How can you tell an email was AI-written?

Common tells include opening lines that praise vague accomplishments, sentences that hedge with multiple qualifiers, transitions that read smooth but say nothing, and a closing pitch that does not quite match the supposed personalization. The pattern is fluent prose with no specific knowledge, paired with a generic ask. Most templates are recognizable within the first two paragraphs.

Will AI tells improve over time?

Yes. Generation quality is rising, and the obvious tells are getting subtler. Five years from now, content-pattern detection of AI writing will be unreliable. The structural answer is to filter on intention rather than on stylistic markers. A sender who paid four cents to reach you is signaling intent, regardless of who or what wrote the prose.

Is AI-written outreach always bad?

No. AI-assisted writing is now common in legitimate outreach too. The quality of the prose is not the right thing to filter on. The right filter is whether the sender thought reaching you was worth a small payment. AI or human authorship is downstream of that signal.

Should I block all AI-generated email?

You cannot reliably do that. Even if you could detect AI writing today, you would catch many legitimate professional emails that used AI assistance. The blunt detection approach produces too many false positives. The intention filter (cover charge gate) avoids this problem because it is indifferent to authorship.

How does Rythm handle AI-generated cold email?

Rythm does not try to detect AI writing. The cover charge gate applies to any unknown sender. If the sender pays the cover charge (roughly four cents by default), the email reaches the inbox regardless of how it was written. If the sender does not pay, the email waits in the held-for-review folder. The filter is on intention, not on authorship.

AI-Generated Cold Email: How to Spot the Tell

AI-written cold outreach has identifiable patterns. Here is how to recognize the tell, and why content-pattern detection is not the durable answer.

The tell of an AI-generated cold email used to be obvious. A few years ago, the prose was clunky, the personalization was wrong in specific ways, and the closing pitch did not match the supposed context. In 2026, the tell is subtler but still present in most templates. This post is the practical guide to recognizing the patterns and the structural reason why content-pattern detection is not the durable answer.

What AI-Generated Cold Email Looks Like

The current generation of cold-outreach AI templates produces a recognizable shape:

Opening that praises something vague. “I came across your work on X and was really impressed by your approach.” The X is usually scraped from a public profile. The praise is vague because the system does not actually know what was impressive about it.

A bridge that names a current trend. “Given the rising importance of [topic], I thought I would reach out.” The trend is plausible but generic. It functions as transitional connective tissue rather than as a statement about your specific situation.

A pitch that almost matches. “We help [your category] companies do X. Would you be open to a quick chat?” The category is correct because the sender filtered by it. The pitch is generic because the same template runs across thousands of recipients.

A closing that hedges. “If now is not the right time, I would love to revisit this in a few months.” The hedge handles the rejection case in advance. It reads polite but indicates a templated pattern.

Subtle prose tells. Sentence structures that hedge with multiple qualifiers (“might be,” “could be,” “would love to”). Transitions that flow smoothly but contain no specific information. Word choice that is fluent but unmemorable.

The pattern is fluent prose with no specific knowledge, paired with a generic ask, paired with hedge language designed to handle multiple recipient categories.

Why the Pattern Persists

AI cold outreach exists because the economics work. The cost to generate 10,000 personalized-feeling emails is now under $50 in API fees. The response rate does not need to be high; even 0.1% on 10,000 sends produces 10 conversations. The unit economics support the volume.

Once one outreach team adopts AI writing successfully, others copy. The result is a saturated market for AI-written cold outreach where the patterns become standardized as the underlying tools converge on similar templates.

Recipients learn to recognize the pattern within a few months of receiving the volume. The tell becomes recognizable. Then the AI tools update, the prose becomes harder to spot, and the cycle continues.

The Specific Tells in 2026

Recognizable patterns:

The “I noticed your recent…” opener. The recent thing is usually pulled from a public profile or recent post. The praise is vague because the system does not actually evaluate the quality of what it is praising.

The “given [trend], I thought…” bridge. The trend is plausible because trends are common. The bridge connects the opener to the pitch without saying anything specific about you.

The category-correct pitch. The pitch matches your category (founder, engineer, attorney, freelancer) because the sender filtered by category. It does not match your specific situation because the system does not know it.

The hedge close. “If now is not the right time, I would love to revisit.” Handles the polite rejection case in advance.

