You open an article, and by the second paragraph you can tell nobody really wrote it. You couldn't point to the sentence that gave it away.
That's the AI tell. It's never one sentence. It's a handful of small signals, and once enough of them pile up, it's unmistakable.
The uncomfortable part for anyone using AI in their content: your audience has developed the same sense, and they don't stay to finish.
Here are eight signals that content is AI-generic, and the fix for each.
1. The throat-clearing opener
AI loves to open by announcing that the topic is important. "In today's fast-paced world..." "As businesses navigate an increasingly complex landscape..." "Content marketing has never been more important."
These openers claim nothing. They say the topic matters without saying why, to whom, or on what evidence.
The fix: open with something specific: a thing that happened, a number, a question, or a plain statement of what the piece delivers. If you tested 40 hooks on one serum and a single variable decided the winners, open with that. "In today's fast-paced world" is a stall.
2. Passive voice where someone should be acting
AI drifts into the passive whenever an action should belong to someone. "Decisions were made." "Results were achieved." "A new approach was implemented."
The passive hides who did what. Human writing, especially a founder's, names the actor, because the actor is the point. "We got pricing wrong in month two" lands very differently from "a decision was made regarding pricing".
The fix: find every "was [done]" in the draft and give it a subject who did the thing.
3. No numbers
"Many clients saw improvement." "Significant results were achieved." "We grew substantially."
Vague size is a strong AI signal. A model with no real data reaches for adjectives. A person who lived through the result knows the number.
The fix: replace every unmeasured claim with a real figure, or cut the claim. "We've delivered more than 1,500 ads for Adkrity and Xyxx" says far more than "we've made a lot of ads for brands", and a reader can check it. If you don't have the number, you don't have the claim.
4. No named people or things
AI writes about categories, not instances. "Brands are using automation to scale." "Founders are building in public." "Tools like this help teams move faster."
People name things: this brand, this tool, the conversation with this client on a Tuesday. Named specifics are close to impossible for AI to fake, because they need real experience behind them.
The fix: for every category noun, ask whether you can swap in a specific instance. Usually you can, and it makes the piece twice as credible and much harder to copy.
5. Every paragraph the same shape
Read enough AI-generated articles and you notice every section has the same shape: topic sentence, two or three supporting sentences, a neat close. No one-line paragraphs for emphasis. No long, winding sentence followed by a short one.
Human writing varies, because thoughts don't all weigh the same.
The fix: read the draft aloud. Wherever it ticks like a metronome, break the rhythm. Add a one-sentence paragraph after a dense section. Let an idea run longer when it's earned it. Cut a paragraph in half if it only repeats the one before.
6. Lessons without the story behind them
"It's important to test before launching." "Communication is key." "Start with your audience, not your product."
These are conclusions without evidence. AI produces them because they're common in writing about any topic. But a lesson with no origin story has no authority, and the reader has no reason to trust it over the thousand other articles saying the same.
The fix: for every piece of advice, add the situation that taught it. "Test before launching" is a platitude. If you learned it because an untested error branch let a client's Shopify sync fail mid-campaign without anyone noticing, say that. A lesson that cost something is worth reading.
7. Bold that emphasises nothing
AI often bolds phrases in the middle of paragraphs that aren't key terms or important claims, just words that seemed weighty to the model at the time. It doesn't guide the eye. It breaks the text up.
The fix: remove any bold inside a paragraph that isn't a defined term or a label. If a sentence matters that much, give it a heading. Otherwise, let the sentence carry itself.
8. Ending where it should begin
AI-generated pieces usually end by summarising what they just said, then offering vague encouragement. "As we've seen, this matters. By applying these principles, you can reach your goals."
That ending tells the reader the piece has no next step to offer.
The fix: end with something to do or something unresolved. A specific thing to try. A question without a clean answer. A prediction the reader can check in six months.
The one-question test
After all of the above, read the piece and ask: could someone without my experience, my data and my point of view have written this?
If the answer is yes, it isn't yours yet.
The goal isn't content that sounds human. It's content that could only have come from you, and the gap between those two is what makes it worth reading. AI can draft, structure and vary. It can't remember the Tuesday the client called, the pipeline failed and you worked out why. That memory is the content. The AI just helps you write it down faster. It's the same standard we hold our AI UGC ads to.
