Writing Content with ChatGPT: The Smart Way to Use AI for Freelance Work

Three months into freelance writing, I had a system. A client would send a topic, I'd paste it into ChatGPT with a decent prompt, skim the output, fix a few obvious things, and send it back within the hour.

It worked for a while. Then a regular client, someone who'd given me four projects already, sent back a one-line message: "This doesn't sound like your other pieces. Can you redo it?"

I was confused, honestly. The grammar was fine. The information was accurate. Nothing was technically wrong with it.

That confusion is what made me actually sit down and figure out what was different.

---

What "sounding like AI" actually means

It's not about errors. AI-generated text is usually grammatically perfect, which is part of the problem. Real writing has rhythm: short sentences mixed with long ones, occasional repetition for emphasis, a specific example instead of a general statement.

The piece I sent that client was smooth in a way that felt, looking back, slightly hollow. Every paragraph was roughly the same length. Every section followed the same "here's the concept, here are three benefits, here's a summary" pattern. No opinion anywhere. No specific story.

My other pieces, the ones the client liked, had little asides. A mention of a tool I'd actually used. A sentence that started with "honestly" because that's how I talk. Those things were missing entirely from the AI draft, and I hadn't noticed until someone else did.

---

The workflow that actually holds up

What changed wasn't whether I used ChatGPT. It's where in the process I use it.

Now the AI handles the research dump and a rough structural skeleton, basically a scaffold I can build on. I ask it for an outline, sometimes a few different angles on the same topic, and occasionally a first pass at a tricky paragraph I'm stuck on.

Then I rewrite almost everything. Not lightly edit, rewrite. I add a real example from something I've worked on or read about. I cut sections that feel generic. I change the structure if all the H2s feel too uniform. And I read it out loud at the end, which catches more awkward phrasing than any amount of silent reading.

This sounds like more work than copy-paste. It is, somewhat. But it's still meaningfully faster than starting from a blank page, and the output doesn't get rejected.

---

The phrases that give it away

There's a specific vocabulary that AI models default to, and once you notice it, you can't unsee it. "In today's fast-paced world." "It's important to note that." "Dive into." "Leverage." "Furthermore." "In conclusion."

None of these are wrong exactly. But stacked together, they create a tone that reads like a corporate brochure rather than a person explaining something they understand.

My fix was almost embarrassingly simple: I keep a short list of these phrases and search for them before submitting anything. If I find one, I either delete it or replace it with how I'd actually say the thing out loud. "It's important to note that pricing varies" becomes "pricing varies a lot, and that's worth knowing upfront."

Small change. Makes a bigger difference than I expected the first time I tried it.

---

Where AI genuinely helps, and where it doesn't

For research and structure, it's genuinely useful. If a client wants an article on a topic I know nothing about, ChatGPT gives me a decent map of the subtopics in about two minutes, something that used to take 20-30 minutes of scattered Googling.

For voice and specificity, it's close to useless on its own. AI doesn't know what I've personally tried, what surprised me, what didn't work. Those are the parts readers (and clients) actually respond to.

There's also a category where AI is actively risky: anything involving numbers or claims that need to be accurate. I've seen ChatGPT confidently state statistics that turned out to be outdated or just wrong. Every number that goes into a client piece now gets checked separately, no exceptions.

---

What changed after that one rejected piece

That client kept working with me, by the way. I redid the piece, this time writing the first draft myself and using ChatGPT only to check if I'd missed an obvious subtopic. It took longer. The client didn't comment on it, which I took as a good sign.

A few months later, a different client actually asked, casually, whether I used AI tools. I said yes, for research and outlines, and explained the rest was my own writing and editing. They seemed fine with it. More than fine, actually. They said most writers either pretend they don't use AI at all or clearly don't edit the output, and that the middle ground was rare enough that it stood out.

I think that's the actual opportunity here. Not hiding that you use AI, and not relying on it either. Just being one of the people who treats it as a tool instead of a replacement, which apparently isn't as common as it should be by now.