Paul Graham: AI-written emails turn off investors

On May 26, Simon Willison gathered on his blog a series of tweets from Paul Graham that have circulated widely across startup and tech communities. The message is straightforward: Graham, cofounder of Y Combinator and a regular reader of hundreds of founder emails, claims he can now easily identify when an email has been written by AI, and when he spots it, he stops reading.

His reason is concrete: AI-generated emails adopt a style he describes as a hard-hitting journalistic style, a forceful journalistic tone that founders never used before. This stylistic uniformity acts as an involuntary alarm signal. Once triggered, the effect is irreversible for that particular read.

The problem isn't AI, it's perceived deception

Graham doesn't criticize AI use in general. What he articulates is something more specific: when an email is signed by a person but written by a machine, the recipient feels misled. In his own words, "it feels like being lied to". That discomfort, he says, leads him to think worse of the author: not only because they can't write without help, but because they're trying to hide it.

This distinction matters. It's not an argument against AI as a work tool, but against its use as a substitute for your own voice in contexts where that voice is exactly what's being evaluated. An email to an investor or YC partner isn't an administrative task; it's a presentation of who you are and how you think. Outsourcing it entirely inverts the purpose of the exercise.

Why this especially affects the startup ecosystem

In the world of early-stage startups, written communication is one of the few direct channels between founders and investors. There's no mature product that speaks for itself, nor a long track record of traction. What exists is the founding team's ability to explain their problem, their solution, and their reasoning. A well-written email can open a conversation; one that sounds like a generated template apparently closes it before it starts.

Graham adds a point worth attention: "it's not impressive to use AI to write stuff for you; any teenager can do that". In a context where differentiation is critical, using AI for personal communication offers no perceived competitive advantage. On the contrary, it puts the founder on equal footing with any casual chatbot user.

The broader context: signal and noise

This episode illustrates a tension that repeats across different professional fields: AI enables the production of apparently quality text with minimal effort, but that same ease erodes the signal that text was meant to convey. When everyone can generate a flawless email in ten seconds, the flawless email stops being a signal of competence. What remains as signal is precisely the ability to write with your own voice, with judgment, with something that can't be copied with a prompt.

It's not a phenomenon exclusive to the investment world. In hiring processes, in client proposals, in editorial pitches, the same pattern begins to emerge: experienced readers develop a detector for synthetic style, and when it activates, the content loses credibility regardless of its technical quality.

Who should pay attention to this insight

Primarily founders who use AI as a shortcut in high-stakes relational communications. But also any professional who depends on writing as a way to convey their own judgment: consultants, journalists, candidates for technical positions, proposal authors.

The tool isn't the problem. Using it to refine structure, correct errors, or translate already-formed ideas is different from using it to generate from scratch a text that should reflect your thinking. The difference is perceptible, at least to readers with sufficient reading volume.

---

From our perspective at ClaudeWave, Graham's observation strikes us as reasonable and well-calibrated: he doesn't reject AI, he rejects impersonation. For those of us building tools on Claude, it's a reminder that real utility lies in amplifying human judgment, not in replacing the voice of the person who signs.