Subtly altering word choices for watermarking degrades the quality of AI writing

Overview Debates Related claims Topics Sources

In earlier work, the research group behind IdeaLens found that human-written stories differ structurally from AI-generated stories and that different chatbots tend to use identical arguments when writing essays.

An 'AI' label appears on the player page directly below a track's name in YouTube Music, and tapping it opens a menu detailing how AI contributed to the track

Under YouTube Music guidelines, standard pitch-correction tools like Auto-Tune are categorized as 'No Gen AI' and do not trigger generative AI disclosures.

YouTube plans to expand its music AI labeling system from the YouTube Music app to the main YouTube app and video player.

Apple Music plans to launch 'Made With AI' labels as part of its AI Transparency Tags framework later in 2026.

YouTube Music launched a 'Made with AI' labeling system on October 14, 2026 to identify tracks where generative AI was used for major components like composition or vocals.

Spotify introduced voluntary AI labels and updated its policies to prevent 'AI personas' from receiving algorithmic recommendations.

Music partners uploading to YouTube can declare content as 'Fully Gen AI,' 'Partly Gen AI,' or 'No Gen AI,' while YouTube can supplement disclosures using its own detection tools.

A survey conducted by Deezer and Ipsos earlier this year found that 97% of respondents could not distinguish AI-generated songs from human-made music in a blind test, while 80% desired clearer labeling

Substack partnered with AI detection tool Pangram to let readers scan posts for estimated AI usage, though creators can disable automated scanning in publication settings.

Meta applies 'AI info' labels across Facebook, Instagram, and Threads to detected AI-generated content and, starting in June 2026, to AI-generated advertisements.

An AI detector analysis by Pangram found that 41% of long-form posts and 30% of short-form posts on LinkedIn between April and June 2026 were AI-generated.

Substack enables creators to add disclosures to their content explaining how AI was used during writing or confirming if no AI was used.

TIDAL labels recordings identified as entirely AI-generated, enables users to disable them, and excludes fully synthetic tracks from royalty attribution.

TikTok requires creators to label AI-generated content, appends invisible metadata watermarks, and allows users to toggle the amount of AI content in their feeds.

Tests run on excerpts of Thélyson Orélien's novel C'était ça ou mourir using the detection program Pangram indicated a 100-percent probability of AI-generated text.

French writer Samuel Fitoussi revealed himself as one of five individuals behind the anonymous X account that publicly posted AI detection tests of C'était ça ou mourir on September 21, 2026.

The jury of the Prix Goncourt removed Thélyson Orélien's novel C'était ça ou mourir from its long-list citing a strong probability of AI use and past plagiarism.

Thélyson Orélien acknowledged using research tools to help with foreign idioms while writing C'était ça ou mourir, but stated no sentence was created by a machine.

Author Thélyson Orélien issued a statement on October 13, 2026 denying that he used artificial intelligence to write his novel C'était ça ou mourir.