's a value system working." is the most recognizable AI tell in persuasive writing. It creates dramatic contrast through negation, mimicking punched-up copywriting. Variations include "It's not about X — it's about Y" and "This isn't X. This is Y."

The pattern is not inherently wrong. It becomes a tell when it appears formulaically — especially as a paragraph closer meant to land with weight. Human writers use this structure occasionally. LLMs use it reflexively.

Caught in our own writing: "That's not a feature. That's a value system working." — flagged by a reviewer before publication.

Puffing Importance

Connecting a specific observation to "broader themes" or "larger movements" without evidence. Phrases like "represents a significant shift," "marks a pivotal moment," "plays a crucial role in the broader landscape." The claim inflates the subject's importance beyond what the evidence supports.

Formulaic Transitions

Overuse of a small set of conjunctive adverbs: "Moreover," "Furthermore," "Additionally," "Notably," "Importantly." Human writers vary their transitions or eliminate them. LLMs insert them mechanically between paragraphs.

Em Dash Overuse

LLMs use em dashes more frequently than human writers and in positions where humans would use commas, parentheses, or colons. The em dash becomes a tic rather than a deliberate choice. Most noticeable when multiple em dashes appear in a single paragraph.

Uniform Pacing

Every sentence roughly the same length. Every paragraph roughly the same size. The rhythm is metronomic rather than varied. Human writing has short punchy sentences followed by longer elaborations. LLM writing marches at a constant tempo.

Bold-Then-Explain

A bolded phrase followed immediately by a sentence that restates the same idea. Common in list-heavy LLM output. The bold carries no additional information — it just reformats the sentence that follows.

Generic Positive Descriptors

"Rich cultural heritage," "enduring legacy," "vibrant community," "innovative approach," "comprehensive framework." These phrases sound authoritative but convey no specific information. They are filler dressed as substance.

Overclarification

Writing as though the reader has reached an incorrect conclusion that needs correcting. "While it may seem like X, it's actually Y." This pattern assumes disagreement where none exists and creates a patronizing tone.

Sycophantic Openers

"Great question!" "That's a really interesting point." "Absolutely!" These leak from conversational AI into written drafts when the writer pastes without editing.

Summary Clichés

"In summary," "Overall," "In conclusion," "Taken together, these findings suggest." Human writers either skip the summary or integrate it naturally. LLMs announce it.

This is a preflight and validate checklist item, not a ban list.

Before publishing any public-facing text:

1. Read the text aloud. Does it sound like a person or a press release?
2. Check for clustering. One em dash is fine. Three em dashes plus a negation parallelism plus a formulaic transition in the same section is a signal.
3. Kill sycophantic openers. Always.
4. Replace generic descriptors with specific claims. "Innovative approach" → what specifically was new?
5. Vary the pacing. If every paragraph is four sentences, something is wrong.

The goal is not AI-free writing. The goal is writing where the ghost writer disappears and the author's voice comes through — grounded in real experience, specific enough to be credible, human enough to earn trust.

• Writing Canon — the quality checklist for all canon documents
• Guide Posture — we enter their story, not the other way around
• Author Identity Language — approved and prohibited identity descriptions
• Wikipedia: Signs of AI writing — the source catalog