What I Learned Writing Articles With AI For a Company That Trusted Prompts More Than Writers

Freelance writers are used to adapting. We adapt to tone guides, editorial preferences, changing SEO strategies, and the occasional editor who insists every article needs three subheadings and a quote from a “source” that may or may not exist. Adaptation is part of the job.

Recently, I experienced something different: writing for a lifestyle website that relied entirely on AI to generate its articles—and trusted its prompts more than the writers it hires.

In theory, the job sounded simple. The listing said articles would take about fifteen minutes to produce. Writers would use the company’s prompts, generate the article, tidy it up, and submit it. In practice, it didn’t work like that at all.

The Illusion of the “15-Minute Article”

The promise was efficiency. AI would do the heavy lifting, and writers would essentially polish the result. But anyone who has actually worked with AI-generated content knows the truth: AI drafts are rarely finished pieces. They’re starting points, which are sometimes very rough starting points.

The articles produced by the company’s prompts were full of repetitive phrasing, strange narrative quirks, and oddly generic storytelling. One phrase in particular showed up constantly: “I remember…”

When every article starts the same, readers will know something is fishy.

Every article seemed to contain a personal anecdote at the beginning, even when the topic didn’t remotely call for personal reflection. A piece about organizing your kitchen? “I remember the first time I opened my pantry and realized…” A piece about choosing throw pillows? “I remember when I first discovered…”

It gave every article the same artificial voice. After a while, they all sounded like the same person reminiscing about everything from scented candles to laundry detergent.

Plus, every writer used the same prompts, so the articles weren’t just repetitive individually—they were repetitive across the entire site. And, yes, this was one of the things they wanted us to edit, BUT that "I remember" statement was IN THEIR PROMPT! Why put it in there if you don't want AI to use it. I don't know much about AI myself, but I know that it's going to do some of the things you tell it to... so why put in stuff you don't want?

The Prompt Problem

The company insisted writers use their prompts exactly as written. They supply their 59-page prompt (I kid you not), and then give it the article title and let it go. Part of the issue was in the titles (like, come on, no one wants to read a clickbait article with a head that's 12+ words long).

From a writer's perspective, this is like giving several authors the same paragraph and asking them to produce unique essays—but not allowing them to change a single word in the instructions.

When writers noticed patterns in the AI output—repeated phrasing, predictable structure, awkward storytelling—the obvious solution would have been simple: Adjust the prompts.

If writers could experiment with different phrasing, add stylistic direction, or shape the instructions to create more varied results, perhaps things would work out better. Instead, the company handled prompt changes internally and only made tiny adjustments. And, if your AI wasn't cooperating, it's your fault as the writer, even though you didn't write the prompt.

The Editing Reality

Because of these problems, the “15-minute article” never existed. To turn those AI drafts into something readable required serious editing:

• Removing repetitive phrases
• Rewriting awkward (and incomplete) sentences
• Fixing tone inconsistencies
• Cutting generic filler paragraphs
• Replacing obviously fabricated anecdotes

... and don't get me started on the sourcing

By the time the editing was finished, each article took closer to an hour to complete... which equalled out to about $12 an hour in pay (which is lower than minimum wage in the state I live in).

The expectation remained that they should be finished much faster. That kind of pressure creates a strange paradox for writers: work quickly, but somehow also fix everything the AI got wrong.

The Feedback That Wasn’t Feedback

The most difficult part of the experience wasn’t the AI itself. It was the lack of communication (I spent five days being soft ghosted before they finally decided to end my contract).

If an editor returned an article, the explanation was often simply: “I fixed it.” That’s not feedback.

Writers improve by understanding what needs to change. Without that information, a writer can’t adjust. And, instead of working with me, they decided that it was too much work to explain exactly what they were looking for.

I'm not mad. I learned a lot from the experience. I mostly learned that a site I wrote for two years ago went from a credible lifestyle site to a place filled with garbage articles that have no human content in them.