Finally! A breakthrough with ChatGPT.
With the release of GPT-5.5, I’ve found the first model that actually gets how I want it to write.
Usually, you have to choose between a wall of text or a response that looks like a corporate PowerPoint deck and lame bullet points. AI seems to really think that humans write like R2D2. This time, I just described what I wanted in plain English, and it actually listened.
The Struggle with “Bot-Speak”
In the past, I spent hours meticulously crafting custom instructions. It was a constant game of tug-of-war. If I asked for formatting, the model would go overboard with bolding and headers. If I told it to dial it back, it would swing too far the other way, like giving me bulleted lists where the actual bullets were missing. It never quite felt human.
My goal has always been simple: I want the AI to write like a normal person. I prefer a narrative style with a bit of personality. I’m fine with formatting, but only when it genuinely helps someone scan the page or highlights a crucial point. It should be used sparingly, not as a default setting.
The “Magazine Journalist” switch up
Yesterday, I took a different approach. Instead of rigid rules, I just talked to it. I told it:
assume I have a background in magazine journalism and that shapes my taste in writing.
ChatGPT understood, saved it to its memory, and the results have been spot on.
The problem with the modern internet is that it’s being flattened into a series of “key takeaways” for people with the attention spans a goldfish on adderall. We’ve traded the rhythm of a well-told story for the aesthetic of a middle-manager’s quarterly report. This corporate-sanitized dialect (let’s call it Silicon Valley Generic) is designed to be unoffensive, efficient, and utterly devoid of anything resembling a human pulse. When you ask an AI to write, it defaults to this because it thinks we’ve all collectively agreed to stop being interesting.
AI can write humanly, you just need to tell it
By forcing the AI out of its “Helpful Assistant” persona and into the shoes of a veteran journalist, I finally stopped getting a summary and started getting a perspective. It’s not just about the words; it’s about the cadence, the irony, and the refusal to treat the reader like a data-entry clerk.
It turns out that the secret to making AI sound less like a spreadsheet is to stop treating the interface like a command line. We’ve been so conditioned by the “move fast and break things” era that we forgot that language is an art, not a productivity hack. When you stop asking for “scannable content” and start asking for a narrative, the AI suddenly remembers that it has access to the greatest library of human expression ever assembled, not just the technical manuals.