AI writing tools are useful. The problem is that the default output has a texture readers have learned to spot, and once they spot it they trust the page less. The good news is that the giveaways are concrete. Here are nine of them, with a fix and a quick before-and-after for each. (This post uses no em dashes anywhere, because that is sign number one.)

1. Em dashes everywhere

Language models love the em dash. They use it to splice clauses together at a rate no human writer naturally hits. Two or three per paragraph is a strong tell.

The fix: replace most of them with a full stop, a comma, or a pair of parentheses. Read the sentence aloud; the pause usually wants to be shorter than a dash implies.

Before (dashes splicing four clauses): "The strategy worked, mostly, because the team moved fast, faster than anyone expected." After (broken into real sentences): "The strategy worked, mostly. The team moved faster than anyone expected."

2. "It's not just X, it's Y"

This construction shows up constantly in generated text because it sounds profound while saying little. Its cousins are "it's not about X, it's about Y" and "this isn't merely X."

The fix: state the actual point directly. If Y is what matters, lead with Y.

Before: "It's not just a tool, it's a complete rethinking of how teams work." After: "The tool changes one thing: teams stop emailing files and edit the same document."

3. Rule of three on every line

Three-item lists feel balanced, so models reach for them relentlessly. "Fast, simple, and reliable." "Plan, build, and ship." When every sentence has a trio, the rhythm becomes hypnotic and obviously machined.

The fix: keep a list of three only when all three items genuinely matter. Otherwise cut to two, or to one strong word.

Before: "Our platform is powerful, flexible, and intuitive, helping you create, manage, and publish content faster, easier, and smarter." After: "The platform helps you publish faster. It is also flexible enough that you will not outgrow it in a year."

4. Hedging on everything

Generated text often refuses to commit. "This can sometimes potentially help in certain cases." "It may be worth considering." The hedging piles up until no actual claim survives.

The fix: decide what you believe and write it. If a caveat is real, give it once, specifically, then move on.

Before: "Generally speaking, it could be argued that shorter emails may tend to perform somewhat better in many situations." After: "Shorter emails get more replies. In our last 40 sends, anything under 120 words beat the longer versions on reply rate."

5. Generic openers

A huge share of AI articles begin the same way: a sweeping statement about how important the topic is, or a clause about the modern world. Readers skip these on sight.

The fix: open with a specific claim, a number, a question with a real answer, or a small scene. Make the first sentence carry information.

Before: "In the modern business environment, content marketing has become more important than ever before." After: "We published 30 blog posts last year. Four of them drove 80 percent of the signups."

6. Every paragraph the same length

Models tend to produce paragraphs of nearly identical size, usually three to four sentences each. The page becomes a wall of even grey blocks, and the uniformity reads as mechanical.

The fix: vary it deliberately. Let some paragraphs run long. Let others be one line.

Like this.

Before: four paragraphs, each exactly three sentences, each making one point at the same pace. After: a long paragraph that develops an idea fully, followed by a short punch that lands it.

7. Empty intensifier adjectives

"Robust solutions." "Powerful capabilities." "Seamless experiences." "Cutting-edge technology." These words add emphasis without adding meaning, and they cluster thickly in generated copy.

The fix: cut the adjective and check whether the sentence still works. Usually it does. If you need to convey quality, show it with a fact instead.

Before: "Our robust, powerful platform delivers a seamless, cutting-edge experience." After: "The platform handles 50,000 articles per account and has not dropped a publish in 18 months."

8. "In conclusion" and other signposting

Generated text announces its own structure. "In conclusion." "In summary." "Firstly, secondly, finally." "It's worth noting that." Human writers rarely label their closing paragraph; they just close.

The fix: delete the signpost word and let the content do the signalling. A final paragraph reads as final because of where it sits, not because you announced it.

Before: "In conclusion, it is worth noting that consistency is the most important factor." After: "Consistency beats intensity here. Publish twelve decent posts a year, not three great ones and then silence."

9. No specifics at all

This is the deepest tell, and it is what makes all the others worse. AI drafts describe things in the abstract. They say "businesses" instead of naming one, "studies show" without a study, "significant improvement" without a number.

The fix: add proper nouns, real figures, dates, named tools, and first-hand observations. Specificity is the single strongest signal that a human with actual experience wrote the piece.

Before: "Many companies have seen great results by improving their content." After: "Buffer rebuilt its blog around original data in 2023 and grew organic traffic by about 60 percent over the next year, according to their own reporting."

A two-pass edit that catches most of this

You will not fix all nine while drafting. Do it in a second pass. Read the draft once just for em dashes and the "it's not just X" pattern, since those are mechanical to find. Read it a second time asking one question per paragraph: what is the concrete claim here, and is there a real example backing it? If the answer is nothing, the paragraph is filler.

Tools can help with the mechanical layer. Austen drafts in your brand voice and runs the kind of style checks above before you ever see the text, so you spend your editing time on the specifics rather than hunting em dashes. It is free to try with five articles.

Run a draft through these nine signs once and you will start catching them as you write, which is the real goal. The aim is not to hide that you used AI. It is to publish something specific enough and human enough that nobody thinks to ask.

9 Signs Your Content Sounds Like AI Wrote It (and How to Fix Them)