AI Did It! The First 4 Blogs

When I set out to write the first set of blogs for OXI Media, I knew traction might be low. Starting a new business is rarely a big-bang event. Getting eyeballs on content was always going to be a challenge.

I settled on the RAM crisis as a focus area because it was topical and a good fit for what Rich Narain and I were working on. It gave us an opportunity to build some leverage around the subject matter and express an opinion.

Our opinion, in case you did not read those posts, was that the broader chip crisis, not just RAM, was getting worse and that recovery would take a long time because of the AI boom and geopolitical events.

I did have another agenda tied to this work. With all the AI slop in the world, could anyone tell what was AI-generated and what was not in those first posts? So, I came up with a small experiment to find out for myself.

What Was Generated by AI

The first blog was written entirely by me and used only Microsoft Word spellcheck.

For the second blog, I asked ChatGPT to apply AP style and fix spelling and grammar. It was keen to make other edits, but I chose not to include them.

For the third, I ran it through ChatGPT again with AP style, spelling and grammar checks, but this time I accepted almost all of its additional changes.

For the fourth, I used AI to create almost the entire piece. I asked Gemini to find the latest news on Samsung’s planned strikes in South Korea and compare union demands with competitors such as Micron and SK Hynix.

I took the entire response, pasted it into ChatGPT, and asked it to apply the same adjustments as blogs two and three while adding relevant information about the impact on the media and entertainment industry.

The result was copied and pasted into the blog I posted yesterday, May 18, 2026, and I added a simple description to finish it off.

What I Learned

The result was better than expected.

ChatGPT does not have a good command of my voice yet. More on that another time.

If I had simply taken the Gemini output and turned it into a blog post, it would have been obvious I made no effort, versus the admittedly low effort I did put in by running it through ChatGPT.

I do not care for the AI-generated blog as much as the others. While the facts are accurate, it is dry and does not spend much time on the human element affecting Samsung workers.

Had I taken the same approach as the first three blogs, I would have included more data, such as likely bonus comparisons per worker, the perspective of South Korean courts partially blocking the scope of the planned strikes over concerns about damage to facilities, and the possibility of government intervention that could suspend the strikes for 30 days because of the potential economic impact.

The Time Spent

No surprises here: the first three blogs required real effort. Hours of research and fact-checking, more time to write and verify, and then someone to proofread my handiwork before I finally posted the results.

In contrast, I spent more time generating an AI image I liked for the fourth post than I did on the prompt, post-AI cleanup and reading the output.

Did Anyone Notice?

Turns out Rich knew something was up with the fourth blog. He did not know it was AI-generated until I told him.

Can you tell the difference? Drop me a line at awarman@oximedia.net or DM me on LinkedIn: Andrew Warman on LinkedIn