AI content is everywhere. How do you actually make yours not suck?

[Dev]

Everyone's pumping out ChatGPT content like it's a slot machine. And SEO blogs, landing pages, cold emails - it's all starting to feel like AI wrote it for other AIs to read. No voice, no flavor, no reason to scroll past the first paragraph.

But some of you are doing better. Way better.

I run a reputation management and SEO company, and here's our approach to making AI content sound like it didn't crawl out of a robot’s recycling bin:

- Start with a well-ranking, deeply engaging article from the niche, something that clearly hits both humans and search.
- Then feed that into GPT as a reference, asking it to match the tone, structure, and depth - not regurgitate, but follow the vibe.
- The output is usually 90% human-feeling. Not perfect, but solid.
- Then read through it and clean up the obvious tells - em dashes, stale GPT phrases like 'in today’s fast-paced world, and anything that smells like it was written by AI.

That combo gives us scalable content that ranks, reads well, and doesn't scream AI.

So, how do you make your AI content not suck? What's your process?

 

[roydan]

I made my own GPT that's based on my own content. I can't get rid of the -- and all the other things, but other than that, I add a paragraph or two on manually written unique takes on any subject I write about.

 

[polecat]

I would recommend this this to help probably my best appsumo purchase https://app.frase.io/

 

[Octavia]

I made my own GPT that's based on my own content. I can't get rid of the -- and all the other things, but other than that, I add a paragraph or two on manually written unique takes on any subject I write about.

What LLM did you use?

 

[roydan]

What LLM did you use?

Gave 4.5 a test, it was fine but didn't notice much of a difference than 4o.

 

[t2van]

- Start with a well-ranking, deeply engaging article from the niche, something that clearly hits both humans and search.
- Then feed that into GPT as a reference, asking it to match the tone, structure, and depth - not regurgitate, but follow the vibe.
- The output is usually 90% human-feeling. Not perfect, but solid.
- Then read through it and clean up the obvious tells - em dashes, stale GPT phrases like 'in today’s fast-paced world, and anything that smells like it was written by AI.

You are doing all this wrong from the start.


Feed it 6-8 (I found) of articles you have actually written. To teach it your writing style, tone and language.

Matching other sites is a waste of time. Ask it to write how you do.

Then adjust the output as you feel fit to fit a little more into your style. Like for example adding Conclusion: or Summary or even Emoji is not something I would do.

Ask it to include table data half way through the article.

Now your 90% of the way to stop getting it from sucking.

I would say all of our AI articles rank. I dont want to give away the formula because its working! When and if that fails I'll share it.

What I will say the more info you can give it to your actual writing style 1500+ words or more the better it becomes are sounding human for niche.

 

[SirTop]

Has anyone tried using Deep Research to write an article...

It’s actually very nice once you add all the instructions of what you want and don’t.

It always delivers clean content.

 

[TheVigilante]

Has anyone tried using Deep Research to write an article...

It’s actually very nice once you add all the instructions of what you want and don’t.

It always delivers clean content.

Is this some other tool that I don't know lol

But seriously this AI thing is next level I have tested Claude and can attest to it that it spits better output than ChatGPT

 

[t2van]

Has anyone tried using Deep Research to write an article...

It’s actually very nice once you add all the instructions of what you want and don’t.

It always delivers clean content.

Do you mean using to research or write?

We use Deep Research to research our niche, history and specific areas, then load these in as project resource files for it to refer back to when it comes to creating content.

Although it's not ideal way to go about using that to be honest. I did a write up on here on using deep research for marketing

 

[kimse]

I’m actually nodding along with a lot of what’s been said here.

I’ve found that the biggest difference-maker is feeding it your own writing. Not just “good articles” from around the web, but stuff you’ve written, emails, docs, old blog posts, whatever. Once it gets that flavor, the content starts sounding like you again.

Also, I never just paste the final AI output. I always rewrite intros and conclusions myself. Even small changes, tweaking sentence length, adding some slang or dry humor, go a long way in killing that generic tone.

And yeah, in today’s fast-paced world… immediate delete.

Haven’t tried Deep Research yet but now I’m curious. I appreciate all the insights in here; solid stuff.

 

[polecat]

Has anyone tried using Deep Research to write an article...

