Tone settings won't save a ChatGPT draft. Ask for "casual" and you get casual scaffolding: the same stock openers, the same metronome rhythm, the same neat bow at the end. The model predicts the most likely next word, and likely is exactly the problem. Humans are less likely, and that's what reads as human.
So stop asking it to sound human and start editing it like one. Seven moves, in order of impact.
AI openers are universal claims: "In today's fast-paced world...", "We're thrilled to announce...". Nobody scrolls for universal. Your opener should be the most specific thing you know: a number, a date, a place, a thing someone actually said. Let ChatGPT draft the body, but write line one yourself.
Model output settles into even, medium-length sentences. Read your draft and mark every full stop; if the lengths barely vary, rewrite two sentences: one into five words, one into a longer, looser one. Contrast is what a human ear expects.
Furthermore, moreover, in conclusion, it is important to note that: connective tissue with nothing to connect. In a LinkedIn post it's pure filler, and it's the single fastest tell. Full list with fixes in the AI filler dictionary.
Nothing on the right loses meaning. Everything on the left loses trust.
ChatGPT closes by restating the opening, which is what school essays do, not people. Close with a question you'd genuinely like answered, or one flat sentence of opinion. If the last line could belong to anyone's post, rewrite it.
Anything your mouth trips over is machine residue. Fix only the stumbles; leave the rest.
Once your voice is in there, paste the draft into the cleaner. It strips the 38 stock tells in one go and shows each removal as a chip so you can audit every edit before posting. You bring the specifics; it brings the sweep.
That order matters. Human first, machine second. The other way around reads fake because it is.
Clean your draft freeFree Toolbox
The playbook above is the thinking. These browser-based tools do the mechanical part: free, no sign-up, nothing uploaded.