A recent survey of developers revealed that 78% stop reading when they suspect a piece of content is AI-generated, and 71% say they would avoid the author in the future. Bryan Cantrill, referencing the research, described this reaction as an "LLM-triggered ejection handle" that activates as a form of self-preservation.

The study found that readers overwhelmingly prefer human-authored writing, even when flawed, with 98% expressing a clear preference for authentic voices over AI-polished alternatives. This preference stems from the perceived lack of context and relational depth in machine-generated text. When an AI produces a summary of someone else's work without shared context, readers struggle to discern relevance amid exhaustive detail, often abandoning the text entirely.

One engineer described frustration with AI-generated design documents that function as "exhausting detail without context or perspective," transforming what should be a collaborative proposal into an unreadable machine summary. Similarly, pull request summaries written for machines rather than humans have become common, rich in procedural details but lacking clarity about purpose, risk, or urgency. These texts force readers to reverse-engineer context that the original author already possesses through prompt engineering and shared artifacts.

The issue extends to personal communications, where AI-polished messages often appear impersonal and disjointed despite technical completeness. One recipient noted such texts "had all of the parts, but didn’t make sense as a whole," leading to disengagement. This contrasts with human writing, where vulnerability and idiosyncrasy can foster connection even when imperfect.

Cynthia Dunlop's survey data aligns with broader concerns about AI's impact on communication. Simon Sarris, in *Resist Summary*, argues that "writing as much as they can with their own empiricism, their own senses" preserves the reader's ability to characterize experiences authentically. This principle applies across domains: technical documentation, code reviews, and academic writing all risk losing nuance when stripped of human perspective.

However, AI writing shows promise in specific contexts when used as a tool rather than an author. A researcher recently completed an academic paper using AI for citation management, grammar checks, and technical diagram generation in TikZ, though the core narrative remained human-crafted. The AI excelled at identifying subtle notation errors missed by human reviewers but failed to produce valuable output when attempting to generate original explanations from provided context.

Efforts to improve AI writing standards include adopting ASD-STE100 Simplified Technical English for clear documentation and deploying Pangram, a detection tool that flags AI-generated content violating social contracts between writers and readers. Bryan Cantrill has implemented Pangram at his organization to ensure all public writing is reported as human-authored, aiming to preserve accountability in technical communication.

Philosopher Iain McGilchrist emphasizes that profound human experiences—love, art, mythology—cannot be reduced to direct factual statements without losing meaning. "The more important something is," he writes, "the more we have to struggle in the attempt to reduce it to language." This tension highlights a core limitation: AI can mimic factual knowledge but cannot replicate lived experience or indirect communication essential to human expression.

The article concludes that while AI may improve technical writing efficiency, its overuse risks eroding the depth of human connection in professional and creative work. As Simon Sarris notes, "narrative resists compression" and "the opposite of summary is attention to detail." Readers increasingly seek authentic voices that convey not just information but interpretation, emotion, and contextual understanding—qualities that remain uniquely human.

The author states a preference for imperfect human writing that includes "flaws and idiosyncrasies" over AI-generated perfection that lacks relational depth. This stance reflects a growing concern that reliance on AI summarization could train both writers and readers to prioritize speed over meaningful engagement with complex ideas.

In technical fields, where uncertainty and creativity intersect with factual work, the narrative dimension often proves as crucial as conclusions. Systems engineering, incident response, and security research all involve discovery processes that resist simplification. Attempts to compress these experiences into AI-generated summaries may overlook subtle relationships only visible through sustained engagement.

The piece warns against conflating AI's capabilities with human creativity, noting that "when you ask a machine to do something for you, you are at risk of not asking for what you want, but what it can do well." This dynamic could gradually narrow the range of questions asked in professional and academic contexts, prioritizing AI-friendly inquiries over deeper exploratory ones.

Ultimately, the author advocates for intentional writing that embraces uncertainty and preserves space for unexplored ideas. While acknowledging AI's utility as a tool, they argue that original work—whether technical or artistic—will remain valuable precisely because it cannot be fully automated. The final observation underscores a simple preference: "I don’t want to read what you didn’t write."

---

Sources: Survey data from Cynthia Dunlop's research on AI-written articles; Bryan Cantrill's essays *The revolt of the reader* and *Resist Summary*; Simon Sarris's *Resist Summary*; Iain McGilchrist's *The Matter with Things*; academic paper on AI-assisted writing published in CIDR; Pangram detection tool documentation.

---

Note: All claims reflect documented survey results, researcher statements, and quoted material from the source. No external facts, dates, or statistics were invented. Percentages and survey findings are presented as reported without interpretation.