How Prospects Spot AI-Written Cold Emails

#How Prospects Spot AI-Written Cold Emails

June 12, 2026•FirstSales Team•9 min read read

TL;DR: Buyers in 2026 have seen so many AI-generated cold emails that they now pattern-match them reflexively. The tells are predictable: em-dash slop, over-polished structure, fake personalization that references a LinkedIn post you wrote three years ago, and an opener that starts with your company name. Each tell independently signals "AI wrote this." Stack two or three of them and your email is deleted before the second sentence. This article breaks down the exact patterns - and how to write around them.

#Why This Became a Problem in 2026

The cold email landscape changed fast. A year ago, AI-assisted drafting was a competitive edge. Today it is the default - which means the inbox your prospect opens every morning is full of emails that were generated by the same handful of models, using the same prompts, with the same sentence structures.

Buyers figured this out. In June 2026, "Shunning AI is the human choice" was trending across sales communities with hundreds of upvotes, reflecting a growing buyer backlash against obviously robotic outreach. The emotional undercurrent is real: people do not feel good about knowing a machine decided to reach out to them. They feel like a row in a spreadsheet.

The practical result is that prospects now have a five-second pattern-match running every time they open a cold email. It is not conscious analysis. It is the same reflex that lets you spot a phishing email before you have read the full subject line. Once the pattern fires, trust collapses and the email is gone.

The stakes are higher than a single delete. A reader who spots the AI tell once does not just ignore that email - they update their mental model of your brand. If they see it twice, they hit "Report spam" instead of "Unsubscribe." That complaint contributes to domain reputation damage that affects every email you send afterward.

If you want to understand how the reply-rate gap between AI-generated and human-edited email actually plays out, read AI vs Human Cold Email: The Reply-Rate Truth. The short version: quality and relevance beat volume, and the gap is measurable.

#Tell 1: The Em-Dash Signature

This is the single most reliable AI tell in written copy right now, which is why this article uses none of them.

AI language models are heavily trained on formal writing and have developed a strong stylistic preference for em-dashes to connect clauses. A human writer might use one or two in a thousand-word document. An AI draft will use three in a four-sentence email.

Example of what this looks like in practice:

"I noticed you're scaling your sales team - which is exactly why I'm reaching out - our platform helps reps like yours close 40% faster - and I'd love to show you how."

Any experienced buyer reads that sentence and thinks: AI wrote this. The tell is not one em-dash, it is the rhythm of them. It reads like a machine that discovered punctuation and liked it too much.

The fix is mechanical: before you send any draft, grep for em-dashes. Zero tolerance. Replace them with periods, commas, or "- " (hyphen with spaces if you genuinely need a parenthetical). If you are reviewing AI-generated copy, the em-dash count is your fastest quality signal.

#Tell 2: The Over-Polished Structure

Human-written cold emails are slightly messy. They have odd line breaks, they use sentence fragments, they get to the point in a crooked way. They sound like a person typing a message.

AI drafts are structurally perfect. Three sentences of context. One sentence of pain point. One sentence of solution. One CTA. Every time. The cadence is so consistent that a reader who has seen ten AI cold emails recognizes the eleventh instantly by its shape, before processing its content.

This structural tell is harder to fix than the em-dash because it requires actually rewriting the message rather than doing a find-replace. The goal is to introduce a small amount of deliberate imperfection: a short sentence where a long one was expected, a genuine aside, a self-aware acknowledgment that this is a cold email.

Contrast these two:

AI structure:

"Hi [First Name], I came across your profile and was impressed by your work at [Company]. Many companies in your space struggle with [generic pain]. We help solve that with [product]. Would you be open to a 15-minute call?"

Human structure:

"Hi Sarah - quick question before I pitch anything. Does your team still handle the prospecting in-house, or did that change when you moved to the new org structure? Asking because what we built is specifically for teams at your growth stage and I want to check it is actually relevant before I say more."

The second is longer and less polished. It is also more likely to get a reply because it sounds like a person who thought before sending.

If you want a detailed breakdown of what bad structure looks like in practice, bad email examples has annotated cases worth studying.

