Half the Uploads, Almost None of the Listening: How Tencent Music Can Call AI Music "Negligible"
Updated:
In July 2026, Deezer announced that fully AI-generated tracks had, for the first time, passed half of everything delivered to its service in a day — roughly 90,000 tracks daily at the June peak (Deezer Newsroom). Four months later, on November 17, the company that runs China's largest music apps told investors the opposite-sounding thing. Tencent Music Entertainment (TME), owner of QQ Music, Kugou, Kuwo, WeSing and now Ximalaya, said that streams of AI content on its platforms "remain negligible," with a declining share of listening (Motley Fool transcript).
At first glance, one of these statements has to be wrong. It doesn't. Deezer's headline figure measures what arrives at the door; TME's measures what gets played. Deezer itself reports that the same flood of AI uploads earns only about 1–3% of its total streams. The real question is not whether AI music is "taking over," but what sits between supply and demand — and who controls that space. On a Chinese platform, the answer is: almost entirely the platform.
What TME actually said, and to whom
The remarks came from Executive Chairman Cussion Pang during TME's third-quarter 2026 earnings call. In his prepared remarks he framed the point as a story about taste: as technology makes content more abundant, he said, "users are becoming more selective" (Digital Music News, quoting the call). He then pivoted to what TME would rather sell — music that "stands the test of time," and an expanding catalog of K-pop, Mandopop and TV-drama soundtracks by human artists.
Two details from the full call change how that line should be read.
First, "negligible" is a characterization, not a measured quantity. No percentage, no stream count and no definition of "AI content" accompanied it, and Pang's answers in the Q&A were delivered through an interpreter, so the exact English wording of those answers should not be over-read.
Second, the topic came back in a context that is anything but reassuring. Asked by a Bank of America analyst about the outlook for 2027 — not about AI — Pang named copyright protection as the first priority and said, in the interpreted rendering, that "low-quality and unauthorized AI-generated content" still weighs on subscription monetization. In the same answer he repeated that streams of AI songs "remain very low," citing "proactive content management" and legal action taken alongside leading global rights holders.
So within one call, AI music is both too small to matter in listening and large enough to drag on the subscription business. That is not a contradiction to dismiss; it is the most useful clue in the whole disclosure, and this piece returns to it below.

Image: Tencent Music Entertainment (via Digital Music News)
The financial backdrop matters too. TME's Q3 release, filed with the SEC, shows total revenue of RMB 8.94 billion, up 5.6% year on year, and IFRS net profit of RMB 2.22 billion, essentially flat against RMB 2.21 billion a year earlier (TME Q3 2026 release, SEC 6-K). Music-related services grew 10.2% and membership revenue 9.2%, helped by the newly consolidated Ximalaya, while advertising stayed weak. A company growing in single digits, telling the market that its human catalog is what users want, has an obvious interest in the AI question being answered "not a threat."
Two meters: what gets uploaded vs. what gets played
The cleanest public dataset on this gap comes from Deezer, and it is worth reading closely, because the two numbers in it move independently.
On the supply side, the curve is steep. Deezer says fully AI-generated tracks made up about 10% of daily deliveries in January 2025, roughly 44% — close to 75,000 tracks a day — by April 2026 (Deezer Newsroom, April 2026), and more than half at the June 2026 peak. On the demand side, the line is almost flat: fully AI-generated music accounts for only 1–3% of total streams. And a large slice of even that sliver is not people listening — Deezer says up to 85% of streams on fully AI-generated tracks in 2025 were detected as fraudulent and demonetized, against about 8% fraud across the whole catalog.
Three things sit between the upload count and the play count, and they are worth naming because they also explain TME's number:
- The recommendation layer. Since January 2025, Deezer has kept tracks it detects as AI out of algorithmic recommendations and editorial playlists. On a modern streaming service, a large share of listening that isn't a deliberate search flows through those two channels. Cut them off and a track can exist in the catalog without ever reaching a listener.
- The fraud filter. A meaningful share of AI uploads is built for streaming-fraud schemes, not for audiences. When those plays are stripped out, "consumption" shrinks further.
- The definition. Deezer's detector targets fully AI-generated music and, by its own description, "does not typically tag AI-assisted tracks." A human-written song with an AI-generated arrangement or vocal guide is not in the 1–3%.
In other words, the share of listening that goes to AI music is not just a measure of taste. It is partly an output of platform policy. A platform that throttles distribution and strips fraud will report low AI consumption almost by construction (this is the author's inference from Deezer's disclosures).
This site has already looked at the listener side of that equation — whether a "human premium" exists and whether audiences will pay for it. This article focuses on the plumbing instead: how a Chinese platform's structure shapes the number it reports.
