Everyone wants to know how to improve AI writing. But the irony is that AI doesn’t write badly anymore – if anything, AI writing tools like ChatGPT, Claude and CoPilot write too well. They’ve mastered the mechanics of “good writing,” which means your blog posts, newsletters and LinkedIn updates risk sounding like everyone else’s.
So, if you want your AI writing to stand out, you need to move beyond “good” and aim for effective. That means knowing when to lean on AI, when to take back control, and how to put the guardrails in place so your content doesn’t drown in a sea of same writing.
AI doesn’t write like it used to…
In 2023, soon after ChatGPT-3 launched publicly, I began experimenting with it as a writing tool. Back then – in what now seems like AI’s dawn of time – it had a writing problem.
GPT-3’s sentences were long, confusing, impersonal, earnest and bland as hell.
Fast-forward to today and AI writing looks nothing like it did three years ago. Forget AI slop – LLMs like CoPilot, ChatGPT and Claude now produce prose that actually looks polished, professional and dangerously competent.
That’s because, in 2025, AI writing has improved to the point where the models employ techniques professional writers spend their entire careers mastering. And they do it relentlessly, in every single thing they write.
Unconvinced? Look closely and you’ll see formulas everywhere in its writing – the very same formulas copywriters, journalists and authors rely on in their best work.
AI writing technique 1: The rule of three
AI can’t resist explaining things in threes like “Business success requires strategy, execution, and resilience” or “A healthy lifestyle comes down to diet, exercise, and sleep.”
There’s a reason it writes this way.
The human brain loves patterns – and no pattern is more simple, powerful or recognisable than the rule of three. (Think Goldilocks and the Three Bears, The Three Little Pigs or The Three Musketeers).
Three’s the magic number because it provides enough contrast without overload (too big, too small, just right). It reinforces without being obvious (bricks, not straw or sticks). And it’s very easy to understand.
That’s why stories usually have three acts: beginning, middle and resolution. Why orators and politicians sell ideas with three phrases: “Veni, Vidi, Vici” or “Liberty, Egality, Fraternity”. And why advertisers and businesses give you three choices about what to buy: Small, Medium or Large.
When AI writes this way it’s leaning on one of the writing’s oldest – and most effective – traditions.
AI Writing Technique 2: Juxtaposition
AI loves using juxtaposition: the writing technique that involves placing two things side by side for contrast or comparison. This anchors the reader’s understanding of one concept in the other, even if they actually have nothing in common at all (apples and oranges, anyone?).
Used well, juxtaposition makes writing powerful, punchy and unexpected. Which is why it’s always been a technique favoured by the very best. Think:
Dickens: “It was the best of times; it was the worst of times.”
Orwell: “All animals are equal, but some animals are more equal than others”.
Shakespeare: “Merry and tragical? Tedious and brief?”
Juxtposition is also used a lot in advertising copywriting: “Have a break, have a Kit Kat” or “Maybe she’s born with it. Maybe she’s Maybeline.”
You’ll notice AI writing with juxtaposition when its sentences started to look something like these:
- “The more connected we become, the more isolated we feel”
- “A product is good; a brand is better,” or
- “AI isn’t replacing humans, it’s augmenting them.”
Again, this isn’t necessarily AI writing that needs to be improved, it’s a tried and tested form.
AI Writing Technique 3: The em dash (—)
Finally, there is the most obvious of AI’s writing techniques—the em dash.
AI’s—particularly ChatGPT’s—love for this long punctuation mark has become the subject of countless think pieces, LinkedIn rants and grammar memes.
But guess what? There’s absolutely nothing wrong with the em dash. In fact, it’s a fantastically versatile piece of punctuation.
Lynne Truss, author of grammar-related best seller, Eats, Shoots and Leaves, praised the em dash because “… you can’t use it wrongly—which for a punctuation mark, is an uncommon virtue.”
Writers have long used em dashes to:
- Mark a break in thought or sentence structure—setting a new direction
- Draw attention to particular material—more punch than a comma, less clunky than parenthesis.
- Qualify or expand on a clause—just like this.
It’s this flexibility that has made the em dash the punctuation of choice for some of the most creative and innovative writers ever to put pen to paper.
Emily Dickinson was famous for them:
After great pain, a formal feeling comes —
The Nerves sit ceremonious, like Tombs —
James Joyce crammed Ulysses with them:
—The ways of the Creator are not our ways, Mr Deasy said.
And Jane Austen employed em dashes in Pride and Prejudice for pacing, drama and effect:
They attacked him in various ways,— with bare-faced questions, ingenious suppositions, and distant surmises…”
The only downside to the em dash is that it’s not very common in modern writing—the typewriter had no dedicated em dash key, so most people stopped using it. Today, it lives mainly in literary and academic prose.
