The hard part of AI content isn't ranking, it's sounding human

Getting AI to write content that ranks well and gets cited reliably by other AI is the easy part. I’ll show you how.

The hard part, the part I’ve been wrestling with since ChatGPT first landed, is getting that content to sound like something you’d actually have written. Something on brand. Something you’re happy to put your name to.

And here’s the catch nobody tells you: those two goals pull against each other. The writing that machines love is the writing that makes humans wince. This post is about how we learned to have it both ways.

First, the easy part: writing to rank and get cited

I’ve written about ranking and getting cited by AI in more depth elsewhere, so this is the short version.

Your titles and headings need to closely match what people, and AI, are actually searching for. Not clever, not the official industry term, just the phrase a real person would type.

Then, straight after each heading, answer the question that heading poses. Clearly and immediately. Don’t make anyone hunt for it.

And straight after the answer, back it up. Supporting facts, data points, a relevant statistic, a quote, or a table. This is also where you answer the related questions branching off the main one, the ones AI quietly generates when it fans out from a search.

There’s more nuance to it, but that’s the engine. Match the search, answer it fast, prove it with specifics.

The problem: that writing is dry as a bone

Tell AI to write this way and you’ll notice something. It comes out flat. Academic. No warmth, no personality, nothing of you in it.

If your only goal is to rank and influence what other AI says about you, that’s fine. Machines happen to love dry, factual writing. They eat it up.

It’s only a problem if you expect actual humans to read the thing and you care what they make of your brand. Which, in my experience, is almost every business.

So the challenge is keeping the structure the machines reward while adding back the character the humans need. Here’s the ladder we climbed to get there.

Step one: strip out the obvious AI tells

The instinct most people have is to tell the model “add more warmth and personality.” Don’t. What comes back is cringe. It does add personality, just a deeply unlikeable one, and one that screams AI from the first line.

Telling it “don’t sound like AI” doesn’t work either. You can’t ask something to simply stop being itself. You have to name the specific habits.

So spell them out. Tell it to avoid:

Contrastive framing. The “it’s not X, it’s Y” construction it reaches for constantly.
Empty intensifiers and hedging. Words like “absolutely,” “actually,” and “truly” that add nothing.
Forcing every single line to land. Not every sentence needs to be a mic drop.
The dash habit. No em dashes.

Give it enough of that and it stops sounding like a machine. But it still won’t sound like you. It’ll just sound like a more neutral machine.

Step two: don’t just hand it your old articles

The next thing everyone tries is feeding it samples. “Here are some things we’ve written, sound more like this.”

It seems sensible. It mostly backfires.

The AI doesn’t know what you liked about those pieces. Is it the subject matter? Particular words? The rhythm of the sentences? So it guesses, usually wrong, and latches onto surface features. The result is repetitive: the same handful of phrases lifted from your samples, over and over, until it reads like a parody of itself.

Step three: turn your style into curated guidelines

Here’s what actually works.

In a separate chat, give the model a good stack of things you’ve genuinely written. Then ask it to study them and write a long list of guidelines it can extrapolate about your style. How you open, how long your sentences run, the words you favour, the ones you never use.

Now the important bit: curate that list by hand. Go through every guideline and keep only the ones you actually agree with and want it to follow. Bin the rest.

Then, when it’s time to write real content, give it only the curated guidelines. No writing samples, no example phrases to parrot. Just the rules. This makes a big difference, and it sidesteps the copycat problem from step two entirely.

Step four: feed every correction back in

You’re close now, but it still won’t be quite you. The last step is what closes the gap, and it’s a habit rather than a one-off.

Every time the AI writes something, give it detailed feedback. Not “make it better,” but exactly what was off and why.

And when you end up rewriting a passage yourself, hand your version back and ask it to work out what new guideline it should have followed to get there itself. Add that to the list.

Do this consistently and the list grows. For some brands I’ve now got well over a hundred guidelines. The frontier models handle lists that long without breaking a sweat, and the writing gets closer to your voice with every round.

The rule that ties it all together

Here’s the trap, and it’s the one thing you have to watch.

As you pile up guidelines chasing your brand voice, it’s easy to quietly undo the SEO structure you started with. A voice guideline that says “vary your openings” can wreck the answer-first structure AI needs. A rule about being conversational can bury the direct answer three paragraphs down where no machine will find it.

So every guideline gets one test before it goes on the list: does this undermine the ranking and citation structure? Titles still match the search. Answers still come straight after the heading. Facts still follow the answer. If a voice rule breaks any of that, it doesn’t make the cut, however nice it sounds.

Get that balance right and you stop choosing between the two. You get content the machines rank and cite, and content a human reads and thinks, “yes, that sounds like them.” Cake, eaten.

Want content that ranks, gets cited, and actually sounds like your business? Get your free audit, or email hello@inkon.co.uk. I read every message myself.