How stylistic features impact user engagement in AI-generated content

There is a clear divide in opinions about what AI-generated content looks like. The web community often identifies certain “tells” — such as the use of em dashes — as indicators of automated content. However, it is important to understand that the effectiveness of content does not always depend on specific stylistic features. The real question is how these characteristics affect user engagement.

Purpose of the study and methodology

To determine which stylistic features of content impact user engagement, an analysis was conducted of over 1,000 URLs. The study included content written entirely by humans, content created with the help of AI, and content produced collaboratively by both humans and AI, across various industries such as tech, e-commerce, healthcare, education, and analytics.

Key stylistic features studied included:

• Use of “not only… but also” constructions.
• Sentence starters with “then,” “this,” or “that.”
• Introductory phrases such as “In this article,” “Let’s explore,” “We’ll look at.”
• Use of “In conclusion” within the text.
• Frequent use of em dashes.

Identifying the problem: trends in AI content

It was found that many of the identified constructions, commonly associated with AI-generated content, did not have a significant impact on content performance. At the same time, some characteristics did have a notable negative or positive impact on user engagement

Analysis of results: The impact of “AI Tics” on engagement

Based on the analysis, the following key results were identified:

• “Not only… but also” constructions: While this construction can serve a purpose occasionally, its frequent use showed a negative correlation with engagement. The more often these constructions were used, the higher the bounce rate. Constant repetition of this phrase could tire readers and push them away from continued interaction.
• Use of “In conclusion”: Frequent use of “In conclusion” to begin sections had the strongest negative correlation with engagement rate. This suggests that such introductions might create the impression of content coming to an end, which deters readers from further engagement.
• Em dashes: The use of em dashes surprisingly had a slight positive correlation with engagement. This indicates that em dashes can be useful for creating more expressive, nuanced sentences that engage readers better. However, it is important to use them in moderation to avoid giving the impression of artificiality.

AI tics: What they are and how they affect content

AI Tics is a term used to describe certain stylistic features or “tells” that are characteristic of content created using artificial intelligence (AI). These could be repeated constructions, phrases, or other elements of language that are commonly seen in AI-generated texts.

Consequences for content marketers and recommendations

The analysis revealed that certain AI tics, such as “not only… but also” and starting with “In conclusion,” have the potential to turn off users, especially when overused. However, it was found that em dashes are not necessarily harmful to content and may even enhance engagement when used appropriately.

Content marketers should pay attention to stylistic features that could affect how content is perceived. It is advisable to avoid overusing certain phrases and constructions that might make content seem automated, while using em dashes appropriately to improve the effectiveness of the text.

Prospects for content development in 2026

Considering changes in search algorithms and requirements, content in 2026 will focus on deeper interpretation of topics and a more interactive approach to audience engagement. Businesses that rely solely on “quick wins” from search engines will be less competitive, so it is important to focus on creating content that provides real value to users.

3 Practical takeaways for content teams

Here’s what content marketers can act on today:

1. Don’t Over-Optimize for AI Detection
   Google does not apply a monotonous punishment score for “AI style.” The phrases we analyzed did not show a correlation with user engagement.
   Therefore, it’s not recommended to rewrite content simply because someone labeled a phrase as “AI-written.” The priority should be creating content for users, focusing on its usefulness, relevance, and clarity.
2. Be Mindful of How You Wrap Up Content
   Explicit conclusion blocks aren’t inherently bad, but overly formulaic and generic structures are likely turning readers away.
   Consider integrating conclusions into analysis, using subtler transitions, or adding new value with headers, instead of signposting obvious structure.
3. Use Punctuation that Makes Sense
   If your style requires em dashes, use them. In this dataset, they were actually associated with better reader engagement.
   Don’t be afraid to use punctuation elements that aid clarity, structure, and communication with the audience.

Don’t miss the forest for fake plastic trees

AI is likely here to stay in content workflows. However, the problems with “bad” AI writing are not limited to linguistic tics and punctuation.
While everyone has their stylistic opinions, we should be cautious about turning stylistic preferences into editorial laws. Write valuable content, think about your readers first, and don’t panic every time someone on Twitter or LinkedIn decrees that “X phrase = AI.”

Read this article in Ukrainian.