AI Is Destroying Trust Between Writers and Readers: The Atlantic Essay and Hacker News Debate
The Atlantic published Rachel Kadish's essay "AI Is Destroying the Trust Between Writers and Readers" on November 12, 2026. The essay's subtitle states the thesis directly: as literary scams proliferate, authors no longer know whether their readers are real, and readers no longer know whether authors are. The Hacker News discussion that followed (285 points, 177 comments) treated the piece less as a narrow publishing story than as a case study in how generated text destabilizes the basic reciprocity between a writer and a reader.
The triggering anecdote: a novelist was confronted over a plot she never wrote
Kadish opens with a book-signing scene from late 2025. A middle-aged couple approached her table. The husband had Parkinson's disease and wanted to discuss her novel The Weight of Ink. He believed the novel featured a character with Parkinson's who chooses assisted suicide rather than endure decline. Kadish had not written that. The character, Helen Watt, does not end her own life in the novel. Kadish suspected the misunderstanding came from an earlier ChatGPT session in which the model gave a false rendition of Helen's plotline. The encounter turned an abstract concern about model hallucination into a real-world confrontation: a reader had formed a moral objection to a book based on a plot that did not exist. Kadish asks: "How many other people had been similarly misled?"
The essay's frame: literary scams, not only hallucinations
The Atlantic's subtitle frames the problem as literary scams as well as misinformation. The scraped text of the essay available for this analysis contains the opening anecdote and the publication metadata but not the full body of the article. What can be stated from the source is that Kadish connects AI misrepresentation of an author's work to a wider loss of confidence in the literary marketplace. The Hacker News comments fill in the marketplace half of the argument with concrete claims about Amazon's self-publishing pipeline.
Hacker News: AI slop as a denial-of-service attack on discovery
Commenters described a supply shock. @sarreph noted that Amazon lets individuals upload a maximum of six books a week: "That's a book a day (give or take), per account. No human author is publishing a book per day." @danielodievich relayed a friend's experience self-publishing a science-fiction trilogy over years. The friend said the gap between his first book six years earlier and his third book had become filled with self-published slop that was denial-of-servicing everyone. He reported people publishing 50 books a year and summarized the scam economics: "5 dollars in compute to make a garbage book to publish on amazon to produce 30 bucks of revenue from unsuspecting half a dozen people is worth the squeeze so scammers/sloppers do it."
@internet_points quoted Mary Rasenberger, CEO of the Authors Guild: "Amazon spends a lot of time and effort trying to prevent these [fake] books from getting posted, but it's always a game of catch-up." The commenter then challenged the premise: "Do they really? They take a cut; they have incentives to let people get scammed." That tension, between platform enforcement and platform revenue, runs through the thread.
@GeoAtreides said the fanfiction and webserial scene is flooded: "Everything written after 2024 REEKS of AI, and most of them, over 90% are obvious AI." The commenter cited stylistic tells and meta tells such as unrealistic output volume and Patreon accounts for fanfiction. @beacon294 argued that nonfiction had already been damaged by a "fraudulent blog post to nonfiction 'bestseller' pipeline" before generative AI arrived.
The structural problem: writers can now produce faster than readers can read
@auntienomen made the sharpest structural point: scams are not the worst problem. LLMs let writers generate prose faster than readers can read it, which breaks an implicit bargain. However long it takes a reader to consume a piece of writing, it almost certainly took the author longer to write it. Even bad prose carried a signal: the author thought the work was worth real time. The commenter and colleagues adopted a guideline: "you should spend more time writing something than you expect the audience to spend reading it. Anything less is disrespectful." @glitchc compressed the same idea: "When reading takes more effort than writing, trust is inevitably a victim."
Counterpoint: the trust was already broken
Several commenters resisted the framing that AI destroyed a previously healthy trust. @bluecalm argued that generic trust in the written word was always a mistake: "Trusting the written word just because it's written by someone or published by some newspaper was a huge mistake before AI was in the picture." The commenter described discovering errors in newspaper coverage of their own expertise and extrapolating to coverage they could not judge. AI, in this view, exposed vulnerabilities rather than created them. @ericmay said many publications "were never deserving of trust anyway" and that AI is formalizing a distrust earned through biased writing and ad-filled pages: "now that the tide is going out we can see who has been swimming naked." @CGMthrowaway noted that novels, op-eds, memoirs, and songs have long used uncredited ghostwriters, that journalism can be written by intelligence operatives or PR professionals, and that legislation is written by lobbyists. Greater vigilance about provenance is not, in this view, a bad thing. @photochemsyn extended the critique to media ownership: AI makes it easier to trace who owns a publication and whether its articles favor the owners' economic interests.
What would rebuild trust?
Commenters proposed several mechanisms, none sufficient alone. @barbariangrunge framed the problem as a "reverse turing test: can a human convince another human (or bot screener) that they are human?" and called for legislation banning AI from imitating or impersonating humans and for labeling AI-generated art. @unknownian called generative AI that imitates human art an "anti-invention" and argued it should be regulated "like digital drugs." @juanre built a site, Possibly Made by a Human, to test whether a piece of writing took a human-compatible amount of time and carries a human writing signature. The site's premise is that faking those signals is possible but costly: it takes at least as long as human writing would. @j45 argued that people should keep reading books and writing with their own pen and cursor to preserve language skills. @DeepOntology proposed a reframing: the issue is not that AI destroys trust but that it forces a reconfiguration of how trust is established. The commenter said disruption driven by external capital injection is pernicious, while disruption of asymmetric power relationships can be welcome.
The cultural mood: satire, low-background steel, and the human signature
@dsign wrote a satirical 2110 scenario: Writers' Street in Alexandria, Egypt, where novelists type at street cafes using self-invented codes in chromatic scales; Benedictines on Montecristo rent mechanical typewriters by the hour; famous writers jump from cliffs rather than bear accusations of using AI; AI literature is free and abundant for the masses, while the discerning prefer manually bound tomes inked by human brains. @FeteCommuniste offered a one-line cultural periodization: "Pre-2023 writing is the low-background steel of culture." @chistev invoked Roald Dahl's "The Great Automatic Grammatizator," a story about a machine that produces sellable fiction. The references point to the same anxiety: once text is cheap, the scarce good becomes evidence of human expenditure, attention, and accountability.
Bottom line
The Atlantic essay's opening anecdote is small and specific: one reader was wrong about one plot point after a chatbot misrepresented it. The Hacker News thread shows why that small error scales into a systemic problem. When platforms allow industrial-scale uploads, when scam economics favor volume over quality, and when generation outpaces consumption, readers lose the ability to use effort and provenance as trust signals. The counterpoint from the thread is equally important: trust in written authority was already unevenly earned and often exploited. AI does not create that history, but it makes the old verification problem unavoidable.
Sources
Related
- Dispatch
- Dispatch
- Dispatch
- Dispatch
- Dispatch