How to fix your AI-generated content

AI-written content has a tell. Several, actually.

Once you know what to look for, you can't unsee them. The em dashes everywhere. The "not just X, but Y" constructions. The lists of three. The "here's the thing" opener. The colons in headings. The same handful of words that nobody used much before 2023 and now appear in every other blog post.

None of these are dealbreakers on their own. Every human writer uses some of them sometimes. The problem is the accumulation. When all of them show up in one article, the reader's pattern-matching brain clocks it before they've consciously registered why. The content reads as off. Trust drops a notch. They skim faster, or close the tab.

This article is about how to fix that. Specifically, the linguistic patterns that give AI content away, why AI defaults to them, and what to do instead. By the end, you'll be able to read your own AI-assisted drafts and spot the tells before anyone else does.

A quick framing note. This isn't about avoiding AI detection tools. Those tools are mostly unreliable, and "passing" one doesn't mean your content is good. This is about the less talked about, more important problem. Your audience can spot AI content even when the detectors can't. And once they do, the content stops working.

The patterns that give AI content away

These are the tells, ordered roughly from most obvious to most subtle.

1. The em dash, everywhere

The single biggest signal. AI loves the em dash. It uses it to add asides, attach clauses, create dramatic pauses, and connect half-thoughts. A 1,200-word AI draft will routinely contain eight or ten of them, at a minimum.

Most professional human writers use em dashes occasionally. AI uses them constantly. The fix is simple, find every em dash and replace it with a full stop, a comma, or rewrite the sentence so it doesn't need one. You'll be amazed how many you find.

2. "Not just X, but Y" (and its cousins)

This is the construction AI reaches for whenever it wants to sound insightful. "It's not just about content, it's about strategy." "This isn't a tool, it's a philosophy." "Not just a workflow, but a way of thinking."

It feels like a profound contrast but it's almost always a hedge. The writer is upgrading a basic claim by pretending the obvious version was wrong. Real writing makes the upgraded claim directly. "Strategy matters more than content volume." Done. No setup required.

3. "Here's the thing" / "Here's what most people miss"

This one makes me laugh cos my husband says this out loud in conversations now. Conversational openers that pretend to share an insider truth. AI uses them as a transition device. They appear at the start of paragraphs to create a false sense of intimacy with the reader.

Once you notice them, they're everywhere. The fix is to delete them. The sentence underneath almost always stands up on its own.

4. The rule of three

AI defaults to lists of three. Three benefits. Three reasons. Three steps. Three adjectives in a row ("clear, compelling, and conversion-focused").

This isn't because three is genuinely the right number every time. It's because three is the most common rhythm in business writing and AI has learned it as a default. The fix is to break the pattern. If your list naturally has two items, leave it at two. If it has four or six, use that. If three is genuinely right, vary the sentence structure so it doesn't sound chanted.

5. Punchy fragment stacking

Short. Punchy. Sentences. Like this. For emphasis.

Single-word paragraphs. Used for impact.

It can work occasionally. Honestly I use this one a lot, purely because of how I've always written. AI uses it as a default rhythm, which kills the effect. Real punchy writing earns its short sentences by surrounding them with longer ones. The contrast is the whole point.

6. Colons in headings

"Content Strategy: The Hidden Power of a Clear Message." "AI Writing: What Most People Get Wrong." "Building Trust: A Framework for B2B Brands."

This isn't something that screams AI on its own. My own headlines have colons in sometimes. But the colon-headline format is everywhere in AI-generated content. It mimics academic and corporate writing conventions and the AI has been trained on a lot of both. Headlines that read naturally don't necessarily need the colon. "What most people get wrong about AI writing." "How to build trust as a B2B brand." Faster, less corporate, more clickable.

7. The question opener

"Ever wondered why your content isn't converting?" "Have you ever tried to write a blog post and felt completely stuck?" "What if there was a better way?"

These openers feel engaging but they're hollow. They ask a question the reader didn't come to your page to answer. They're a stalling tactic before the writer commits to a real position.

If the article is genuinely answering a question, ask the question in the headline and start the article with the answer. Don't recycle the question at the top of the body.

8. The specific vocabulary

Some words have become so heavily associated with AI writing that using them now reads as a tell, even when used correctly by a human.

The current list (subject to change as AI training catches up):

• Delve
• Leverage
• Bottleneck
• Compounds
• Momentum
• Manufacture (used metaphorically, as in "manufacture clarity")
• Robust
• Seamless
• Empowers
• Tapestry
• Realm
• Quiet

If you find these in your draft, replace them. "Delve into" becomes "look at." "Leverage" becomes "use." "Bottleneck" becomes "stuck point" or just gets rewritten around. Robust becomes strong, or sturdy, or gets cut.

9. The empty intensifier

"Simply put." "Literally." "Quite frankly." "Honestly." "Of course."

AI sprinkles these to mimic natural speech. Human writers use them sparingly, if at all. Strip them out and the sentence is almost always clearer.

10. The "in today's landscape" opener

Any version of: "In today's fast-paced digital landscape..." "In an increasingly competitive market..." "In a world where attention spans are shrinking..."

These are the OG AI-tells... AI-generated scene-setting that adds nothing. The reader knows what year it is. They know markets are competitive, so just get to the point.

