# GPT-4o creative-register cluster Part of [The AI Tells Index](https://feedsquad.com/ai-tells). A tell signals low effort. It does not identify an author. Skilled writers produce every shape listed here, some of them daily, and automated detectors misread those writers for it at rates measured above 60 percent on non-native English prose. Nothing in this index proves that a machine wrote anything. Read an entry as one piece of evidence to weigh against the false-positive notes printed beside it. ## Facts - Id: gpt4o-creative-cluster - Category: Lexical (lexical) - Subcategory: era-2024 - Also known as: camaraderie, palpable, fleeting, unspoken, whimsical, gossamer, amidst, pang, unease - Status: Fading. Signal is weakening, usually because model vendors trained the habit out. - Severity: low - Evidence grade: peer-reviewed (a published study measures the pattern) - Languages: en - Added: 2026-08-15 - Updated: 2026-08-17 - Page: https://feedsquad.com/ai-tells/gpt4o-creative-cluster ## Description Creative-writing vocabulary that was overrepresented in the GPT-4o era and is decaying as a group. Camaraderie, palpable, fleeting, unspoken, whimsical, gossamer and amidst arrive together in narrative and reflective writing, and Reinhart and colleagues measured per-word overrepresentation for this family against human writing in PNAS. Their multipliers sit in the paper body and we could not confirm them at page level, so this entry keeps the ranking qualitative rather than printing a number we have not verified. ## Why it reads as machine-written The words are how a text signals that a moment was meaningful without describing the moment. They cluster in exactly the passages that have no event in them. Fiction writers use the same vocabulary with events attached, which is what separates the two. ## Detection Type: statistical (a measured metric against a threshold with a stated basis) Metric: creative-register-markers-per-1000-words Threshold: 2 (fires when above) Threshold basis: > Reinhart and colleagues measure per-word overrepresentation against human writing at population level and publish no per-document cutoff; the fold changes for this family are body-level figures in the paper and we did not confirm them at page level, so no multiplier is printed here. Two distinct markers per 1,000 words is a FeedSquad review trigger, set low because the family clusters tightly when it appears at all. It flags a passage for reading and decides nothing, and it should not be run over fiction. ## Examples Constructed specimens. Written for this index. Never quoted from anyone's posts. ### Specimen 1 Before, exhibiting the tell: > There was a palpable camaraderie in the room, a fleeting and unspoken sense that something had shifted. Amidst the noise a whimsical thought surfaced and dissolved again before anyone could name it. Underneath it all sat a pang of unease, gossamer thin, the kind of feeling that lingers long after the moment that made it has gone. The air held a certain electric quality, ineffable and faintly bittersweet, and the whole scene seemed to shimmer at its edges in a way that nobody present would have been able to articulate afterwards. After, repaired: > Four of us stayed after standup and argued about the pricing page for an hour. Nobody won. The page shipped with the old headline and a new second line that somebody muttered near the end. Note: The repair supplies the event the atmosphere words were standing in for. ## False positives Who legitimately writes this way. Fiction, memoir and literary journalism use this vocabulary as ordinary equipment, and a novelist writing palpable is doing their job. Reinhart measures model output against human writing at population level and makes no claim about any individual text, which is the correct reading of the entry too. Amidst is standard in British and Indian English and carries no register signal in either. Reflective and devotional writing is built from unspoken and fleeting. The discriminator is whether an event is attached: a described room with named people in it can carry any of these words, and a room with only atmosphere in it is the thing worth flagging. Never run this entry over fiction, where a hit means only that the text is fiction. ## Model attribution Tied to the GPT-4o generation by Reinhart, who report larger differences from human writing for instruction-tuned models than for base models. Later generations show a different vocabulary, so read this entry as dated. ## Sources 1. Reinhart et al., Do LLMs write like humans? PNAS 122(8) (arXiv:2410.16107) https://arxiv.org/abs/2410.16107 (tier: peer-reviewed; accessed 2026-08-14) 2. vale-ai-tells, 111 machine-checkable rules https://github.com/tbhb/vale-ai-tells (tier: community; accessed 2026-08-14) 3. chrisgherbert gist, ChatGPT cliches https://gist.github.com/chrisgherbert/c734ec50ae464135be57cd03b84281f9 (tier: community; accessed 2026-08-14; inverted evidence, cited for what it catalogues rather than what it advocates) ## Status history Ids are permanent. A retired tell keeps its id and its page. - 2026-08-15, Fading: Fading. The family is tied to the GPT-4o generation that Reinhart measured, and the frontier science split we extracted in August 2026 shows a different profile entirely. The vocabulary also sits on published avoid lists, which is the condition under which a marker stops separating anyone from anyone. ## License CC BY 4.0. https://creativecommons.org/licenses/by/4.0/ Attribution: The AI Tells Index, feedsquad.com/ai-tells Reuse the data, including commercially. Keep the attribution line. --- Dataset version 1.0.0. Schema version 1. Part of [FeedSquad](https://feedsquad.com). Built by [Herman Foundry](https://hermanfoundry.com) from Levi, Finnish Lapland.