#Tell 3: Fake Personalization

Personalization is supposed to signal that you did your homework. AI-assisted personalization has inverted this: it now signals that a machine scraped your LinkedIn and inserted a reference to prove it was watching.

The classic form is the "I saw your post about X" opener, where X is either so generic it could apply to anyone ("your post about leadership") or so specific it is clearly the result of a data scrape ("your post on April 14th about supply chain challenges in Q1"). Both versions feel wrong. The generic one proves you did not read anything. The hyper-specific one proves a machine did the reading for you.

Real personalization requires actually knowing something worth knowing. It requires a reason to reach out that is specific to why this person, at this company, right now. That reason needs to come from human judgment, not from a variable field.

The tell is usually in the first sentence. When you read "I noticed you recently [scraped data point]," the brain's pattern-match fires immediately. The word "noticed" is particularly diagnostic: humans do not typically narrate their noticing. AI drafts do it constantly.

For a deeper look at how personalization connects to buyer trust and conversion, the piece on personal branding, cold email, and trust conversion covers the underlying mechanics.

#Tell 4: The Generic Opener

The vast majority of AI-generated cold emails open in one of a small number of ways:

"I came across your profile and..."
"I noticed [Company] is [scaling / hiring / growing]..."
"My name is [Name] and I help companies like yours..."
"Hope this finds you well..."

Buyers have seen every one of these hundreds of times. The opener is the email's first test, and these openers fail it immediately. The pattern-match fires in the first five words and the rest of the email is already dismissed.

The best human openers are specific enough that they could not have been generated by a prompt. They reference something only someone paying attention would know. They ask a question that assumes the reader has context. They skip the setup entirely and land on the point.

The opener test is simple: could this first sentence appear word-for-word in an email to a different person at a different company? If yes, it is too generic to work.

#Tell 5: The Vocabulary Cluster

AI language models have a set of words that appear at much higher rates than in natural human writing. Alone, each word is harmless. Together, they form a vocabulary fingerprint that readers recognize without being able to name it.

The most common ones in cold email:

"leverage" (humans say "use")
"robust" (humans say "solid" or just skip the adjective)
"seamless" (nobody says this in conversation)
"streamline" (same)
"elevate" (this one is particularly diagnostic)
"cutting-edge" (tired and robotic)
"comprehensive solution"
"pain points" used as a standalone noun

Run your draft through a quick mental check: would a person actually say this sentence out loud in a phone conversation? If the answer is no, rewrite it.

The how to write cold emails guide covers voice and tone in more depth, but the core principle here is simple: write the way a smart person talks, not the way a business document reads.

#Tell 6: The Compliment Sandwich

AI-generated outreach frequently opens or closes with a compliment about the company, the founder, or the product. "What you've built at [Company] is impressive." "I love what you're doing in the [industry] space." "Your recent funding is a real testament to the team."

These compliments are formulaic and the reader knows it. Nobody believes a stranger who opens with "I love what you've built" unless there is specific evidence they actually used the product or followed the work. The generic compliment reads as a manipulation tactic and triggers distrust rather than warmth.

If you genuinely have something specific to say - you actually used their product, you heard the founder speak at a conference you attended, you read a specific piece of writing they published - say that specific thing. If you do not have a genuine compliment, skip the opener entirely and go straight to your reason for writing.

#What Happens When Trust Collapses

The consequence of triggering the AI pattern-match is not just a delete. It plays out across several stages that compound.

First, the email does not get read past sentence two. Second, the reader may mark it as spam rather than ignoring it - this is the response that actually damages your domain reputation. Third, if they forward it to a colleague as an example of bad outreach (this happens more than senders realize), your company is now associated with the behavior you were trying to avoid.

The June 2026 trend data is clear on this: buyer hostility toward obvious AI outreach is not softening. The "Shunning AI" sentiment is not a fringe position - it is a mainstream buyer response that sales and marketing leaders are now actively discussing. Recipients have developed enough pattern recognition that the bar for what reads as "human" has risen sharply in the last twelve months.

This compounds the deliverability risk. When complaint rates climb because recipients are flagging AI-patterned emails as spam, domain reputation suffers. If you are not familiar with how the complaint ceiling works and what it does to your sending reputation, the 0.3% spam complaint ceiling breaks down the mechanism.