The awkward fact: TME is also an AI music factory
It would be easy to read Pang's comments as a platform holding the line against generative music. TME's own filings tell a more complicated story.
In its fourth-quarter 2025 results, TME said its "one-stop AI music production platform" had served more than 150,000 artists and over 10 million users (TME Q4 2025 release via Tiger Brokers). The same release describes a self-developed multimodal model used to sharpen recommendations — producing what TME called a record share of recommendation-driven streams — AI-generated video clips built around song choruses, and a QQ Music AI agent running on Tencent's Yuanbao assistant that the company describes as a "system-level hub." By the Q2 2026 release, TME was saying its AI agents "now act as personal DJs" (TME Q2 2026 release, SEC 6-K). And on the March 2026 call that accompanied the Q4 results, CEO Ross Liang (Liang Zhu) described "explosive growth" in AI-generated songs (Caixin Global) — eight months before the chairman called their listening share negligible.
So TME is not anti-AI. Its position is narrower, and it has been consistent for some time: AI is welcome as a tool inside a licensed system, not as a source of unlicensed supply. Back in 2024 the company said AI "should be a supportive tool rather than a substitute for musicians" (RightsTech). Its 2025 ESG report, published in April 2026, says it uses content for AI training only once rights holders have authorized it, and has added AI-specific clauses to its user agreements (Music In Africa, summarizing the ESG report).
That distinction matters for the "negligible" claim. A platform that builds AI creation tools for 10 million users and also says AI streams are negligible must be drawing a line somewhere — between "AI-generated content" and "music made by people who used AI." TME has not said publicly where that line is.
Why a Chinese platform's meter reads differently
If AI listening share is partly a function of platform policy, then the structure of the Chinese market should push TME's number lower than a Western open-upload service's. Three structural features point that way.
1. A regulated front door. Since September 1, 2025, China's Measures for Labeling AI-Generated and Synthetic Content — issued by the Cyberspace Administration of China with three other ministries — have required AI-generated text, images, audio and video to carry labels, both visible ones and embedded metadata. Distribution platforms are expected to check incoming content for those markers and flag material that is confirmed or suspected to be AI-generated (Loeb & Loeb). On top of the legal duty sits the platform's own review machinery. According to TME's 2025 ESG report, the company removed more than 250,000 songs for rule violations or "reputational risks" using AI detection plus human review, handled over 600,000 high-risk copyright cases, removed about 27,000 tracks for "song theft," "song laundering" and trend-chasing knockoffs, and uses audio models to identify and tag AI-generated works (Music In Africa). These are the company's own figures and cannot be independently checked, but they describe a gate that is far more actively policed than a typical open-distribution pipeline.
2. A licensed catalog, not an exclusive one — but still a concentrated one. It is often said that Chinese streaming runs on exclusive licenses. That was true until July 2021, when China's State Administration for Market Regulation ordered Tencent to end exclusive licensing deals with record labels, after finding that its acquisitions had given it more than 80% of exclusive music library resources (Business Standard/Reuters). Exclusivity is gone; scale is not. TME still competes on the depth of its licensed catalog and on artist partnerships, which is exactly what Pang pointed to on the call: K-pop, Mandopop, drama soundtracks. Its regional deals point the same way — this site's earlier piece on T-Pop noted TME's 2024 stake in Thailand's GMM Music. A platform whose pitch is "the catalog you can't get elsewhere" has every reason to keep anonymous AI filler out of its storefront.
3. A paid membership base the platform must protect. At the end of 2025, TME reported 127.4 million online music paying users, up 5.3% year on year, and a monthly ARPPU of RMB 11.9, while online music monthly active users fell 5% to 528 million (TME Q4 2025 release via Tiger Brokers). Its growth engine has been its higher-priced SVIP tier; the Q3 2026 release again reports growth in both SVIP paying users and ARPPU, though without numbers. When revenue depends on persuading members to pay more for a "premium" experience, letting low-quality AI tracks clutter recommendations is a direct threat to the product. And because TME says a record share of streams is now recommendation-driven, its own algorithm largely decides how much AI music anyone hears (this is the author's inference from TME's disclosures).
Put together, "negligible" may be less a discovery about Chinese listeners than a description of a closed, curated system working as designed.
Reasons to discount the claim
None of this means TME's statement should be taken at face value. There are at least four reasons for caution.
It is an adjective, not a number. Deezer publishes a range (1–3%), a definition (fully AI-generated, not AI-assisted) and a detection method with stated accuracy. TME has offered none of these. "Negligible" relative to what — total streams, new-release streams, streams on one app? Is AI-assisted music counted? What about tracks made on TME's own AI production platform and released under a human artist's name? Without answers, the claim can't be compared with anything.
