Claude · The tell-spotter
Claude's writing style: the patterns that give it away
Structural habits, not vocabulary, are what give the model away.
Claude's recognizable style comes down to habits of writing, not a hidden mark. It announces before it says anything, opening with agreement and only then getting to the point. It hedges to keep claims safe from contradiction. It turns paragraphs into lists, each point wearing a bold label. It reaches for em dashes to fold extra clauses into one sentence. And it closes by restating the answer. None of this is a watermark, and none of it proves anything on its own. Detection is statistical: detectors and readers score how predictable the text looks, not whether a provider's signal is present. Habits are exactly what editing changes.
Why this page exists
This page defines the style before anyone decides what to do. It is the reference for the rest of the hub and the natural bridge into the tells glossary.
The tells that show up here
Not every AI tell appears everywhere. These are the ones that cluster on this surface, and what it is about the format that produces them.
- chatbot-artifact
- Validation openers, the model's most consistent reflex: agree with the question before answering it.
- em-dash-chain
- Dense dash chains, the formatting habit our lint scores hardest.
- not-only-but
- Parallel structure as a default rhetorical move, including the 'not X but Y' construction.
- throat-clearing
- Pivot openers like 'let's dive in' and transition lines that delay the point.
Before and after
A generic AI draft on the left. On the right, the real output of running that draft through underslop, checked so it adds no name or number the draft did not already carry. The voice is The Steady Hand.
Great question! When it comes to Claude's writing style, it's important to note that the model's voice — its careful cadence, its polished delivery, its unmistakable rhythm — stands as a testament to the sophistication of modern AI. It's not just about word choice; it's about the underlying structure of communication itself. First, there's the opener that validates you. Second, there's the hedge that keeps every claim safe. Third, there's the list that organizes your thoughts for you. And in today's fast-paced world, that structure serves as a beacon of clarity — a hallmark of machine-generated prose. In summary, Claude's style is recognizable because it is consistent, and that consistency is what gives it away.
Claude's writing style is consistent enough to spot. The opener validates you before it says anything. The hedge keeps every claim safe. The list organizes the thoughts, and the structure stays the same every time. That structure is what gives it away. None of it is accidental, and none of it is hard to miss once you know what to look for. The cadence is even. The delivery is polished. The rhythm holds steady. It reads like a well-built machine, not a person. That consistency is the tell. It holds because the pattern does not break.
How to do it by hand
- Scan for the structural tells Look for the habits in the order they appear. The agreeing opener comes first, then the hedges, then the imposed lists, then the dash chains, then the summary closer. Each one is named on the tells glossary page, along with its fix.
- Strike the reflexes Cut the reflexes in the order they appear. The opener goes first, then the hedges, then the imposed lists. The point stays.
- Rewrite in your own rhythm Rewrite in your own rhythm. Keep the facts. Vary the sentence length. Write the way you talk.
- Check the result against the tells Re-read for residual patterns. The structural ones are the ones a light edit will not reach, so run the scan once more.
Questions
What does Claude's default writing style look like?
It is a set of structural habits, not a watermark. Openers that agree with the question before the point. Hedges like "it's worth noting" that keep claims safe. Imposed lists, dash chains, and a closing summary that restates everything. Recognition of those habits is statistical, not a fingerprint.
Why does Claude text all sound the same?
The same habits repeat in every draft. The model defaults to one register: agree first, hedge second, list, dash, summarize. Those habits are what make the text read as one voice.
Does Claude have a distinctive voice?
It has a default register that shows up in statistics. A detector gives a probability, not proof of authorship. That record is disputed, and it carries real false positives.
Is Claude text easy to detect?
The 2023 claim that Claude did not appear on detectors is out of date. Detection is statistical and contested. What holds steady is the record on bias and false positives, not any verdict against the model itself.
How do I stop my text from sounding like Claude?
Edit the habits, not the vocabulary. Drop the agreeing opener. Cut the hedges where you're sure. Put the lists back into paragraphs, and remove the dash chains. Trim the summary closer. Each pattern has a fix page on this site.
Claude marks are real and the coverage is still settling. underslop is the edit, not a remover: it edits an AI draft so it reads like you wrote it, and it makes no promise about what any detector will report. Your text is held for 0 seconds.