Direct answer
Claude's output signature is distinct from ChatGPT's: more hedging, more long compound sentences stitched with semicolons, and a more formal register. The 6 moves that reliably humanise Claude text are: cut the hedges, break long sentences into shorter ones, shift the register one notch casual, reach for one odd word per paragraph, add one piece of specific detail only you would know, and run the result through a cadence-aware humaniser.
Claude is one of the three mainstream large language models alongside ChatGPT and Gemini. Its output has a specific signature readers often describe as "thoughtful and careful": heavier hedging, longer compound sentences, more formal register. That signature is also what detectors pick up and what human readers unconsciously feel as "AI." This piece covers the six moves that address Claude's specific tells, in order of how much they shift the signals.
What Claude writes like, specifically
Compared to ChatGPT, Claude tends to produce longer sentences stitched by semicolons; more hedging phrases ("it is worth noting," "one could argue"); a more careful register with academic connectives; and a tendency to open paragraphs with meta-statements about what the paragraph will cover. These are not failures of the model: they are features of its training. They are also tells detectors read as low-perplexity, low-burstiness, high-connective text.
Move 1: cut the hedges
Claude hedges more than ChatGPT. "It is worth noting," "one could argue," "it would be reasonable to suggest," "there is a case for the view that" - delete every one on sight. The resulting paragraph will read more direct and sound less like a model avoiding commitment. If you believe the sentence underneath, say it. If you do not believe it, cut the sentence.
Move 2: break up long compound sentences
Claude strings three clauses together with semicolons. Break each long sentence into two or three shorter ones. The result is higher burstiness, which is the single strongest signal detectors read. For why burstiness matters, see AI detection explained.
Move 3: shift the register one notch casual
Claude writes in a formal register by default. Push the register one notch toward how you actually speak. "Furthermore" becomes "And." "In the context of" becomes "when it comes to." "Various considerations suggest" becomes "the point is." The shift is subtle enough not to break the content but large enough to change the texture of the prose.
Move 4: one odd word per paragraph
Claude, like every modern language model, picks the most probable next word. Humans do not. Reach for one noun or verb per paragraph that is specific, slightly odd, or carries personal connotation the model would not have inferred. The paragraph becomes measurably less uniform in perplexity, which detectors measure, and reads more alive, which humans notice.
Move 5: one specific detail only you know
Claude cannot know the specific detail of your context. The espresso machine in your lab, the paper your supervisor keeps citing, the student who always disagrees. Put one such detail in every piece. The reader feels it even without knowing why.
Move 6: cadence humaniser pass
After the five editorial moves above, run the result through a cadence humaniser as a last structural check. The humanise.ai tool is free, requires no account, and caps each pass at 10,000 characters. The point of this last step is not to replace the human editing above; it is to catch the structural tells your eye may have missed. For the broader free comparison see best free AI humanisers; for the ChatGPT variant see how to humanise ChatGPT text. For the paid upgrade when you need API access or team workflows, Walter Writes is the one we point readers at.
Frequently asked questions
Is Claude text harder to detect than ChatGPT text?
Not meaningfully. Mainstream detectors read the shared perplexity + burstiness signature. Claude outputs are flagged in broadly similar ways to ChatGPT and Gemini. The small per-model differences are in connective-phrase density and hedging frequency, addressable by the editing moves above.
Will the humaniser work on Claude 3 vs Claude 4 output?
Yes. The humaniser addresses signatures shared across the Claude family, not specific to one version. Newer Claude releases produce slightly less-hedged output on average, which reduces the editing burden on move 1 but does not change the structural moves.
Will humanised Claude text pass Turnitin?
Often, but not reliably. The six moves above address burstiness, perplexity and connective density, which are the signals Turnitin reads. No approach guarantees 100 percent pass-rate because detectors update. See Turnitin vs GPTZero and Turnitin AI detection 2026.