It arrives while disclosure is shrinking. In its Q4 2025 release, TME said that from the following quarter it would "discontinue the disclosure of certain quarterly operating metrics," including online music MAU, paying users and ARPPU, and would instead report total paying users once a year (TME Q4 2025 release via Tiger Brokers). The Q2 and Q3 2026 releases on file with the SEC duly contain none of those figures. TME's stated reason is that revenue and profit now better reflect a business built on tiered memberships, advertising and IP. Still, the effect is that investors now get one qualitative sentence about AI listening at the same time as they lost the quarterly user numbers that would put it in context. The last MAU figure published showed a decline.
The audience has a stake in the answer. An earnings call speaks to two constituencies at once: investors worried about the subscription model, and the labels and publishers whose licenses TME depends on. "AI slop isn't getting traction here, and we're suing alongside rights holders" is the message both want to hear. That doesn't make it false — Deezer, which has its own motives, reports a similar pattern — but self-reported data from an interested party needs an independent check that is not yet available.
The claim sits uneasily with TME's own warnings. If AI streams are negligible, why does low-quality and unauthorized AI content weigh on subscription monetization, as Pang said in the same call? One plausible reading is that the pressure comes from outside TME's walls: free AI covers and unlicensed versions circulating on short-video apps and on rival services, giving casual listeners less reason to pay. In the same answer, Pang went on to mention short-term pressure on membership and advertising, though the transcript does not tie that pressure explicitly to AI. If that reading is right, TME's on-platform AI share could keep falling even as the AI problem grows in the wider market — a declining number that is measuring the wrong place (this is the author's interpretation; TME has not spelled out the mechanism).
What would settle it
The gap between "half the uploads" and "negligible listening" is real, and it is mostly a gap between supply and curated demand. But whether TME's version of it is as small as claimed is an open question. A few things would close it:
- A number with a definition. A percentage of streams, the definition of AI content behind it, and whether AI-assisted and in-house-platform tracks are included.
- Visible labels. China's labeling rules already require AI audio to be marked, and TME says it tags AI works internally. It has not said publicly how, or whether, those tags are shown to listeners. Consistent public labeling, like Deezer's album-level tag, would let outside observers see how much AI music is actually in circulation.
- The year-end paying-user figure. TME's next annual disclosure will show whether its membership base held up through a year in which management acknowledged short-term pressure on membership.
- Off-platform data. If the real competition from AI music is on short-video and rival apps, the meaningful measure will come from the wider market, not from any single platform's earnings call.
The bottom line
Deezer and Tencent Music are not contradicting each other. They are describing the two ends of the same pipe. AI tools have made music supply nearly unlimited, while listening is still filtered by recommendation systems, fraud controls, regulation and paid catalogs — and in China, all of those filters sit with a handful of platforms. TME's "negligible" says more about how effectively it controls its own storefront than about whether Chinese listeners have rejected AI music. The more important signal from the call may be the line that got less attention: AI content is already weighing on what people are willing to pay for.
Sources
- Original article: Ashley King, "Tencent Music Says AI Music Consumption is 'Negligible' and 'Declining' Across Its Platforms", Digital Music News, November 17, 2026
- Tencent Music Q3 2026 earnings call transcript, The Motley Fool, November 17, 2026 (Q&A answers are interpreted)
- Tencent Music Q3 2026 unaudited financial results, Form 6-K exhibit 99.1, SEC
- Tencent Music Q2 2026 unaudited financial results, Form 6-K exhibit 99.1, SEC
- Tencent Music Q4 and full-year 2025 unaudited financial results, press release as republished by Tiger Brokers, March 17, 2026
- Guan Cong, AI music boom tests Tencent Music's business model, Caixin Global, March 17, 2026 (only the opening paragraphs were accessible)
- Deezer: AI music has surpassed 50% of new music uploads for the first time, Deezer Newsroom, July 21, 2026
- Deezer: AI-generated tracks now represent 44% of all new uploaded music, Deezer Newsroom, April 20, 2026
- Tencent Music removes 250,000 tracks amid tighter copyright and AI controls, Music In Africa, May 8, 2026 (summarizing TME's 2025 ESG report)
- Tencent Music is 'deploying advanced AI tools' to crack down on infringement, RightsTech, October 9, 2024
- China's AI Labeling Measures and Mandatory National Standards Take Effect September 1, Loeb & Loeb, March 2025
- Chinese regulator orders Tencent to give up exclusive music licensing, Business Standard / Reuters, July 2021