{"slug": "the-slush-pile-just-became-infinite-and-nobody-knows-how-to", "title": "The slush pile just became infinite and nobody knows how to", "summary": "The volume of unsolicited book queries has surged from about 200 per month to 200 per week since ChatGPT went mainstream, with most submissions now AI-generated or heavily AI-assisted, according to industry insiders. Publishers are automating first-pass screening and self-publishing platforms like Amazon's KDP added roughly two million titles last year, with an estimated 60-70% involving significant AI, while detection tools struggle with false positives and the arms race favors generators. The middle of the market is being hollowed out first, and the first copyright test case on 'substantial curation' will set precedent for a decade.", "body_md": "# The slush pile just became infinite and nobody knows how to\n\nThe volume spike started eighteen months ago. Before [ChatGPT](/en/tags/chatgpt/) went mainstream, a typical imprint might see two hundred unsolicited queries a month. Now they're getting that many in a week. Most are AI-generated or heavily AI-assisted. The tell isn't bad prose anymore — it's the absence of idiosyncrasy. Every character speaks in the same measured rhythm. Every conflict resolves through reasonable compromise. No weird obsessions, no awkward phrasing that somehow works, no authorial fingerprint whatsoever.\n\nAgents are adapting faster than publishers. Two I know have started requiring a \"process statement\" with every query: draft timeline, revision history, tools used. One agency now runs submissions through a detection pipeline before they even hit a human reader. False positives are a nightmare — they've accidentally flagged three genuinely written debuts this quarter because the authors happened to be meticulous outliners.\n\nThe economic pressure is brutal. A slush reader costs roughly forty thousand a year including benefits. An API call to a classification model costs fractions of a cent. Do the math. Houses are automating the first pass whether they admit it publicly or not. The detection arms race favors the generators, not the detectors. Every time a new watermarking scheme drops, someone publishes a workaround on Hugging Face within days.\n\nSelf-publishing platforms are the real canary. Amazon's KDP store added something like two million titles last year. Industry estimates suggest sixty to seventy percent had significant AI involvement. The platform's response? A checkbox at upload: \"Does this contain AI-generated content?\" Honor system. No verification. The scammers don't check the box. Legitimate authors using AI as a brainstorming partner do — and get buried in the algorithm anyway.\n\nReaders are noticing. Goodreads reviews increasingly mention \"AI feel\" as a complaint. But here's the uncomfortable part: in blind A/B tests I've seen, readers can't reliably distinguish AI-assisted from human-only above chance when the human author is midlist genre fare. The distinction only appears at the literary fiction and voice-driven memoir ends of the spectrum. Which means the middle of the market — cozy mysteries, romance series, techno-thrillers — is getting hollowed out first.\n\nSome editors are quietly experimenting with \"AI-native\" imprints. Not labeled as such. They commission detailed outlines from humans, generate drafts with tuned models, then hire line editors to inject voice. One house has three such titles slated for fall. Early reader copies tested well. Nobody involved wants to go on record.\n\nThe copyright office's stance — no protection for purely AI-generated text — creates a fascinating loophole. If a human provides a detailed enough prompt sequence and does substantial curation, the resulting work might be copyrightable. But \"substantial curation\" has no legal definition yet. The first test case will set precedent for a decade.\n\nWhat nobody discusses openly: the backlist problem. Houses own millions of out-of-print titles. Training data. They're sitting on gold mines they can't legally exploit without author permission — and most contracts from before 2020 don't address AI rights. The litigation will be spectacular.\n\nFor now, the slush pile keeps growing. The detection tools keep losing. And somewhere in a suburban home office, someone is prompting \"write a bestselling thriller about a retired detective with a dark secret\" for the ten thousandth time this month. The resulting manuscript will be competent, forgettable, and indistinguishable from the nine thousand nine hundred ninety-nine before it.\n\nThe industry survives on the exceptions. It always has. But finding them just got exponentially harder.\n\n[Next WHRG'26 kicks off Saturday — humanoid field finally gets its →](/en/news/6926/)\n\n[these AI tool field notes](https://tanyan888.com/), with plenty of directly applicable cases.", "url": "https://wpnews.pro/news/the-slush-pile-just-became-infinite-and-nobody-knows-how-to", "canonical_source": "https://promptcube3.com/en/news/6928/", "published_at": "2026-08-19 13:47:32+00:00", "updated_at": "2026-08-19 14:13:44.980289+00:00", "lang": "en", "topics": ["generative-ai", "ai-ethics", "ai-policy"], "entities": ["ChatGPT", "Amazon KDP", "Hugging Face", "Goodreads"], "alternates": {"html": "https://wpnews.pro/news/the-slush-pile-just-became-infinite-and-nobody-knows-how-to", "markdown": "https://wpnews.pro/news/the-slush-pile-just-became-infinite-and-nobody-knows-how-to.md", "text": "https://wpnews.pro/news/the-slush-pile-just-became-infinite-and-nobody-knows-how-to.txt", "jsonld": "https://wpnews.pro/news/the-slush-pile-just-became-infinite-and-nobody-knows-how-to.jsonld"}}