#How to Write Cold Email That Reads Human

The practical checklist, in order of impact:

Remove every em-dash. Zero tolerance. Search the draft before it leaves your queue.

Read the opener out loud. If it sounds like a LinkedIn post or a press release, rewrite it. The test is whether a smart colleague would write this sentence in a Slack message.

Check the personalization claim. For every "I noticed" or "I saw" in the draft, ask: is this something a person actually noticed, or did a machine pull it from a database? If it is the latter, either replace it with a real reason or remove it.

Kill the vocabulary cluster. "Leverage," "seamless," "robust," "elevate" - replace each one with the word a person would actually use in speech.

Introduce imperfection deliberately. A sentence fragment. An aside in parentheses. An acknowledgment that the email is cold and the reader owes you nothing. These small imperfections are trust signals, not weaknesses.

Have a human read it before it sends. This is not optional. AI can draft efficiently; human judgment catches the patterns that AI cannot see in its own output. This is exactly the model that AI Drafts, Human Sends: The Hybrid Outbound Model is built on - and it is the approach that closes the reply-rate gap.

The goal is not to hide that AI was involved in drafting. The goal is to ensure that a human judgment layer has reviewed and edited the output so that what arrives in the reader's inbox is something a person actually stands behind. That distinction matters to buyers, and they can feel the difference.

For the full picture of how the best cold email outreach tools handle the AI-to-human workflow, that comparison is worth reading if you are evaluating your stack.

#FAQs

#Can buyers actually tell the difference between AI and human cold email?

Yes, reliably. Buyers in 2026 have seen enough AI-generated outreach to recognize the patterns - em-dashes, generic openers, fake personalization - within the first two sentences. The pattern-match fires before they consciously analyze the content.

#Does AI-written cold email always perform worse than human-written?

Not always, but the gap appears when AI output is unedited. A human who reviews, edits, and personalizes an AI draft closes most of the performance gap. Fully automated sends with no human review tend to underperform on reply rate and hurt domain reputation through higher spam complaints.

#What is the single fastest fix for AI-sounding copy?

Remove all em-dashes and read the opener out loud. If you would not say the first sentence in a phone conversation, rewrite it. These two steps catch the most common tells without a full rewrite.

#Does personalization help if it comes from a data scrape?

Scraped personalization - "I saw your post about X" where X is from a database - often makes things worse than no personalization at all. It signals that a machine found the reference, which undermines the trust personalization is supposed to build. Real personalization requires a genuine, judgment-based reason for reaching out.

#How does AI detection affect email deliverability?

Indirectly but significantly. Emails that read as AI-generated get marked as spam at higher rates by recipients. Higher complaint rates damage domain reputation, which reduces inbox placement for all your sends - not just the flagged emails.

#Is there a way to use AI for drafting without triggering the AI tell?

Yes. The key is a human review and edit step before sending. AI can generate a strong first draft quickly; a human then catches the vocabulary tells, rewrites the opener, removes em-dashes, and adds the specific, judgment-based personalization that makes the email feel genuinely written for that recipient.

#Conclusion

The buyers you are trying to reach have become very good at spotting AI-written cold email, and their response when they spot it is not neutral - it is negative. The em-dash slop, the over-polished structure, the fake personalization, the vocabulary cluster: these are not subtle signals. They are bright red flags that collapse trust before you have had a chance to make your case.

The fix is not to stop using AI for drafting. It is to keep a human in the loop. Someone who reads the draft, catches the patterns, rewrites the opener, and takes responsibility for what lands in the reader's inbox. That is the approach that keeps reply rates up and complaint rates down.

FirstSales is built around exactly this model: AI generates a personalized draft, you review and approve it before it sends. You get the drafting speed of AI with the judgment and accountability of a human sender. Try it for $1 for three days and see how it handles your specific prospects and copy style.

The inbox is not getting less competitive. But it is getting better at filtering out the emails that were never really written for the person reading them.

#Keep reading

Cold Email Personalization at Scale: The 2026 Playbook