Last updated: July 31, 2026
So you picked your niche. Now comes the part where it gets real: writing the first batch of articles. And if you’ve researched this step at all, you’ve already run into the warning that hangs over it: low-effort AI content reads like mush, readers skip it, and Google added a spam policy in March 2024 (scaled content abuse) aimed squarely at people publishing it in bulk without adding anything original. You want the speed AI offers. You don’t want a site that smells like a content farm.
Here’s the quick answer. AI content sounds mass-produced when you hand the machine a topic and nothing else. Generic input, generic output. The fix is not a magic prompt, and it’s definitely not a $500 course on “undetectable AI writing.” You feed the AI raw material only you have, and you run an edit pass with your own judgment before anything goes live. That’s the whole trick, and the rest of this post is how to do it.
I’ve also stood on the wrong end of this. I spent years building a review site the way everyone built review sites back then: review the low-quality products, warn people off them, point to what I actually used. It was the standard playbook, and I ran it honestly. Then Google’s 2023 and 2024 updates decided that whole model wasn’t worth ranking anymore, and my traffic went down with the rest of the industry’s. I wasn’t careless, I was doing what worked, right up until it didn’t. That’s why I take this seriously now: when the model itself is thin, careful execution doesn’t save you.
Quick Navigation
Why AI articles all sound the same
Type “write me a 1,000-word article about the best email tools” into ChatGPT and you’ll get something grammatically clean, reasonably organized, and completely interchangeable with what everyone else in your niche gets from the same prompt. That’s not a flaw in the tool. A language model with no other input gives you the statistical average of everything it has read on the topic, and the average has no fingerprints on it.
You can spot the pattern within seconds once you know it. Words no human says across a kitchen table. A lot of “it depends” where an opinion should be, paragraphs that all land at the same length, confident lists with nothing at stake, and not one specific number or lived detail anywhere in the piece. Your reader has seen hundreds of these articles, and they close the tab without being able to tell you exactly why.
The stakes run in both directions. Google’s spam policies now call out scaled content abuse directly, meaning masses of low-effort pages published to game search, whatever tool produced them. And the reader you actually want, the skeptical one with money to protect, has been burned before and leaves at the first whiff of filler. Mass-produced content doesn’t just fail to help you. On a new site, it actively works against you.
Give the machine something only you have
The difference between mush and a usable draft is what you feed in before anything gets written. A topic is not raw material. Raw material is your opinion on the question, the experience that formed that opinion, who you’re talking to, and what you’d tell that person if they asked you directly.
So before you ask for a draft, brain-dump into the chat. Type it or dictate it, and messy is fine. What you think about the topic, why you think it, the mistake you made that taught you, the advice you’d give a friend who asked. Then tell the AI to draft from those notes and nothing else. The draft comes back sounding like a cleaned-up version of you instead of an averaged version of the internet, and that difference survives all the way to the published page.
I use Claude and ChatGPT for this, and the free versions are fine for a first batch. The tool matters less than the input you give it. If you do want to compare the two on writing, I broke it down in which AI actually writes content that doesn’t sound like AI. If this sounds like more work than the “push a button, get a business” pitch you’ve seen, it is, and that’s the point. AI is a tool, not a strategy, and the strategy part stays yours.
A first-batch workflow that holds up
Here’s a concrete way to run batch one without burning a month or your motivation. For a first batch, I would think in terms of five articles, not fifty. You are building your process first, your content library second.
- Collect ten real questions people in your niche actually ask. You probably tripped over plenty of them while picking your niche: forum threads, Reddit posts, Facebook groups, and the questions you typed into a search bar yourself when you were starting.
- Pick five for the first batch. Small on purpose. Five finished, edited articles teach you more than thirty raw ones, and thirty raw ones can hurt a new site more than they help it.
- Write one article at a time. Notes first, then an outline you actually agree with, then the draft. Never ask for five articles in one go. That’s the mass-production button, and everything wrong with AI content comes out of it.
- Edit the next morning with fresh eyes. Editing the same day usually means reading what you meant instead of what’s actually on the page.
Try this prompt
Once your notes are down, this is the drafting prompt I’d start with:
Here are my rough notes on [topic]: [paste your notes]. Write a first draft of a blog post for [describe your reader]. Use only the ideas and opinions in my notes. Keep my opinions as opinions and don’t soften them. Plain conversational English, short paragraphs, no hype words, no exaggerated claims. If a point in my notes needs a fact I didn’t give you, flag it in brackets instead of filling one in.
One caution: do not paste private details, client information, passwords, account numbers, medical details, or anything you would not want stored inside an AI tool. Use real experience, but strip out anything sensitive. You do not need to expose your life to make the article sound human.
That last line matters more than it looks. AI states a guess with the same confidence as a fact, and one fake number or an invented statistic in your first batch is all it takes. That reader is gone and they’re not coming back. Flag it first, then verify before you publish. The whole loop runs on free tools, and if you haven’t set that side up yet, my free AI workflow post walks through the basics.
The edit pass that kills the mass-produced smell
The edit pass is where the articles stop being AI content and start being your content. It takes me 20 to 40 minutes per post, and I don’t skip it even when the draft looks good. Especially when the draft looks good.
- Read your draft out loud, or better, have your computer read it to you. Word has a Read Aloud button under the Review tab, and Edge has one built in, so you can run it right on your post preview. Any sentence you would never say across a table gets rewritten in the words you’d actually use. This one habit catches more AI flavor than everything else combined.
- Add one specific detail per section: a number, a tool name, a price you paid, a thing that happened to you. Specifics are the one ingredient the model cannot supply on your behalf.
- Make sure at least one plain opinion survived. If the draft never actually takes a side, the reader finishes it knowing nothing about you, and you were the reason to read it.
- Verify prices, features, and statistics against the actual source. AI training data goes stale fast, and it will quote last year’s pricing without blinking.
- Cut the filler ending. If the last paragraph could close any article on the internet, delete it and stop one paragraph earlier.
Then run the test that covers all of it: would someone who has been burned by online promises read this and feel like a real person wrote it for them? If yes, publish it and move to the next one.
One expectation to set before you start: your first batch won’t be your best work, and that’s fine. Batch one exists to put something real in front of real people so you have actual signal to improve on. My early posts make me wince a little when I reread them, which mostly proves the later ones got better.
The machine can draft faster than you will ever type. What it can’t do is care whether an article is true, useful, or sounds like you. That judgment is yours, it doesn’t scale, and it’s exactly why a small site run by one careful person can still beat a thousand-article content farm on the only measure that pays long term: whether readers trust it.