Workflows · 5 min · Updated 2026-06-15 · Atlas Editorial
The 9 tells that make AI writing detectable — and how to delete them.
TL;DR
§01
These words and phrases triggered AI-detector tools and human readers' suspicion in our 2026 audit:
delve · leverage · harness · seamless · seamlessly · in essence · navigate (as a verb for non-physical things) · dive in · game-changer · revolutionize · next-gen · cutting-edge · in today's fast-paced world · it's important to note that · unlock the power of
Paste them into every system prompt as forbidden. The model will route around them and the prose will sharpen immediately.
Tip
Add new tells to your forbidden list every time a friend says 'this sounds AI'.
§02
Em-dashes — like this — are the single biggest giveaway. Replace with periods. Two short sentences read more confidently than one with em-dashes anyway.
Three-item lists ("X, Y, and Z") are the second-biggest. The model loves rhythm-of-three. Real writers pick the best of the three and drop the rest.
Bullet lists where every bullet is the same length are a third tell. Vary them.
§03
Paste 3 samples of writing you love. Then ask the model to extract: typical sentence length, signature openers, what it never does, rhythm, and 3 favourite metaphors. Save that as a "voice profile" and prepend it to every writing prompt for the rest of the year.
This single move beats every "write like a human" hack.
Still curious?
Ask anything about this guide — clarifications, examples, what to do next.
Content · Claude 4.5
Repurpose long-form into a thread without the giveaway tells.
Marketing · ChatGPT 5
Paste your URL, get a hero rewrite that actually converts.