The Month Everyone Became a Prompt Engineer

Written in 2026, looking back at the GPT-4 launch and the flood it started, knowing now how the flood played out.

GPT-4 shipped on March 14, 2023, and within about a week my inbox changed shape. Not more emails, different emails. Suddenly half my clients had discovered that a chatbot could produce a passable blog post in thirty seconds, and the question arriving from all of them, phrased a dozen polite ways, was the same underneath: so, can’t ChatGPT just write it?

It was a fair question. It deserved a real answer, not a defensive one. Working out that answer, in real time, in front of people who were paying me, was one of the more useful months of my career.

The flood was instant

Within weeks the internet started filling up. You could feel it before any tool measured it. Blog posts that were technically fine and completely weightless. Product descriptions with that smooth, confident, slightly plastic texture. Whole content calendars, clearly, being run through a prompt and pasted out.

I understood the temptation completely, so I want to be careful here: my opinion is about the approach, not the people reaching for it. If you have thirty pages to fill and a small budget, a tool that fills them for free looks like a gift. The problem wasn’t that people were lazy. The problem was that the tool was very good at producing text and had no idea whether the text was worth producing. Those are different skills, and in early 2023 an enormous number of people, myself included at first, hadn’t fully separated them in their heads.

How I answered the question

The honest answer I landed on, and kept giving all year, went roughly like this. Yes, ChatGPT can write it. It cannot know it. It has never used your product, never sat with your customer, never watched the thing go wrong at 3am and had to fix it. It can arrange words that sound like experience. It cannot have the experience. And on the open web, where a million pages now sound equally fluent, the only thing left that’s actually scarce is the part the tool can’t fake.

So my advice, from about March onward, stopped being write more and became write from something only you have. Use the tool for the scaffolding, the outline, the tidy-up, the second draft. Do not use it for the part that’s supposed to be you. If a page could have been generated by anyone with the same prompt, it will end up competing with everyone who used that prompt, which is to say everyone.

Some clients heard this and were relieved. A few heard it and quietly went off to generate two hundred pages anyway. I made a private note to check back on both groups in a year.

What I’d add now

I checked back. The clients who leaned into pure volume mostly got flattened later that year, when Google’s Helpful Content work arrived and did exactly what you’d fear. The ones who used the tool to move faster on real, first-hand material did fine, some better than fine.

The thing I got slightly wrong in March 2023 was the tone. I was a bit too worried, a bit too much in defend-my-job mode. The tool didn’t take the job. It took the easy half of the job and left the hard half more valuable than before. That’s a better outcome than I let myself believe that spring, and I wish I’d said it with less of a flinch at the time.