It’s actually very nice once you add all the instructions of what you want and don’t.

It always delivers clean content.

Yes a few times found good.

I noticed as well deepseek would often produce sort of college level work from each article

 

[Zwielicht]

I made my own GPT that's based on my own content. I can't get rid of the -- and all the other things, but other than that, I add a paragraph or two on manually written unique takes on any subject I write about.

I do the same thing.

I generally just write my own content still, but whenever I've used ChatGPT to write content, I feed it similar articles I've written and ask it to write in the same tone and style.

From there, I make sure to specify things I know about my own writing style. So things like British English spelling and grammar conventions and using Oxford commas, as an example.

 

[Dev]

You are doing all this wrong from the start.


Feed it 6-8 (I found) of articles you have actually written. To teach it your writing style, tone and language.

Matching other sites is a waste of time. Ask it to write how you do.

Then adjust the output as you feel fit to fit a little more into your style. Like for example adding Conclusion: or Summary or even Emoji is not something I would do.

Ask it to include table data half way through the article.

Now your 90% of the way to stop getting it from sucking.

I would say all of our AI articles rank. I dont want to give away the formula because its working! When and if that fails I'll share it.

What I will say the more info you can give it to your actual writing style 1500+ words or more the better it becomes are sounding human for niche.

Fair points, and I actually agree with quite a bit of that.

Training GPT on your own writing absolutely levels up the output - no doubt. Feeding it longform content you've written to capture your tone, cadence, and quirks? That's a smarter baseline than just mimicking what's already ranking. No argument there.

That said, I don't agree matching top-ranking content is a waste. For us, it's less about copying and more about reverse-engineering structure, flow, and search intent - especially for clients in competitive niches where you're not just writing, you're positioning.

We're just coming at the same goal from two angles - yours is more "brand voice first," ours is "search and reader behavior first," and both clearly have their strengths.

Appreciate you sharing a glimpse of your process though. Solid stuff.

 

[t2van]

Fair points, and I actually agree with quite a bit of that.

Training GPT on your own writing absolutely levels up the output - no doubt. Feeding it longform content you've written to capture your tone, cadence, and quirks? That's a smarter baseline than just mimicking what's already ranking. No argument there.

That said, I don't agree matching top-ranking content is a waste. For us, it's less about copying and more about reverse-engineering structure, flow, and search intent - especially for clients in competitive niches where you're not just writing, you're positioning.

We're just coming at the same goal from two angles - yours is more "brand voice first," ours is "search and reader behavior first," and both clearly have their strengths.

Appreciate you sharing a glimpse of your process though. Solid stuff.

Ah if you want to analyse/reverse engineer it. Have you tried:


Take the article and get GPT to scan for common entities and the writing structure to use that as a baseline for its "ranking ability" then have it craft an article for you.

Once it's done that article ask it to redo it using the common entities and writing structure found in the top ranking pages and have a play with it that way.

 

[SirTop]

Do you mean using to research or write?

We use Deep Research to research our niche, history and specific areas, then load these in as project resource files for it to refer back to when it comes to creating content.

Although it's not ideal way to go about using that to be honest. I did a write up on here on using deep research for marketing

I mean for actually creating content for an idea you have... It helps a lot to cut out those AI jargons and write very clean articles.

Could you please share your post on deep research for marketing?

If I can improve the use of deep research

 

[t2van]

I mean for actually creating content for an idea you have... It helps a lot to cut out those AI jargons and write very clean articles.

Could you please share your post on deep research for marketing?

If I can improve the use of deep research

Try here:

Using GPT my way. - Another prompt guide

Since I have introduced AI into our daily work flow getting better at prompting and setting up the whole interface for what we need to do has been key. Me personally right or wrongly I talk to AI like it's a member of my team, although a tool I have found the more information you give it the...

officeoutlaw.com

If you need more specifics let me know

 

[Skipper]

I’m lazy and just put out whatever ChatGPT gives me. No tweaking the prompt or anything.

Best part is when people start debating whether my content is AI or human. Whatever gets the views is good enough for me!

 

[chaktty]

It's about a robust prompt that explains in details how you want the content to be created, then you read through, edit and give the human feel, and write like a researcher.