Writer and software practitioner Rafal Cymerys has described an emerging personal reaction to low-effort AI-generated text: recognizing familiar patterns and mentally filtering the document before fully absorbing its contents. He calls the experience becoming AI-blind, drawing a comparison with the way web users learn to ignore advertising banners.
Cymerys noticed the pattern while reading workplace documents. Although he appeared to be progressing through the material, he found himself unable to focus on its meaning. The result was repeated discussion with senders about points that their documents had technically already covered. After reviewing several cases, he concluded that each showed strong traces of AI assistance.
His examples include a design document carrying stock phrases and analysis associated with Claude, a 20-page marketing presentation that combined a reasonable strategy with irrelevant technical architecture, and an unusually verbose requirements document for a simple concept. The common problem was not merely a particular vocabulary. It was a structure that treated small details as breakthroughs, retained uncertain internal reasoning and expanded straightforward information without adding meaning.
The essay acknowledges ongoing research and debate over whether people can reliably recognize machine-generated writing. Cymerys disagrees with broad claims that humans are poor at the task, at least for material produced with limited editing. His observation is personal rather than a controlled detection study, and it does not establish that every text he disliked was generated by a model.
The central concern is attention. Exposure to formulaic posts, emails and websites has, in his account, trained him to dismiss similar material quickly. That filter may protect against overload, but it can also hide useful information when substantive details are embedded in a document that otherwise looks synthetic. In a workplace, the cost then reappears as clarification meetings and messages.
Cymerys compares the effect to banner blindness, in which repeated exposure makes a familiar visual category easier to ignore. His version applies that mechanism to tone, sentence flow and inflated framing. The irony, he writes, is that a tool promoted for productivity can slow collaboration when its unedited output imposes more reading and verification on colleagues.
The essay does not argue that all AI-assisted writing is empty or that models should be excluded from documentation. Its examples instead point to an editing problem: generated material that preserves recognizable filler and fails to prioritize the reader's need. AI blindness, as Cymerys defines it, is therefore both an individual habit and a warning that producing more words can reduce rather than improve communication.
Its practical implication is aimed at writers: visible automation residue can become a barrier even when the underlying document contains the answer.
