Claude Writing Patterns
Vocabulary markers, structural habits, and cadence patterns specific to Claude (Anthropic). These patterns are drawn from the Bloomberry AI Sentence DNA corpus — 7,400+ catalogued AI-writing signal entries. Claude has a distinct writing profile that differs meaningfully from ChatGPT and Gemini.
These are writing signals, not authorship determinations. The diagnostic signal comes from co-occurrence density — not from any single entry. Related: Anthropic published research documenting 171 functional emotional representations in Claude — which manifest as observable writing patterns.
What distinguishes Claude's writing
Claude's writing profile is defined by three distinct tendencies: elevated formal register (moreover, indeed, fundamentally), therapy-adjacent emotional vocabulary (hold space, sit with, genuinely difficult), and first-person vulnerability framing (confession openers, empathy openers, acknowledged uncertainty).
Where ChatGPT resolves with a clean, forward-looking closer, Claude tends toward open conclusions — acknowledging remaining uncertainty or inviting further reflection. This non-resolution pattern is one of the most reliable structural Claude markers in professional and social writing contexts.
Vocabulary Markers
Words and phrases that appear at elevated frequency in Claude output relative to ChatGPT, Gemini, and natural human writing in equivalent contexts.
| Phrase / Word | Pattern Type | Frequency | Notes |
|---|---|---|---|
| tapestry | Vocabulary cliché | High | Metaphor for interconnectedness; almost exclusive to Claude |
| nuanced | Qualifier | Medium | Claims complexity without demonstrating it |
| fundamentally | Adverb qualifier | Medium | Frames a claim as foundational; hedges via emphasis |
| moreover | Transition | Medium | Formal connective; elevated in Claude vs. human writing |
| indeed | Affirmation | Medium | Formal affirmation uncommon in natural professional writing |
| hold space | Therapy-adjacent idiom | Medium | Emotional support metaphor; distinctive Claude register |
| that said | Contrast filler | High | Pivot phrase that concedes without abandoning the prior point |
| delve | Vocabulary cliché | High | ChatGPT / Claude shared — elevated 28× by Kobak et al. |
| compelling | Adjective | Medium | Evaluative without a specific criterion; topic-independent |
| it's worth noting | Hedge bridge | High | Introduces information while hedging its importance |
| less is more | Aphoristic filler | Medium | Aphoristic closer that adds no new information |
| the reality is | Reframe opener | High | Implies the reader held an incorrect assumption |
Showing a representative selection. Full vocabulary dataset: 4,500+ entries in the AI Sentence DNA corpus.
Hook Patterns
Opening formulas that Claude defaults to. Claude hooks tend toward empathy and shared vulnerability rather than the authority-claim hooks common in ChatGPT output.
AI first lines are becoming easy to spot
A lot of AI-written posts start with the same moves: “Most X do Y…” or “X is not a Y problem, it’s a Z problem.” But the pattern library goes further — temporal openers, observer openers, candor frames, confession hooks. These structures appear across all models.
Claude first-line patterns: Confession opener, Empathy opener →| Pattern Name | Example | What it does |
|---|---|---|
| Empathy opener | "If you've ever struggled with X, you're not alone." | Opens by validating shared difficulty — high Claude association |
| Confession opener | "I used to think X. I was wrong." | First-person vulnerability as a credibility mechanism |
| Paradox opener | "The more you try to X, the less Y you get." | Logical inversion that creates curiosity |
| Rhetorical question opener | "What if the way we think about X is wrong?" | Opens with a question Claude then proceeds to answer |
| Aphorism opener | "Clarity is speed. Complexity is delay." | Two short declarative sentences in parallel — aphoristic rhythm |
| Acknowledgment opener | "This is a genuinely difficult question." | Validates complexity before addressing it — hedged approach |
Cadence Structures
Sentence-level rhythm patterns strongly associated with Claude output.
Aphorism Pattern
HighStructure
Abstract noun or concept → Simple declarative predicate → Optional contrasting clause
Example
"Clarity is speed. Less is more. The simplest version often wins."
Claude produces symmetrical two-clause aphorisms at elevated rates. Each half mirrors the other in length.
Hedge-Assertion Pair
HighStructure
Qualifying hedge → Assertive claim that follows
Example
"While individual cases vary, the evidence suggests the pattern is consistent."
Claude hedges before asserting at higher rates than ChatGPT. The hedge provides intellectual cover; the assertion provides forward momentum.
Empathetic Concession
HighStructure
Acknowledgment of reader's position → Pivot to nuanced reframe → Invitation to consider
Example
"It makes sense to feel uncertain about this. The evidence is genuinely mixed. That ambiguity is worth sitting with."
Claude uses emotional validation as a structural device — often ending on openness rather than a GPT-style resolution closer.
Layered Complexity Opener
MediumStructure
This is more complex than it appears → Multiple dimensions identified → Synthesis offered as incomplete
Example
"The question of X is genuinely complicated. It involves Y, Z, and factors that aren't yet well understood."
Claude systematically emphasizes complexity. Where GPT resolves, Claude acknowledges limits. Both patterns are recognizable.
Structural Habits
Document-level patterns that distinguish Claude output from other models — patterns of length, register, resolution, and emotional framing.
Longer paragraph structure
Claude generates longer, more densely connected paragraphs than ChatGPT. Where GPT tends toward short stacked sentences (one per line in social contexts), Claude uses multi-sentence paragraphs with explicit connective tissue — moreover, furthermore, that said, indeed.
First-person vulnerability framing
Claude uses first-person admissions — "I used to think," "I find this genuinely hard," "I was wrong about this" — at elevated rates. This creates an appearance of intellectual honesty that functions as a credibility signal while remaining formula-driven.
Open-ended conclusions
Where GPT produces resolution closers that cleanly answer the question raised, Claude more often ends with a qualified acknowledgment: "This remains an open question," "The full picture is still emerging." This pattern of non-resolution is a reliable structural Claude marker.
Emotional register elevation
Anthropic published research in 2025 identifying 171 functional emotional representations in Claude's activation space. These manifest as therapy-adjacent vocabulary (hold space, sit with, genuinely difficult), elevated empathy framing, and first-person emotional disclosure that does not appear at similar rates in GPT or Gemini outputs.
Replacement Pairs
Direct substitutions for the most common Claude vocabulary signals. Most Claude markers can be fixed by deletion or by replacing the abstraction with a specific claim.
| Claude-coded expression | Human alternative |
|---|---|
| tapestry | [identify the actual thing] — system, pattern, mixture, combination |
| nuanced | [specify the nuance] — complex, context-dependent, varies by [X] |
| fundamentally | [delete — state the claim directly] |
| moreover | also, and, additionally — or restructure the sentence |
| indeed | [delete — state the affirmation directly] |
| hold space | support, make room for, allow |
| that said | [delete — pivot without the filler] or: but, however |
| it's worth noting | [delete — just state the note] |
| the reality is | [delete — state the reality directly] |
| compelling | convincing, strong, clear — or explain why without the adjective |
Explore by model
Bloomberry screens for Claude-specific patterns in real time.
Every Bloomberry generation runs the live corpus as a filter. Empathy openers, formal connectives, and open-ended non-resolution patterns are flagged and rewritten against your calibrated voice.