When you already know the feeling before even understanding why … so you automatically scroll past without deciding. This is what we like to call the ‘AI Skip Reflex,’ referring to the instant and subconscious recognition that a certain media was manufactured instead of made, causing a reflexive scroll.

However, while most would fix this problem by ‘just detecting the AI parts,’ most do not understand why viewers are actually skipping instead. Moreover, survey findings from the Journal of Artificial Intelligence Research found that the accuracy of humans detecting AI was often up to chance, as most often assumed for human-written text (Fraser et al., 2025). Meanwhile, Milicka et al.’s (2025) study on learning to detect AI found that language proficiency and prior exposure did not reliably help.

Therefore, the AI Skip Reflex is not a scanned analysis done on media content, but rather the result of a question that most users have encountered even before AI: “Is this worth my attention?” So, readers are not rejecting AI as it was made by a machine, but are rather suffering from a lack of cognitive relevance, too high a predictability, and absence of tone produced by it.

The Anatomy of the Skip Reflex

So, if no one can fully identify AI content, then what is actually triggering a scroll? These are visual features that the brain often does not find appealing such as:

• Visual symmetry - consisting of lines with similar length and rhythms
• Lower lexical complexity - where a linguistic comparison of AI generated medical essays by Doru et al. (2025) found that AI tends to write in present tense, lower lexical complexity, and flatter register.

Catering towards reading ease tends to be the pattern, as eye-tracking research has shown that people do not read web content, but rather scan it. In the Proceedings of the Human Factors and Ergonomics Society Annual Meeting, researchers found that readers tend to move through text in an F-shape pattern consisting of: a horizontal scan at the top, a sweep further down an a vertical scan again at the bottom (Shrestha et al., 2007). Therefore, the opening lines of a paragraph and distinctive elements within paragraphs become more important.

The Psychology of Sameness

Andy Clark’s (2018) research on predictive processing serves as the basis for the cognitive neuroscience model that the brain consistently forecasts what comes next, while flagging when that prediction does not match reality. Additionally, a study in Frontiers in Psychology on expectations and memory found that the brain prioritizes encoding events that violate these expectations as unpredictability within inputs becomes the source for new information (Kafkas et al., 2025). Consequently, text that does not violate expectations and predictions does not give the brain information worth encoding, causing attention to shift elsewhere, further evident in Clark’s (2018) analysis of predictive processing, showing that participants will voluntarily engage in cognitively challenging material as it leads to higher intrinsic rewards.

Breaking the Reflex: How to Craft for Human Engagement

• Incorporating friction and opinions - As mentioned, the brain rewards violated expectations, so swapping neutral summaries for challengeable claims will lead to higher engagement.
• Deliberately disrupt patterns - This can be achieved through varying sentence lengths, as uniform lengths and rhythms cause attention to be shifted elsewhere.
• AI as a pre-draft: While the use of AI within the workplace is inevitable, the output produced should be treated as a source for structure and scaffolding, as the product still lacks credibility, voice, and specificity.

The Future of AI Marketing

As everyone now has access to the same AI models producing similar phrasing and habits, the use of AI stops becoming a source of value and becomes a free and universal tool. However, what is now scarce has always been so, as having a specific voice and defensible claim tends to generate more interactions, in comparison to an output based on an average of the public’s opinions. Therefore, as humans cannot reliably detect AI, the objective switches from ‘hiding the use of AI’ towards crafting a product that is strikingly unique from the standard baseline to incentivize discussion and engagement.

References

Clark, A. (2018). A nice surprise? Predictive processing and the active pursuit of novelty. Phenomenology and the Cognitive Sciences, 17(3), 521–534. https://doi.org/10.1007/s11097-017-9525-z

Doru, B., Maier, C., Busse, J. S., Lücke, T., Schönhoff, J., Enax-Krumova, E., Hessler, S., Berger, M., & Tokic, M. (2025). Detecting artificial intelligence–generated versus human-written medical student essays: Semirandomized controlled study. JMIR Medical Education, 11, e62779. https://doi.org/10.2196/62779

Fraser, K. C., Dawkins, H., & Kiritchenko, S. (2025). Detecting detectability with current methods. Journal of Artificial Intelligence Research, 82, 2233-2278. ****http://doi.org/10.1613/jair.1.16665

Kafkas, A., Westerman, M., Cleto, K., Sabaityte, K., & Sergi, K. (2025). When personality meets surprise: Individual differences in memory for unexpected events. Frontiers in Psychology, 16, Article 1652428. https://doi.org/10.3389/fpsyg.2025.1652428

Milička, J., Marklová, A., Drobil, O., & Pospíšilová, E. (2025). Humans can learn to detect AI-generated texts, or at least learn when they can't (arXiv:2505.01877) [Preprint]. arXiv. https://doi.org/10.48550/arXiv.2505.01877

Shrestha, S., Lenz, K., Chaparro, B., & Owens, J. (2007). "F" pattern scanning of text and images in web pages. Proceedings of the Human Factors and Ergonomics Society Annual Meeting, 51(18), 1200–1204. https://doi.org/10.1177/154193120705101831

‍