The debate over AI generated content often centers on style “tells” like em dashes, repetitive phrases, or formulaic intros. But are these stylistic quirks really affecting how readers interact with content? A recent study analyzing 1,000+ content marketing URLs provides data driven insights into which AI writing habits impact engagement, and which are mostly harmless.
Understanding the Study
Content marketers frequently debate what makes AI writing feel “off.” Phrases like:
- “In today’s fast paced digital landscape…”
- “It’s important to note that…”
- “Not only… but also.”
- “In conclusion…”
…often get flagged as AI “tells.” While they may trigger a sense of repetition or formula, their actual effect on user engagement hasn’t been clear. To investigate, the study examined content from:
- 10 domains across industries like tech, healthcare, ecommerce, education, and analytics
- 1,000+ URLs, including human written, AI assisted, and fully AI generated posts
The team normalized AI writing tics per 1,000 words to account for article length differences and excluded pages under 500 words. Engagement was measured using GA4 metrics, where a session lasting 10 seconds or more counted as engaged.
Key Insights: Which AI Tics Matter
1. “Not only… but also” and similar structures
This construction had a negative correlation with engagement, particularly when overused. While it can emphasize points, repetitive usage appears to annoy readers. One blog post in the dataset used this phrase 12 times, correlating with higher bounce rates. For AI assisted writing, limiting this structure is advisable to avoid disengagement.
2. Section headers starting with “Conclusion”
Headers like “Conclusion” showed the strongest negative impact on engagement (≈ -0.118). Readers may skip to the end quickly, and these formulaic endings often signal lower quality content. Blending conclusions into the analysis or using subtler transition headers can improve retention.
3. Em dashes: Surprisingly harmless
Contrary to popular opinion, em dashes showed a slight positive correlation with engagement. Rather than hurting readability, they often appear in nuanced, explanatory sentences, helping to keep readers invested. The takeaway: don’t avoid em dashes purely for fear of being “AI like.”
Other Observations
Many other supposed AI tics, filler intros like “In this article” or repeated sentence starters such as “Then” or “This shows” had negligible effect on engagement. Over optimizing for AI detection or obsessively removing minor stylistic elements is largely unnecessary.
Practical Takeaways for Content Teams
- Write for readers, not AI detection: Avoid changing content just because it contains minor AI like phrasing. Focus on clarity, value, and readability.
- Reconsider formulaic conclusions: Integrate conclusions naturally within the content or add additional insights instead of relying on repetitive “Conclusion” headers.
- Use punctuation appropriately: Em dashes and other stylistic tools can enhance readability and engagement when used thoughtfully.
The Bigger Picture
AI is increasingly part of content workflows, but the perceived “bad writing” often comes from stylistic bias rather than measurable engagement issues. While certain repetitive structures may irritate readers, most AI tics are harmless. The real focus should remain on creating useful, readable, and well structured content, rather than policing every em dash or filler phrase.
In short, the study shows that reader engagement is about substance, not style alone. Smart use of AI tools can enhance content without sacrificing performance, as long as writers focus on audience needs over arbitrary stylistic rules.
