LLM-written cold email is now easy to detect — and inboxes are starting to

When everyone uses the same models with similar prompts, the output converges. The same openers, the same "I noticed you…", the same three-sentence rhythm. At scale that sameness becomes a fingerprint — and both filtering systems and humans have started recognizing it.

What's actually changing

The advantage of AI in outreach was never "write more emails." It was supposed to be relevance at scale. In practice most teams used it for *volume* at scale, which produces high send counts and collapsing reply rates as recipients learn the pattern.

Why it matters for outreach

• Generic AI copy increasingly correlates with lower engagement, and engagement is what mailbox providers actually score. Low replies + low opens → worse placement, regardless of how "good" the copy reads.
• The teams keeping reply rates are using AI for *research and angle*, then keeping the message short, specific, and human — the opposite of templated AI bulk.
• Spintax-style variation and genuine per-recipient specifics matter more now, not less, because identical AI output across thousands of sends is the thing being detected.

The takeaway

AI's edge in outreach moved from "produce text" to "produce *relevance*." If the recipient can tell a model wrote it with no specific knowledge of them, it's the same as a worse version of a template. Specificity is the moat again.