I've been using ChatGPT wrong for 2 years.
The first signal arrived on March 25th: a blunt, first-person confession that two years of daily ChatGPT use had amounted to something close to sophisticated self-sabotage. Not because the tool underdelivered, but because the person had been asking it to do the wrong thing entirely. Outsourcing production — drafts, summaries, polished replies — when the more valuable move, in hindsight, was to outsource resistance. AI as ghostwriter, when the harder and more useful version is AI as the colleague who tells you your argument has a hole in it. Then, five weeks later on April 30th, a second signal sharpened the point: the same essential reversal, this time framed as a deliberate experiment. Stop asking ChatGPT to write. Use it instead to interrogate, to stress-test, to ask the question you were avoiding. The results, according to the account, got noticeably better. Two anecdotes, same conclusion, one month apart.
If confirmed, here is what this means. The dominant mental model for AI productivity — volume in, polished content out — may be actively degrading the quality of thinking for a significant portion of its most enthusiastic users. The people who feel most fluent with these tools might, paradoxically, be the ones most at risk. When AI writes for you, you lose the friction that forces clarification; when it challenges you, that friction becomes the entire product. The practical implication is uncomfortable: the most effective use of a language model might require you to do more cognitive work, not less. For anyone building AI workflows into editorial, research, or strategy functions, that reframes the whole value proposition. The question stops being "what can AI produce" and becomes "where do I most need to be contradicted."
Watch for workplace studies or structured user research that compare output quality across these two usage modes — interrogative versus generative. Any independent data on that distinction would either confirm this is a real pattern or reveal it as compelling but individual anecdote.
<!-- SC_OFF --><div class="md"><p>I’ve been noticing a pattern with how people use AI tools at work.</p> <p>Not obvious misuse — just normal things like:</p> <ul> <li>debugging
<!-- SC_OFF --><div class="md"><p>For years I assumed bad AI output was just a sign that the model was not good enough. I kept trying different prompts, testing different GTPs,
<!-- SC_OFF --><div class="md"><p>not custom instructions. everyone knows custom instructions.</p> <p>something inside custom instructions that almost nobody uses correctly.</p>
<!-- SC_OFF --><div class="md"><p>Not different in a 'worse/better' way necessarily. More like it has a different texture to how it responds.</p> <p>A year ago, conversations with
<!-- SC_OFF --><div class="md"><p>This AI has completely gone to shit, I ask it about a fucking build in a video game or the things you have to do to platinum a game or what the
<!-- SC_OFF --><div class="md"><p>I've been thinking about how we use ChatGPT. Most of the time, we ask it to solve a problem, improve an idea, or tell us how to do something.</p>
<!-- SC_OFF --><div class="md"><p>They go under settings> personalization.</p> <p>Under “More About You”:</p> <p>I value accurate, well-reasoned information and structured
<!-- SC_OFF --><div class="md"><p>I don’t open ChatGPT to have a conversation.</p> <p>I open it to get a result.</p> <p>But with longer or slightly complex tasks, it almost always
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 →