Readers bail on AI-written text — where AI actually helps writing

Colin Breck argues that a flood of AI-generated design docs, pull requests, tickets, and even personal messages has become unreadable, because the output carries no context or perspective. A common anti-pattern: build something with AI, then have AI retrospectively summarize it into a ‘design document’ that reads as exhaustive machine detail rather than a proposal meant to build consensus. PR summaries describe what changed line by line but never answer why the work matters, how risky it is, or where a reviewer’s judgment is actually needed. The core problem is an asymmetry of context — the person who wrote the prompts can skim the output and instantly separate signal from noise, but the recipient has none of that background and is forced to read every line as if peering into a machine’s internals.

Breck points to survey data reinforcing the backlash: readers who suspect a post is AI-authored overwhelmingly stop reading (78%) and avoid the author afterward (71%), while 98% prefer a human’s own flawed, idiosyncratic voice over a polished AI rewrite. His conclusion is not anti-AI but anti-substitution. Used as a tool inside the writing process, AI was invaluable to him — verifying technical claims against source code and logs, generating citations, catching subtle notation errors four expert reviewers missed, fixing grammar, and drafting diagrams. What never worked was reversing the flow and asking AI to write prose from context; the results were consistently inaccurate and unpleasant to read.

The one exception proves the rule: AI wrote his academic paper’s abstract perfectly and he used it unchanged — because an abstract is the most mechanical, abstracted part of a document, and summarization is exactly what these models excel at. The takeaway for technical teams is that writing meant to persuade, align, or connect with other humans still has to come from a human voice; AI belongs in the verification and mechanical layers of that work, not as a replacement for authorship.