That means, when it does start to show up everywhere in your writing, people conclude that AI’s been involved—whether that’s true or not.
Good writing is not the same as effective writing
The problem with AI’s writing in 2025 isn’t that it writes badly because, technically, it doesn’t. Not anymore.
The problem is that AI writes ineffectively – or at least it writes ineffectively when it’s left to its own devices.
Because, while AI now employs the same techniques the greatest writers of all time have used, it does so repetitively, mechanically and constantly.
It hands out em dashes like a real estate agent hands out business cards. It treats the rule of three like a comedian telling every joke exactly the same way – with no variation, no timing and no spark.
And it uses juxtaposition like a see-saw, so its writing takes a predictable pattern: “up and down”, “this vs that” and “once new, now stale”.
The real issue, though, is not that AI writes like this, it’s that most users simply don’t care. They see that the quality of AI’s writing is better than it once was, so they don’t edit its work (a third of companies admit to checking less than 20% of AI outputs). They also skip giving it the guardrails and instructions that would strip out many of its quirks.
Instead, they take a laissez-faire approach – pasting AI text into their emails, posting it on LinkedIn and even dropping it into company reports without question.
And that’s making AI more dangerous than when it wrote badly.
Because, in 2025, everyone’s writing looks good. But it all looks the same.
In the age of AI, true differentiation means sounding like yourself
The good news is that you can still stand out.
In fact, you can even use AI to stand out, so long as you know what you’re doing and you’re prepared to invest just a little more time.
Here are five ways to make your AI-assisted writing cut through.
Tip 1 to improve AI writing: Don’t give it free rein
Left unchecked, AI writes well but not effectively. That’s why you need to rein it in.
Feed it samples of your work. Set boundaries. And be just as clear about what you don’t want it to do, as you are about what you want.
Better still, develop a style guide that locks in your brand voice and use it whenever you write with AI. Also, keep a shared library of tested prompts that you update all the time.
With the right guardrails, you can strip the generic out of AI’s output.
Tip 2 to improve AI writing: Protect your Critical Content
Some writing is too valuable to outsource to a machine. This is the content that shapes your reputation and drives your revenue. It’s your regulated material, your brand-defining pieces, your major proposals and your serious thought leadership.
At Antelope Media, we call this Critical Content, and it requires critical human thought because, not least because if you get it wrong, your neck is on the line.
While AI can help you polish this writing, it shouldn’t be originating it. Fence off this content and keep it human-led.
Read more about how to decide which writing counts as Critical Content
Tip 3 to improve AI writing: Inject lived experience
AI can mimic your style but it can’t fake your life. Even where AI leads your writing, make sure to add details only you can. These could be things like a client story, a hard-earned mistake, a personal observation or a local reference.
It’s touches like these that show you know what you’re talking about and anchor your writing in reality. And that’s a place AI models don’t live.
Tip 4 to improve AI writing: Edit for rhythm and surprise
AI writes in patterns because it’s trained on patterns. That’s why its sentences fall into the same cadences and rhythms over and over again.
Your job is to break them.
Vary its pace. Kill its cliches. Flip its structures.
Use editing to inject unpredictability so your writing still feels alive.
Tip 5 to improve AI writing: Take a stance
AI is built to be neutral. It’s polite, balanced and so middle of the road you can almost see the double white lines.
You’re not. You have ideas, opinions and experience worth sharing.
So back your perspectives, take risks and let your reality – not AI’s pattern recognition – shine through.
After all, people don’t follow sameness; they follow original thinking. And they follow it even more often when it’s expressed in an original way.
The real danger isn’t AI writing badly anymore
Forget bad AI writing. That was 2023’s problem. In 2025, the risk is that AI writes too well… so well, in fact, that everyone starts sounding exactly the same.
And in a crowded market – where content is everywhere – sameness kills.
Today, your edge doesn’t come from whether you use AI, it comes from how you use it.
Because good writing may be easier than ever to produce. Effective writing is still just as hard as it’s always been.
Ready to fix AI sameness?
AI can make your writing faster, but only you can make it stand out. We’ll show you how.
FAQ: How to improve AI writing
1) Why does AI writing make everything sound the same?
AI models are pattern recognition systems trained on massive text datasets. When generating text, they predict likely word sequences based on statistical patterns from their training data. Without specific guidance, they tend to default to the most common, "safe" formulations they've learned — which often results in generic-sounding output.
Factors that contribute to sameness:
- Training data bias: Models learn from text that already contains common patterns and conventions.
