AI Writing Tics That Kill Engagement: A Study
▼ Summary
– Online debates about “AI-written” content often confuse subjective taste with objective performance, focusing on stylistic “tells” rather than their actual impact on readers.
– A study analyzed content marketing pages to identify which common AI writing patterns, like “not only…but also” or “In conclusion,” actually correlate with lower user engagement rates.
– The analysis found that overusing “not only…but also” constructions and using headers starting with “Conclusion” had the strongest negative correlations with engagement.
– Contrary to popular belief, the frequent use of em dashes showed a slight positive correlation with engagement, challenging the notion that they are a harmful AI artifact.
– The practical takeaways are to prioritize reader usefulness over avoiding perceived AI “tics,” be mindful of formulaic conclusions, and use punctuation like em dashes when stylistically appropriate.
The online conversation about AI-generated content is filled with strong opinions, often focusing on specific phrases and punctuation marks as supposed “tells.” While debates about stylistic preferences are subjective, a crucial question for content marketers remains: which of these common patterns actually cause readers to disengage? By analyzing a large dataset of content marketing pages, we can move beyond speculation to identify the specific linguistic habits that correlate with lower user engagement, separating impactful trends from mere noise.
Our investigation began by compiling a list of frequently cited AI writing patterns. These included repetitive sentence structures like “not only… but also,” generic introductory phrases such as “in this article,” obvious conclusion headers, and the infamous em dash. We then built a dataset from over a thousand content marketing URLs across various industries, ensuring a mix of human-written, AI-generated, and collaboratively produced posts. To enable a fair comparison, we standardized the data by calculating the occurrence of each “tic” per 1,000 words and focused on engagement rate as the key performance metric, which measures whether a visitor stays on a page for more than ten seconds.
Initially, we considered simply tracking the total number of these tics. However, this approach proved misleading. For instance, the em dash appeared so frequently that it skewed the averages. A deeper look revealed why: these patterns aren’t exclusive to AI. When we analyzed known human-written works, including a modern novel and Shakespeare’s “Hamlet”, they also scored highly on these same metrics. This confirmed that a high count of certain phrases doesn’t inherently mean content is AI-generated or of poor quality; it often just reflects common English prose. We therefore shifted to analyzing each stylistic element individually to see its specific impact.
The analysis yielded clear, actionable insights. While most of the so-called AI “tells” showed no statistically significant correlation with performance, a few notable exceptions emerged. Constructions built around “not only… but also” demonstrated a negative correlation with engagement rate. Their overuse, frequently seen in AI-assisted drafts, appears to trigger user bounce. For example, one reviewed blog post used this phrasing twelve separate times.
The strongest negative signal in the entire study was the use of section headers that explicitly begin with the word “Conclusion.” Posts featuring these headers showed the largest negative correlation with engagement. This suggests readers may quickly scroll past such formulaic endings or that content relying on these obvious structural signposts tends to be lower in quality overall.
In a surprising twist, the data challenged a major piece of conventional wisdom. Despite their reputation as an AI artifact, em dashes showed a slight positive correlation with engagement. This indicates they are not harming performance and may even align with more thoughtful, explanatory writing that retains reader interest. Their use should be guided by style and clarity, not fear of detection.
For content teams, the study offers three practical takeaways. First, avoid over-optimizing content based on perceived AI detection. Most phrases we examined had no bearing on engagement, so rewriting solely to eliminate them is counterproductive. Prioritize reader usefulness and clarity above all else. Second, be strategic with conclusions. Instead of relying on generic “Conclusion” headers, consider blending final thoughts into the analysis or using subtler transitions that add new value. Finally, use punctuation that serves the writing. Em dashes, according to this data, are not the enemy and can be employed effectively.
The integration of AI into content workflows is a reality. However, the focus should remain on creating valuable material for readers. Chasing after every stylistic hot take declared on social media is a distraction. The goal is to produce resonant, helpful content, not to eliminate every pattern that someone arbitrarily labels as robotic. Write for people first, and let the data, not the dogma, guide your stylistic choices.
(Source: Search Engine Land)