11. The hedge

"This can sometimes be a key factor in driving improved engagement outcomes."

AI is trained to be helpful and not overstate. The result is sentences padded with hedges. "Can sometimes." "Often." "May potentially." "In many cases."

Human experts write more confidently. "This drives engagement." If you're not confident enough to make the claim directly, the claim probably shouldn't be in the article.

12. The false balance

"While AI is a powerful tool, it's equally important to remember that human creativity also plays a crucial role."

This is the AI equivalent of standing in the middle of the room with arms outstretched, refusing to commit to either side. Real writing picks a position. The whole point of an opinion piece is the opinion. If the article doesn't have one, it doesn't need to exist.

Why AI defaults to these patterns

AI doesn't write badly because it's broken. It writes this way because it's trained on a vast corpus of business writing, marketing blogs, LinkedIn posts, content farm articles, and SEO-optimised filler. That writing is the most common form of professional content on the internet. It's also the form most readers have learned to skim past.

When AI generates content, it's essentially producing the average of everything it's read. The em dashes, the rule of three, the "not X but Y" constructions, the colon headings, all of these are statistically common patterns in the training data. The AI isn't trying to sound generic. It's trying to sound like content. And it's succeeding, which is the problem.

The fix isn't to abandon AI. It's to recognise that AI gives you the average, and your job as the editor is to push the content away from average and toward something specific.

The actual fix (a process that works)

Once you can name the patterns, fixing them gets faster. Here's the process I use on every AI-assisted draft.

1. Read it aloud

Most AI tells are audible. The rhythm of punchy fragments. The repetition of three-item lists. The pause an em dash creates. The hollow ring of "here's the thing." If you read the draft aloud and something sounds chanted, mechanical, or oddly cadenced, that's the pattern.

2. Strip the obvious tells

Find every em dash. Replace with full stop or comma. Search for "not just X" and rewrite as direct claims. Cut any sentence that starts with "here's the thing," "here's what most people miss," or "the truth is." Cut "delve," "leverage," "bottleneck," and "robust" wherever they appear.

This step alone removes most of the AI feel. It also reveals which sentences had nothing underneath the construction. Those need rewriting from scratch.

3. Break the patterns

If you've got three lists of three in a row, change one to two and one to four. If every paragraph is the same length, vary them. If you've used a colon in two consecutive headings, rewrite one. Pattern-break wherever the rhythm has become predictable.

4. Add the things AI couldn't know

A specific number. An anecdote from a client (anonymised or not). A piece of evidence from your industry. A reference to something that happened last month. An opinion that's actually yours, with the reasoning behind it.

This is the single biggest lift you can give an AI draft. AI can write around a topic, but it can't write from inside your experience. Adding even one paragraph of genuine specificity turns generic content into something that could only have come from you.

5. Read it aloud again

This time listen for whether the voice holds. Does it sound like you, or does it sound like a content marketing article that someone like you might have written? Those aren't the same thing. The first is recognisable, the second is forgettable.

When AI is genuinely useful (and when it isn't)

I use AI every day. I'm not anti-AI, but I'm specific about where it earns its place in my workflow and where it doesn't.

AI is genuinely useful for brainstorming, expanding outlines, summarising research, rewriting clunky paragraphs, finding the word I'm groping for, generating examples to react to, and turning voice notes into a rough first pass. It saves real time on tasks that don't require a point of view.

AI is not useful for anything that requires a point of view, anything where specifics matter, anything that needs to sound like a specific person, or anything where the reader is supposed to trust the writer's expertise. AI can read 300 blog posts in an hour. It can't tell you which ones matter, or why, or what the people who wrote them got wrong.

The mistake most businesses make is using AI for the second category and assuming the output is good enough because it reads as competent. It reads as competent. It doesn't read as anyone in particular. That's the difference between content that fills space and content that builds trust.

What to do if you're publishing AI content at volume

Some businesses will keep using AI for volume reasons, particularly for SEO content where the publishing cadence matters more than any individual article landing.

If that's you, the minimum you can do is the four-step process above. Read aloud. Strip the tells. Break the patterns. Add specifics. It takes ten minutes per article. The lift in quality is significant, and the lift in how the content reads to humans is bigger than any AI detection score would tell you.

If you're not willing to do those four things on every piece of AI-assisted content you publish, you might be doing more brand damage than the content is worth. There's a real cost to publishing content that signals "we don't care enough to make this sound like us." Once an audience reads three of those in a row from your business, they stop reading the fourth.

The bigger principle

AI is a useful tool. It's also a tool that produces the average. If you want content that builds trust, you have to push past the average, toward specifics, toward a clear point of view, toward writing that could only have come from you.

The patterns in this article are the most common ways AI content fails that test. Strip them out, add what AI can't, and you'll be ahead of most of what's on the internet right now. Which, depressingly, is a lower bar than it used to be.

If your AI content has been creating more problems than it's solving, the Blog Audit is the fastest way to figure out which pieces are salvageable and which should come down. Or book a discovery call if you want to talk through what a better approach to AI-assisted content looks like for your business.

And if you want to read the companion piece to this one, your audience can spot AI-written content covers what's actually at stake when readers clock that content wasn't written by a person.