- Prompt similarity: Many users ask for similar content in similar ways.
- Risk aversion: Models are designed to avoid controversial or unusual outputs (though there are ways around this).
- Lack of specificity: Vague prompts tend to produce generic responses.
The solution is guardrails and better prompting:
- Be specific about style, tone and format.
- Provide examples of what you want (and don’t want).
- Give context about audience and purpose.
- Experiment with creative constraints or unusual perspectives.
- Iterate and refine based on initial outputs.
AI writing can be distinctive and engaging when given clear direction. The key is understanding that these systems excel at following detailed instructions rather than mind-reading what you really want.
2) How can I improve AI writing without losing our brand voice?
The more specific you get, the better AI’s output gets. Here’s a systematic approach to improving AI’s writing so you don’t sound like everyone else:
Before writing:
- Provide a detailed style guide with tone, voice and formatting preferences.
- Share 3–5 strong examples of your brand voice in action.
- Specify what not to do (e.g. competitors’ styles, overused phrases, tones to avoid).
- Give context about your audience and brand personality.
During generation:
- Use iterative prompting — start broad, then refine based on initial output.
- Ask for multiple variations to find the best starting point.
- Request specific adjustments (“make this more conversational” or “add more technical detail”).
After generation:
- Edit for cadence, clichés and structure.
- Add lived examples, insider knowledge and unique perspectives only you can provide.
- Verify facts and claims the AI generated.
- Test the final content against your brand voice guidelines.
The key is treating AI as a collaborative writing partner rather than a replacement. It handles structure and initial drafts, while you add the authenticity and brand-specific nuances.
3) What should stay human-led when using AI for content?
Always human-led:
- Regulated content: Legal documents, compliance materials.
- Brand-defining pieces: Mission statements, core messaging, foundational content.
- Thought leadership: Original insights, industry predictions, controversial takes.
- Major proposals: High-stakes pitches, strategic documents, client presentations.
- Crisis communications: Sensitive messaging requiring careful judgment.
- Personal storytelling: Authentic experiences, company history, founder narratives.
Human oversight essential:
- Fact-sensitive content: Statistics, research claims, technical specifications.
- Cultural/contextual content: Humour, cultural references, sensitive topics.
- Relationship-dependent messaging: Personalised outreach, customer service escalations.
Learn more in our Critical Content framework.
4) What prompts help AI write more effectively?
Essential prompt components:
- Role assignment: “Act as a [senior editor/conference chair/target reader]”.
- Specific constraints: Word counts, grade-level readability, format requirements.
- Success metrics: Define what success looks like (engagement, clarity, conversions).
- Output format: Be explicit — tables, markdown, numbered lists.
Advanced techniques:
- Scoring systems: “Rate each idea 1–10 on relevance, novelty, value”.
- Multiple variants: “Give me 3 versions: concise, vivid, conversational”.
- Structured feedback: “Flag any phrase that’s flabby, cliché or passive with 3 alternatives”.
- ‘Avoid’ lists: “No passive voice, no sentences over 25 words, avoid these overused phrases: [list]”.
See our free ChatGPT prompt templates for practical examples.
5) Is the em dash bad writing?
No — the em dash is a perfectly legitimate punctuation mark. The problem is AI models overuse it, creating a telltale “AI voice.”
Why AI loves em dashes:
- They’re versatile (can replace commas, colons and parentheses).
- They create dramatic pauses AI thinks add “personality.”
- They help AI connect ideas when it’s unsure about transitions.
When to use them well:
- For genuine interruption or dramatic pause: “The results were clear — we had failed.”
- To set off explanatory information when commas would be confusing.
- For abrupt topic shifts in informal writing.
Red flags of overuse:
- More than one per paragraph.
- Using them instead of proper transitions.
- Creating artificial “drama” where none exists.
The real issue isn’t punctuation — it’s lazy sentence construction. Edit specifically for em dashes to keep your prose sharp.
6) Can AI help teams write faster and better?
Yes, but only with the right framework. Most teams get faster immediately but sacrifice quality if they skip the human elements that matter most.
The speed part is easy:
- First drafts in minutes instead of hours.
- Multiple variations generated instantly.
- Research synthesis and outlining automated.
- Formatting and basic editing streamlined.
The “better” part requires discipline:
- Strategic thinking must stay human — AI can’t set your positioning or make judgment calls.
- Voice and brand consistency require active management (style guides, example banks, calibration).
- Fact-checking is critical — AI still hallucinates statistics and claims.
- Editing isn’t optional — AI drafts are starting points, not finished work.
Successful teams use the Critical Content framework, build prompt libraries, and invest in AI writing training for their teams.