VCs can smell a machine: AI written cold emails to investors, and the five-word fix

Sunny Ray · Jun 10, 2026 · 5 min read

The more of your investor email AI writes, the less of it the investor reads.

Use AI to research investors, not to write to them. Let a model read an investor's LinkedIn, their fund's site and their recent posts, then have it give you five words of genuinely relevant personalization, not a paragraph. Write the rest yourself, short, in your own voice, and read every email before it goes out. Investors see more AI-generated outreach than anyone, and the paragraph-long fake-personal opener is now the fastest way to get filed with the rest.

What a VC said about it, on the record

On June 10 I interviewed a VC with more than twenty years in venture on my show. Near the end, we got to the biggest fundraising mistake founders make right now. I expected something about valuations or round size. His answer was outreach.

Founders are sending AI-written emails without reviewing them, and investors can tell. In his words, he's on the receiving end of

stuff that feels like an attempt to personalize, but that falls very flat and generic because it was so clearly written by, you know, a machine.

His fix was simple. Use AI to draft if you want. Then read every single send yourself and make it sound like you.

Later that afternoon, Amir, Charles and I were on a call with a founder who had pitched at Spark Tank the day before, building in the AI inference space. He raised the exact same worry, unprompted: sophisticated VCs might detect AI-generated email and discount the founder behind it. Two signals, same afternoon, from opposite sides of the table.

Why paragraphs of personalization read as machine-written

Here's what most AI personalization looks like in an investor's inbox. A three-sentence opener that restates the fund's thesis, names two portfolio companies, compliments a podcast appearance and connects all of it to the founder's startup with a phrase like aligns perfectly with your focus.

Every piece of it is technically accurate. All of it together is a tell. No human founder writes three sentences of research before getting to the point. Humans write one specific thing and move on, because they're busy and they assume the reader is too. The longer the personalization, the more obviously it was generated, and the more the investor wonders what else in the email was.

And investors are pattern-matchers by trade. They see hundreds of these a week. Once they've seen the template, they see it everywhere, the same way you can spot a stock photo.

Research deep, write short: the five-word method

What we sketched on the call flips the ratio. The AI does more work, and writes far less.

• Pull deep data on each investor: their LinkedIn, their fund's website, their last ten blog posts or public notes.
• Feed it to a model with one job: find the single most relevant point of connection to your company.
• Output about five words. Not a sentence. Five words that slot into an email you wrote yourself.

So instead of a paragraph about how your vision aligns with their thesis, the email says something like: saw your post on inference costs. Then it gets to the point. Five words carry all the proof that you did the work, and none of the smell that a machine did it.

The rest of the email is yours. Who you are, what you're building, one line of traction, one ask. Written once, by you, in the way you talk, and reused across the campaign with the five words swapped in.

Read every send before it goes

The VC's fix and ours land in the same place: a human reads it. Not a sample, every one, at least while the campaign is young. You'll catch the model's mistakes, the wrong portfolio company, the stale role, the post from three years ago, before an investor does.

I've saved a couple of posts on X this year from operators bragging that AI writes nearly all their cold copy, and another from someone sharing a tool that strips AI patterns out of writing. Put those side by side and you see the arms race. One side generates, the other side scrubs the generation out. The founder who writes the email and lets AI only find the five words skips the whole race.

Layer an invitation on top of the pitch

For that founder we added a second layer. Alongside his direct investor outreach, a podcast invite campaign: instead of asking investors to read a pitch, invite them onto a show to talk about the space.

It solves the AI problem from a different angle. An invitation is short by nature. It doesn't need a paragraph of fake rapport, because the ask is about them, not you. And it changes what the investor is evaluating. They're deciding whether to have a conversation, not whether to fund you, and that's a much easier yes. The pitch comes later, after you've spoken.

The VC I interviewed that day was a guest because he was invited, not pitched. That's the whole idea.

A quick test before you send

• Read the first line out loud. Would you say it to the investor's face at a dinner?
• Count the personalized words. More than ten? Cut.
• Check one fact against the source. If the model got it wrong once, assume it got it wrong elsewhere.
• Delete any phrase you've seen in another founder's email.

If you'd rather invite investors than pitch them, with every message written in your voice and approved before it sends, raise with an invitation, not a pitch.

Frequently asked questions