Cold Email AI: What Twitter Got Right

I check X every week to see what the cold email conversation looks like from the outside. Most of it is noise. Some of it is genuinely useful. And a small percentage is so wrong it's worth correcting publicly.

This week had all three. Here's what caught my eye.

The Cursor Founder Sends Cold Emails Like an SDR. Good.

This is the post of the week for me, and it has nothing to do with cold email tactics. The Cursor founder built a $60 billion company and got there by doing exactly what most "smart" people refuse to do: sending cold outreach at scale, accepting rejection, and doing it anyway.

The line that hits hardest is the framing around credentials. People with impressive backgrounds often treat cold outreach as beneath them. They'd rather wait for a warm intro. They'd rather build an audience first. They'd rather do anything except sit in their apartment and send email after email into the void.

I've sent millions of cold emails personally. The rejection rate is real. You send a hundred, you hear back from five. You send a thousand, you get twenty-five conversations. That math is not glamorous. But it is reliable. The Cursor founder understood the math and did the work anyway. That's the whole lesson.

The Follow-Up Is Where the Money Lives

Completely accurate and chronically underappreciated. In my experience running campaigns for 14,000+ entrepreneurs, the initial email books maybe 20 to 30 percent of the total meetings generated by a sequence. The rest come from follow-ups.

When someone shows interest in a $50,000 service and goes quiet, you don't stop. You follow up until you get a clear yes or no, even if that takes months. Most people give up after one or two touches because they're afraid of being annoying. The ones who close deals are the ones who stay in the game longer than everyone else. If you want a system for this, I put together a detailed breakdown at /followup.

AI Wrote the Whole Email. That's the Problem.

The core data point here is right even if the delivery is a gated lead magnet. Full AI-written emails read like full AI-written emails. Prospects feel it instantly. One AI-generated personalized line dropped into a tight human-written frame? That works because the personalization feels earned without making the whole message feel synthetic.

LinkedIn posts as a personalization source is a legitimate find. Someone who posted about hiring challenges last week is a different prospect than someone who posted about a product launch. Timing the signal to the message is where relevance actually comes from, and relevance is the only variable in cold email that consistently moves reply rates. I cover the AI personalization question in more depth at /coldemailgpt.

A CEO With $8 Billion Replied in Under Two Hours

Not surprising to me at all, and this is something I've said for years: senior executives often respond faster than mid-level gatekeepers. They have the authority to make decisions, so they can reply immediately without checking with anyone. They also tend to have lean inboxes because they've trained everyone around them to filter. One genuinely relevant email at the right time and you're in the conversation.

The lesson is not that rich people are accessible. The lesson is that nobody is actually unreachable. The bar is a message that feels like it was written specifically for them. Most messages don't clear that bar, which makes yours easier to stand out.

Optimize for the Reply, Not the Meeting

This is exactly right and I call the underlying principle the Simple Ask. The goal of the first email is not to book a meeting. It's to get a reply. One small, low-friction ask creates a conversation. Once you're in a conversation, you can sell. Trying to book a thirty-minute call in the first message is asking someone to make a big commitment to a stranger, and most people won't do it.

The AI-powered follow-up layer on top of engagement is the smart extension of this. You identify who replied, then you use AI to build a hyper-specific response for that exact person and company. Most reps do the opposite: they spend hours personalizing the initial blast and send zero-effort follow-ups to everyone who actually engaged. That's backwards. You can find the cold email scripts I use to execute this kind of two-step approach at /top5scripts.

What a Real Prospect Thinks When Your Email Lands

Read this twice if you're sending cold email right now. This is a real buyer telling you exactly what filters he uses to delete your message before he reads it. Unsubscribe links trigger immediate removal. The "just say yes" CTA gets an instant "no" reply. Funding-news openers get archived.

The frustrating part of this post is the last paragraph: an actual human-written email that makes sense in context might still get a reply. The bar has not disappeared. It has just gotten easier to clear because everyone else is sending garbage. One well-written, relevant, human-sounding message in an inbox full of AI slop is not invisible. It's the only thing that gets read.

From 530 Cold DMs to 2 Clients, Then Something Smarter

530 cold DMs for 23 replies and 2 conversions. That's a 4.3% reply rate and a 0.4% conversion rate. Not great numbers, but honestly not far from what a lot of people see early on when their list quality, targeting, and messaging aren't dialed in yet.

What's smart here is the pivot. Building in public created inbound, which has higher conversion rates because the lead already knows who you are before they reach out. The honest part of this post is what most people skip: even with a strong inbound channel, dry spells happen, and when they do you go back to cold outreach. That's the right answer. Cold email is not plan A or plan B. It's the floor you can always stand on.

Cold Email Is Dead. Again.

