How to Write Cold Email That Doesn't Sound AI-Generated

Recipients have gotten faster at spotting AI-drafted cold email, partly because so much more of it exists now, and partly because the patterns are genuinely recognizable once you've seen a few hundred of them. An email that reads as generated — even if the offer is good — tends to get skimmed and dismissed faster than one that reads as written by a specific person who thought about the recipient.

Why This Matters More in 2026 Than It Used To

The issue isn't that AI-assisted writing is inherently bad — plenty of well-performing cold email starts from an AI draft and gets edited into something specific. The issue is unedited AI output, which tends to default to a small set of vocabulary and structural habits that recipients now associate, consciously or not, with mass-sent, low-effort outreach. Reply.io's own analysis of AI-sounding email puts it directly: individual words "mean little" on their own, "but when they cluster, they usually signal machine-assisted writing with no human oversight" — which is the same clustering logic that applies to spotting it in any writing, cold email included.

The Specific Tells to Check For

Vocabulary that shows up disproportionately in AI-generated text: delve, leverage, foster, harness, underscore, and inflated-importance words like crucial, pivotal, robust, seamless, comprehensive. Reply.io flags an overlapping set specific to sales writing — "plethora," "myriad," "paradigm," "cutting-edge," "game-changing," "revolutionize" — plus a set of hedging phrases: "generally speaking," "arguably," "it is worth noting," "to some extent." Structural tells matter as much as vocabulary: stacked transition words opening consecutive sentences ("Moreover... Furthermore... Additionally..."), the "not just X, but Y" construction, everything wrapped in a tidy rule-of-three, and a closing line that summarizes rather than asks for something specific.

It's a Cluster, Not One Word

One elevated word in an otherwise plain email doesn't read as AI-written — everyone uses "leverage" occasionally. What reads as generated is several of these tells stacking in a short space: an inflated vocabulary word, a stock transition, a rule-of-three list, and a summary closer, all in a five-sentence email. The fix isn't hunting down every instance of one banned word; it's noticing when an email has that smooth, symmetric, slightly-too-polished quality where every sentence is doing the same amount of work and nothing sounds like it came from a specific person with an actual opinion.

What to Do Instead

• Read it out loud. If a sentence doesn't sound like something you'd actually say to a person, rewrite it in the words you'd use in a real conversation.
• Cut the throat-clearing. "I wanted to reach out because" and "I hope this email finds you well" delay the actual point and are two of the most common openers in generated cold email — start with the observation or the offer instead.
• Let sentences be different lengths. Uniform, medium-length sentences in every line is itself a tell; real writing has some short, blunt sentences and some longer ones.
• Include one specific, real detail that couldn't have been generated from a template — a specific number, a named example, an actual opinion about the recipient's situation rather than a generic compliment.
• If you used AI to draft it, edit hard. Treat the first draft as a starting point to rewrite, not a final version to lightly polish — the editing pass is where the tells actually get removed.

FAQ