Paul Graham says AI outreach emails are deceptive

Paul Graham says he stops reading cold outreach emails when he recognizes machine-generated prose and calls polished, founder-signed messages produced by AI deceptive.

Paul Graham, co-founder of startup accelerator Y Combinator, wrote on X that AI-written cold outreach emails are deceptive and that he stops reading them when he recognizes machine-generated prose.

In a string of posts, Graham described a “hard-hitting journalistic style” as the tell. He wrote the giveaway is not the subject line or the pitch but the prose itself: too polished and precise for the typical pressure-filled writing most founders produce.

Graham wrote that he immediately stops reading messages that fit the pattern. “I have never knowingly finished reading an email signed by a human but written by AI. It feels like being lied to, and who would stand for that?” he added.

He framed using AI for personal outreach as a form of deception rather than a harmless efficiency tool. Graham argued a message that mimics a human voice while being machine-generated becomes manipulative regardless of the sender’s intent and described delegating writing to AI as a signal that a founder cannot communicate independently, saying “even a teenager can do it.”

Graham drew a line between acceptable uses of AI and deceptive substitution, telling followers to “use it correctly, like any technology.” He has previously praised AI for helping some Y Combinator startups grow, and distinguished that kind of automation from using AI to impersonate an individual’s voice in outreach.

His comments reflect a change in startup outreach as more professionals receive AI-generated messages. Graham wrote that founders who craft their own emails now stand out by default and noted that some investors have become wary of pitches that favor presentation over demonstrated judgment.

Founders and fundraisers are increasingly using generative tools to scale outreach. Recipients who handle large volumes of messages may stop engaging with outreach that displays a recognizable machine-written style, and some observers warn automated prose can obscure how much the sender actually participated in the outreach.

Articles by this author