The fluent-but-empty middle. Sentences that read fluent but transmit no information. Phrases that handle multiple recipient categories at once.

The optional follow-up. A second email a few days later, also fluent, also generic, also AI-written.

We covered the broader cold-outreach pattern at the anatomy of modern cold outreach and the recruiter-specific version at the recruiter spam epidemic for software engineers.

Why Content-Pattern Detection Is Not the Durable Answer

The tells are recognizable now. They will not stay recognizable.

Generation quality keeps rising. Each generation of language model produces more idiomatic, more contextually aware, more specific-feeling output. The obvious tells of 2024 are gone in 2026; the tells of 2026 will be gone in 2028.

Detection quality is plateauing. Tools that try to detect AI-written text consistently report 60-90% accuracy in 2026, with high false-positive rates. The detection arms race is structurally harder for the detector than for the generator because the generator only needs to fool the detector, not produce true human writing.

Legitimate outreach uses AI too. Many legitimate professional emails are now drafted with AI assistance. Filtering on AI tells produces false positives against real correspondence. The blunt detection approach is too costly.

The cost of generation keeps falling. Even if detection improves, the volume of AI-generated outreach will keep rising because the economics keep getting better. Detection cannot win an arms race where the cost of attack is approaching zero.

The structural conclusion: filtering on stylistic markers of AI writing is a losing battle. The signal is fading; even if you could read it, it would not tell you what you actually want to know.

What You Actually Want to Know

The question that matters is not “did a human write this?” The question that matters is “did the sender care enough about reaching me that they were willing to commit something to do so?”

A sender who paid a small amount to reach you has signaled intention. The prose can be AI-written or human-written; either way, the sender thought you were worth the cost. That is signal you can use.

A sender who would not pay anything is broadcasting at zero cost. The prose can be AI-written or human-written; either way, the sender’s marginal cost of including you was approximately nothing. That is also signal, and it is the signal you usually want to filter on.

The intention filter does not care about the authorship of the prose. It cares whether the sender chose to commit to reaching you specifically.

How a Cover Charge Filter Handles This

The cover charge gate makes the filtering question economic rather than stylistic.

A sender who paid four cents reached the inbox. The email is in the main inbox, regardless of who wrote the prose. AI authorship is fine; the signal is the four cents.

A sender who did not pay is in the held-for-review folder. The email waits there. The recipient can rescue any message they want to read.

The volume drops. Mass-volume AI cold outreach campaigns become uneconomical at four cents per recipient. A campaign at 10,000 recipients costs $400, which changes the math.

Real outreach still arrives. Senders who genuinely want to reach a specific recipient can pay four cents. The cost is small enough that real outreach is unaffected.

Existing relationships are unaffected. Senders on the recipient’s auto-built guest list walk in at no cost. The cover charge applies only to unknown senders.

What This Does Not Do

The cover charge gate is not anti-AI. Three things to be clear about.

It does not detect AI writing. Rythm makes no attempt to identify AI authorship. The filtering is on the payment, not on the prose.

It does not block legitimate AI-assisted outreach. A real sender using AI to draft a thoughtful pitch can pay the cover charge and reach the inbox. The system is indifferent to drafting tools.

It does not eliminate cold outreach entirely. Senders willing to pay four cents per recipient can still reach the inbox. The volume drops because the unit economics change, but legitimate paid outreach still arrives.

The realistic outcome: less volume, better signal-to-noise, and a structural answer that does not depend on staying ahead of AI quality. The filter is on intention, which is durable. The prose can be human or AI; the signal of caring enough to pay is the same.

A Specific Honest Note

AI-generated cold email has recognizable patterns in 2026. You can spot most templates within two paragraphs. The recognition is genuinely useful for the next year or two.

But the durable answer is not pattern recognition. The patterns will fade. The volume will rise. The cost of generation will keep falling. The arms race favors the attacker.

The structural answer is to filter on intention. A sender who paid is signaling something the prose cannot tell you. A sender who did not pay is also signaling something. Both signals are stable; both are independent of AI quality. That is what Rythm filters on.

For the related guides, see the anatomy of modern cold outreach, why “I loved your recent work” is almost always a template, and the recruiter spam epidemic for software engineers. For the broader frame, see why we don’t use AI to fight phishing and what is an email paywall. Rythm is $1.65 per month, cancel anytime.