Cold email has been declared dead every single year since I started doing this. The logic is always the same: inboxes are full, spam is rising, filters are getting smarter. And the conclusion is always wrong.

The confusion here is between bad cold email and cold email itself. 30 spam calls a day is a deliverability and targeting problem, not a channel problem. If someone sends you one relevant, well-timed, personalized message about a problem you actually have, you are going to read it. That's human nature. It does not change because Gmail added an AI summary tab.

The Meta ads and UGC creator play works, but it requires budget, time to build creative, and a feedback loop that takes months. Cold email can book you a qualified meeting this week. Both channels belong in a mature B2B stack. Declaring one dead because the other exists is how people miss pipeline.

The companies building on cold outbound right now and doing it well - targeting sharp, copy tight, sequences short - are not standing on a cracked foundation. They're standing on the only channel with a measurable, repeatable cost per meeting.

Relevance Is No Longer Expensive to Produce

This is one of the sharper takes of the week and it maps to what I've seen across thousands of campaigns. The consulting industry made a directionally correct observation - spray and pray doesn't work on senior buyers - and drew the wrong conclusion from it. They said the channel was broken. The channel was fine. The messaging was broken.

The structural shift is real: relevance used to cost hours of research per prospect. Now it costs seconds. A tool like Clay can pull a prospect's recent hiring activity, funding news, tech stack, and LinkedIn posts, then feed all of it into a prompt that generates a contextually accurate first line. You're not faking personalization. You're automating the research that used to make personalization impossible at scale.

The 4X reply rate claim for signal-referenced outreach over generic email matches what I see in practice. Get your lead source, targeting, and triggers right before you touch the copy. Everything else follows from there.

How Cold Email Has Actually Changed

The infrastructure changes here are accurate. Fewer emails per inbox per day is the single biggest operational shift in the last few years. The old mindset was volume through volume. The current reality is that inbox reputation matters more than send count, so you protect deliverability by keeping volume controlled per domain and verify contacts twice before anything sends.

The move from VAs writing personalization to Clay for data enrichment and AI for first-line generation is a real upgrade in both speed and consistency. VAs were the bottleneck and the quality was uneven. AI at scale with good data sources is more reliable once the prompts are dialed in. For a full breakdown of what the current stack looks like, I put it together at /coldemailtechstack2025.

One thing missing from this comparison: the offer. The tech changed, the targeting changed, the personalization changed. The founders who still struggle despite having the right tools are usually running them against the wrong offer. Fix that first.

The One Metric That Actually Matters for Sell-the-Reply Campaigns

Short and correct. 50 to 75 words is real. I've tested long emails against short ones more times than I can count and the short version wins consistently. The 2:1 you-to-I ratio is a useful heuristic for catching emails that are too focused on the sender. And selling the reply instead of the meeting is the core principle that separates campaigns that generate conversations from ones that generate silence.

The "read like a text message" framing is worth emphasizing. If your email has more than one paragraph before the ask, you're writing for a prospect who doesn't exist. Real buyers are reading on a phone, between meetings, skimming. Write for that person. The scripts I use are built around exactly this principle - you can find them at /newemailscripts.

Seven Rewrites and Still No Job. But the System Works.

This is funnier than it's meant to be, but it also contains a real problem worth naming. Six replies out of 200 emails is a 3% reply rate. That's not a copywriting problem at this point. That's a targeting and offer problem.

If you're emailing 200 YC startups with the same ask - hire me - you're treating a diverse set of companies at different stages, with different hiring needs, as one homogeneous list. The email optimization matters, but no subject line tweak turns a generic job ask into a compelling offer for a company that wasn't hiring in the first place. The fix is not version eight. The fix is a shorter, more targeted list with a more specific value proposition for each subset of it. Seven rewrites is discipline. Targeting the wrong list seven times is expensive practice.

The Takeaway This Week

There's a theme running through almost every post worth paying attention to this week. The channel is not the problem. The way most people use it is the problem.

The Cursor founder showed that volume plus humility plus persistence builds billion-dollar companies. The senior buyers posting about their inboxes confirmed that one relevant, human message still cuts through. The infrastructure comparison showed the tools have improved dramatically. And the "cold email is dead" crowd is making the same argument they make every year while people who actually run campaigns keep booking meetings.

The thing that has changed is the floor. Generic spray-and-pray is now effectively worthless because the filters are too good and the noise is too high. But contextually relevant, tightly targeted, short and human-sounding outreach is working better than it ever has, because there's less competition for that specific thing.

Your job is not to send more emails. Your job is to send fewer emails to better-fit prospects with a message that feels like it was written for one specific person. AI makes that cheaper to do. The question is whether you're actually doing it.

If you want the exact frameworks I use to build campaigns that convert, start with the templates at /killercoldemails.