Engineer reports developing involuntary filtering of low-effort AI-generated work documents

The engineer describes a cognitive shift: documents that exhibit certain stylistic quirks, verbose reasoning, exaggerated language, or misplaced technical detail, are now being dismissed without conscious analysis. This filtering appears involuntary, emerging after prolonged exposure to AI-generated text in professional settings. The effect is strongest with low-effort outputs that retain artifacts of the generation process, such as internal uncertainty or generic marketing flourishes applied to mundane features.

The cost of this adaptation is immediate and practical. When an engineer skips over a design document or requirements spec because it feels AI-generated, the sender must spend additional cycles clarifying points that were already present in the original text. This creates a hidden tax on collaboration, where the efficiency gains promised by AI tools are offset by the need for repeated explanation. The problem compounds when the sender is unaware of the filtering effect, leading to frustration on both sides.

The phenomenon also reveals a limitation in how AI-generated artifacts integrate into engineering workflows. While tools like Claude or other LLMs can produce technically correct text, their output often lacks the conciseness and specificity that engineers rely on for rapid comprehension. When these artifacts are used without human editing, they introduce noise that trained readers learn to ignore. This suggests that AI-generated documents may require a distinct review step to remove stylistic giveaways before they can be effectively consumed.

The comparison to banner blindness is instructive. Just as web users learn to ignore visual clutter, engineers may be developing similar heuristics for textual clutter. However, unlike banners, which are visually distinct, AI-generated text blends into legitimate documents, making the filtering less reliable. The risk is that useful information gets discarded along with the noise, particularly when the AI output is structurally sound but stylistically off. This could lead to a broader erosion of trust in internal documentation if teams begin to assume that all verbose or overstated text is AI-generated.

The engineer’s observation also raises questions about the long-term impact of AI on professional communication. If engineers adapt by filtering out AI-like text, senders may respond by over-editing or avoiding AI tools altogether, reducing their utility. Alternatively, teams may develop new conventions for signaling the provenance of documents, such as explicit labels or standardized templates. Either way, the episode highlights an unintended consequence of AI adoption: the tools designed to save time may instead reshape how engineers process information, with unpredictable effects on productivity.