Line Breaks

What appears to have begun on April 11th is the kind of complaint that never trends but never quite goes away either. A user posted a thread titled simply "Line Breaks" — the subject line of someone who had already tried the obvious things and was now appealing to strangers on a forum. No version numbers, no reproduction steps, just a person who wanted a clean paragraph and kept getting a bulleted list instead. A month later, on May 12th, a second user arrived at the same frustration from a different angle: "How do I stop the bullet point madness?" The phrasing alone tells you something. Not "bullet point issue" or "formatting problem" — madness. The word someone uses when a tool has started making them feel like they are arguing with furniture. Together, the two posts sketch a consistent pattern: users prompting ChatGPT for plain prose and receiving structured output they did not ask for, apparently regardless of how clearly they specified otherwise. Whether this reflects a deliberate model tuning decision — perhaps an attempt to improve readability metrics — or something more incidental in a recent update is not established. What is established is that two separate users, a month apart, hit the same wall and came looking for a door.

If confirmed, here is what this means. The bullet point is not neutral. It fragments continuous reasoning into discrete chunks, which suits a checklist and destroys an argument. Writers, analysts, anyone using ChatGPT as a drafting partner rather than a task manager would find persistent over-formatting genuinely corrosive — not dramatic, not career-ending, just a slow daily tax on usefulness. If the model has been tuned to default toward structured output because that scores better in some internal readability measure, it would represent a quiet but meaningful misalignment between what power users need and what the model has been optimised to produce. The second-order effect is subtler: users who cannot override formatting defaults through natural instruction start to distrust the model's responsiveness to instruction generally. That erosion of trust, accumulated across thousands of small moments, matters more than any single feature failure.

Watch for OpenAI's system prompt documentation or any model update notes that reference output formatting defaults — and watch for whether this thread pattern appears on Reddit or Hacker News, where a larger sample would either confirm a systematic shift or reveal this as two people having a bad week.

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 →