AI ‘tells’ — and how to write around them

Readers cannot always explain why a piece of writing feels machine-made, but they detect it reliably. The signals are consistent enough to write rules against.

The tells

• The contrast sentence. "It's not X — it's Y." Once is a rhetorical device. Four times is a template.
• Tricolon everywhere. Three-item lists in every paragraph, regardless of whether the subject has three parts.
• Elevated abstractions. Landscape, realm, tapestry, journey, delve, unlock, elevate.
• Hedged nothing-claims. Sentences that could be deleted without losing information.
• Symmetrical paragraphs. Every block the same length, every section the same shape.
• Summary sentences that summarise nothing new, especially opening ones explaining why the topic matters.
• No specifics. No dates, no names, no numbers, no anecdote only this business could tell.

The rules that fix them

Ban the vocabulary explicitly — a stylebook list is more effective than any instruction about tone. Vary paragraph length deliberately, including one-sentence paragraphs. Require at least two concrete specifics per post: a real number, a named example, a dated event.

And cut the first paragraph. Machine-written pieces almost always open by explaining that the subject is important. Human ones open with the thing itself.

Specificity is the only reliable anti-AI signal. Everything else is style.

The proof bank exists for this reason: real quotes, real statistics, real customer stories, available to the model at drafting time. A post with three facts only your business knows never reads generic.