Asked ChatGPT to roast me

It started on March 26 with someone handing ChatGPT a peculiar kind of permission: go ahead, mock me. The post that followed had that particular bruised-delighted quality you only get when something lands too accurately, when the joke contains just enough truth to sting. Then April 8 arrives, and the pattern doesn't just repeat — it mutates. Two posts surface almost simultaneously. One person asks ChatGPT to roast them directly, which is the expected escalation. The other does something stranger and more revealing: they feed it a love letter. Not themselves, but the most earnest thing they've made. The thing they wrote for someone else. By May 3 the story had doubled again, with "Chatgpt roasted me" and "Asked GPT to Roast Me... It Didn't Hold Back" appearing on the same day — that last title doing quiet work, because the surprise encoded in it suggests people keep expecting the model to flinch and keep being wrong about that. May 7 brings someone burned by a response they describe as genuinely cutting. May 10 reframes the whole thing as entertainment, with a video game cover given the roast treatment. By May 13, someone is back at the beginning, asking the same question the March 26 person asked, as if the loop is just... running.

If confirmed as a durable behavioral pattern rather than scattered novelty posts, here is what this means: people are actively seeking friction from a system that was largely designed to avoid it. The roast prompt is, at its core, a test of whether the model will tell you something true that a polite system wouldn't — and the fact that people keep returning to it, keep being surprised by the results, suggests that ChatGPT is passing that test often enough to sustain the behavior. The second-order effect is more interesting: when you roast a love letter, you are asking an AI to evaluate your emotional authenticity. That is a different relationship with the technology than productivity or assistance. It is closer to confession, or therapy, or the specific vulnerability of asking a stranger whether you sound like an idiot. If enough people are using the model this way — not for answers, but for a mirror that will actually show them something — the product implications for OpenAI are significant. The most-used feature might not be the one anyone designed.

Watch for whether OpenAI's safety or tone guidelines begin shifting in response to roast-style prompts — any softening or hardening of how the model handles self-directed mockery requests would signal that the pattern has grown large enough to require a policy response.

NewsHive monitors these sources continuously. All signal titles above link to the original reporting.

Intelligence by NewsHive. Need help navigating what this means for your business? Contact GeekyBee →