Direct answer
Gemini's output signature is distinct from ChatGPT and Claude: it leans heavily on mechanical connectives (first, second, third), drifts toward bulleted lists even when prose is better, and keeps sentence lengths unusually uniform. The 6 moves that reliably humanise Gemini text are: cut the enumerated transitions, convert bullet drift back to prose, aggressively vary sentence length, 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.
Gemini is Google's mainstream large language model, used widely in Workspace integrations and standalone. Its output has specific tells: a tendency to enumerate steps ("first, second, third"), a drift toward bulleted lists even when the content fits prose better, and sentence lengths that cluster near a mean. This piece covers the six moves that address Gemini's specific signature in order of how much they shift the signals detectors read.
What Gemini writes like, specifically
Compared to ChatGPT and Claude, Gemini tends to use more mechanical enumeration ("first, second, third"), drift toward bulleted lists for content that reads better as prose, and keep sentence lengths closer to a mean than human writing does. The result is unusually low burstiness, which is the single strongest signal detectors read.
Move 1: cut enumerated transitions
"First," "second," "third" at the start of sentences in a prose paragraph is a Gemini tell. Delete them. The remaining sentences usually hold together as a sequence without the scaffolding. Where the content really is enumerated, prefer "to start with," "and," "finally" or similar natural connectives.
Move 2: convert bullet drift back to prose
Gemini drifts toward bulleted lists for content that reads better as prose. Rewrite the bullet list as a paragraph with varied sentence length. The content is the same; the structure now matches how a human would actually write it.
Move 3: aggressively vary sentence length
Burstiness is the signal detectors read most strongly. Edit the output so some sentences are 6 words and others are 35. One-word sentences work. Fragments work. Long compound sentences with three clauses stitched by semicolons work. Read your edited paragraph aloud; if it reads with a rhythm, you have fixed the main tell. See AI detection explained for the mechanism.
Move 4: one odd word per paragraph
Gemini 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 paragraph becomes measurably less uniform in perplexity.
Move 5: one specific detail only you know
Gemini cannot know the specific detail of your context. 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, no account, 10,000 characters per pass. For the ChatGPT variant see how to humanise ChatGPT text, for Claude how to humanise Claude text. For the paid humaniser we point readers at when a free textarea tool is not enough, see Walter Writes.
Gemini vs ChatGPT and Claude, side by side
| Model | Signature tell | Detector risk |
|---|---|---|
| Gemini | Mechanical enumeration + bullet drift + uniform sentence length | High (lowest burstiness) |
| ChatGPT | Flat paragraph rhythm + connective density | High |
| Claude | Hedging + long compound sentences + formal register | Moderate |
All three models fail detectors in broadly similar ways because the detectors all read the same perplexity-and-burstiness signal. The practical implication: the six moves in this guide also transfer to ChatGPT output and Claude output with small per-model adjustments. If your workflow mixes drafts from more than one model, treat the moves as a universal editing pass and vary intensity by model.
When each move earns its keep
On short passages (under 300 words), move 3 (vary sentence length) does most of the work; detectors read burstiness most strongly on short samples. On longer-form content (over 1,000 words), move 5 (specific detail) matters more because the reader experience depends on texture more than any single sentence. For academic or research-adjacent writing where you want to preserve a careful tone, use move 3 heavily but go light on move 1 to keep nuance. For journalistic or opinion writing where directness is the goal, invert that balance.
Frequently asked questions
Is Gemini text easier to detect than ChatGPT text?
Slightly, in practice. Gemini's lower burstiness and enumerated-connective tendency produce a cleaner detector signature than ChatGPT or Claude output of similar length. The editing moves above are addressed at those tells.
Does Workspace-integrated Gemini differ from the standalone product?
The underlying signature is similar. Workspace drafts (in Docs, Sheets, Gmail) tend to be shorter and more context-aware, which reduces some of the enumerated-connective tendency but does not change the structural moves needed.
Will humanised Gemini text pass Turnitin?
Often, but not reliably. The six moves address burstiness, perplexity and connective density. No approach guarantees 100 percent pass-rate because detectors update. See Turnitin AI detection 2026.