YouTube Music begins labeling AI-generated songs, built on what labels and distributors declare
YouTube Music now labels AI-generated songs based on what labels and distributors declare, but has not said whether a label changes recommendations or royalty payouts.
YouTube Music starts labeling AI-generated songs using partner disclosures and its own detection tools
YouTube Music has begun labeling songs made "primarily using generative AI," with the labels expanding to the main YouTube app. A label appears on the player page, and tapping it opens more information. YouTube says it identifies tracks using partner information plus its own detection tools, and the test is whether AI played a "significant role."
Music partners declare AI use when they deliver content, choosing Fully Gen AI, Partly Gen AI or No Gen AI, and they interpret the definitions themselves. Pitch correction and automated mastering count as No Gen AI. If a partner provides no information, YouTube relies solely on its own signals. Neither the blog post nor the help page explains how labeled tracks are treated in recommendations or royalty payouts. Apple Music has told partners its AI tags become visible "Made With AI" labels later this year, and Spotify has tested AI disclosures in credits since April.
The label is only as reliable as the declarations behind it, and the platform has not said what a label changes for the money.
- Declaration first.The label starts from what partners say, so its accuracy depends on how each one reads three loosely defined categories.
- Detection as backstop.YouTube's own signals carry the whole load exactly when a partner declares nothing.
- Money unanswered.Nothing YouTube has published says how labeled tracks are treated in recommendations or royalty payouts, and that is the part rights holders will price against any license.
I learned this replacing hand-submitted college radio playlists with independent monitoring: self-reported usage invites gaming, and an independent count changes where the money goes. Rights holders should treat the label as a claim to test, not as a count they can rely on, and any license that leans on it should say who bears the cost of a wrong label.
- CommercialDeal ExecutionAsk each distribution partner how it classifies Partly Gen AI, and align that definition across catalogs before the next delivery cycle.
- BuildDetection & TelemetrySample labeled and unlabeled tracks from your own catalog each month and log any mismatch against your declarations.
- LegalIndustry PositioningPress the platforms for a written statement on whether labels affect recommendations or payouts.
Streaming services file competing proposals for 2028–2032 songwriter royalty rates
Proposals filed October 5 with the Copyright Royalty Board, with public redacted versions on October 7, will set mechanical royalties for on-demand streaming from 2028 through 2032. The headline rate is 15.35% of service revenue in 2027. Spotify, Google (YouTube), Apple and Pandora each propose keeping 15.35%. Amazon proposes 13.4% and would cut the standalone rate from 26.2% to 22.8% of total content costs. Google also proposes that rates apply only to the share of plays the license covers, saying only one of six YouTube content categories qualifies.
The Operator’s Take: The coverage does not mention AI, so treat these filings as the baseline price any AI music license will be compared against.
Commercial · Economic Model Benchmark any AI music licensing ask against the rates each party filed, noting that Amazon's 13.4% is the lowest proposal on the table.
Taylor Swift extends voice and likeness trademark filings to the UK, EU and Australia
TAS Rights Management, LLC is listed as owner of the filings. US applications filed April 24 cover recordings of Swift saying "Hey, it's Taylor" and "Hey, it's Taylor Swift," plus a photo of her holding a pink guitar. Sound marks for the spoken phrase were filed October 6 in the UK, EU and Australia (Australia accepted it October 12, and it is now in its opposition period); image marks followed in the UK and EU on October 7. All claim the April 24 priority date. Lawyer Josh Gerben, who first spotted the US filings, said registering a celebrity's voice is "a new use of trademark registration that has not been tested in court before." The article reports no specific deepfake incident.
The Operator’s Take: Voice is moving from a right argued case by case toward a registered asset, but it is untested in court, and a registration helps only if the owner can find the imitation.
Legal · Industry Positioning Ask talent and catalog counsel whether voices and signature phrases are worth registering, and what monitoring would be needed to enforce them.
ChatGPT appears to drop UTM tracking, pushing AI referral traffic into "direct"
OpenAI appears to have stopped adding UTM parameters to links in ChatGPT; the utm_source=chatgpt.com tag was introduced in June. Links now open in a way that suppresses referral information, so analytics tools may label the visits as direct. Adobe's Cedric Huesler examined about 500 pages where ChatGPT had been at least 5% of visits and compared October 9–13 with the week before: ChatGPT traffic fell by about 31,000 visits a day while direct traffic rose by about 26,000. The article frames this as an attribution problem, not necessarily a real drop, and OpenAI has not confirmed the change was intentional.
The Operator’s Take: Measured referrals just became less reliable, which weakens anyone who uses ChatGPT traffic as proof of value in a licensing or sponsorship conversation.
Build · Detection & Telemetry Review referral, direct and landing-page trends together before reporting ChatGPT performance, and annotate dashboards with the mid-October change.
IAB Australia's audio glossary counts a play after one second of listening
IAB Australia's Audio Advertising Council compiled the Audio Buyer's Guide Glossary 2026: 10 pages, 55 terms, six categories. A play counts when a user listens for more than one second, an audible ad impression needs two continuous seconds, and active streams, bounce rate and average listening time use 60 seconds. A "month" is defined three ways: 28 days, 30 days, or an undefined calendar month. Podcasts are on-demand audio longer than 10 minutes. Host-read ads, video podcasts, sponsorship and attribution are not defined, and the glossary lags the IAB Tech Lab podcast guidelines finalized September 29.
The Operator’s Take: Buyers and sellers who count in different units are pricing different products; a one-second play and a 60-second stream are not the same thing.
Commercial · Deal Execution State the counting rule (one second, two seconds or 60 seconds) in every audio deal and reporting schedule.