Outside Verifiers Get a Look at ChatGPT Ads as the Music Industry Fights a ‘Free Ride’ in Australia
OpenAI lets outside verifiers check ChatGPT ads, the music industry fights a copyright “free ride” in Australia, and the open question is who holds the record of use.
OpenAI moves to make ChatGPT Ads more measurable and more visual
OpenAI is widening how advertisers can check ChatGPT ads. DoubleVerify and Integral Ad Science are piloting brand suitability reviews in controlled test environments, without access to real user conversations, and OpenAI says ChatGPT reaches 1.2 billion people a week. It is also expanding conversion tracking through Kochava and 10 other partners, and is testing geo-based incrementality, which measures sales that would not have happened without the ad, with Haus, Measured and WorkMagic.
Ads today carry an advertiser name, logo, headline, description, landing page and image. OpenAI plans to test image ads in the US later this month, shown during image generation and labeled apart from ChatGPT’s own images. One agency executive noted ChatGPT sees about 2.5 billion prompts a day against roughly 15 billion Google searches, with one ad slot rather than several. Pricing, and what advertisers can see in reports, were not disclosed.
The platform is no longer asking advertisers to take its word, and that is the moment its numbers start to be worth something.
- Independent eyes are the product.Two verification companies reviewing how ChatGPT applies its brand rules, in a controlled test and without seeing user conversations, give a buyer a second opinion the platform did not write. It is a pilot with limits, not an audit of live ads.
- Measurement follows the money.Conversion tracking through 11 partners and early incrementality tests tell advertisers what happened after the click. The inventory is small, about 2.5 billion prompts a day against Google’s 15 billion searches and one ad slot per answer, so a buyer’s first question is whether the results are real, not whether they are big.
- The record belongs to someone.Pricing and reporting terms were not disclosed. Whoever holds the log of where an ad ran, and next to what text, decides who can dispute a bill or a brand-safety call.
I learned this negotiating data licensing with rights organizations: usage data becomes an asset in its own right once both sides trust it. The infrastructure question under every content industry is the same here, whether you own the harness or rent it, and a rights holder whose work appears beside a paid answer will want to see that log.
The open question is who sees the record first.
- BuildDetection & TelemetryAsk any AI platform you buy from for impression-level logs, not summaries, and keep your own record of where each ad appeared and next to what content.
- CommercialDeal ExecutionWrite third-party verification and a right to check the platform’s counts into the next insertion order before spend scales.
- NetworkIndustry PositioningPress your trade body to ask for one common reporting format for ads inside AI answers, so every buyer gets the same view.
ARIA boss tells Australian lawmakers OpenAI and Anthropic are ‘asking Government for a free ride’ on copyright
At a Sydney hearing on October 6, the head of ARIA, the Australian recording industry body, told Parliament’s Joint Select Committee on Artificial Intelligence that OpenAI and Anthropic want Australia to ease copyright rules so they can train models there, warning it could make the country a “Trojan horse for Big AI.” Anthropic suggested a narrow form of conditional approval for training, with conditions such as investment in Australian creators; OpenAI said models should be able to train on publicly available information, with chances for rights holders to collaborate.
ARIA says changing the law would legalize ongoing piracy and weaken licensing markets, and that global licensing deals already cover about 90% of the world’s sound recordings. It does not oppose frontier AI training in Australia as long as it follows local terms. Australia ruled out a text-and-data-mining exception in October 2025, and the committee is due to report by November 30.
The Operator’s Take: The 90% coverage figure is ARIA’s own and has not been checked, and the AI developers’ cost case was not put on the table. Whichever side can show a count of licensed use will set the terms of the debate before the committee reports.
Advocacy · Industry Positioning Before the committee reports, send it a short, evidenced note showing which of your catalog is already licensed to AI developers and on what terms.
With a new CFO, three licensing deals and $775M raised, what’s stopping $5.4B-valued Suno from an IPO?
Suno, valued at $5.4 billion after a Series D of more than $400 million in June, has raised more than $775 million in total, hired its first CFO on September 14, and has licensing deals with Warner Music Group (settled November 2025), BMG (August 2026) and Believe (September 2026). It has not said it plans to list. Its last disclosed figures, from February, were 2 million paying subscribers and $300 million in annual recurring revenue; the CEO says it is now “far beyond” both, without numbers.
Music Business Worldwide lists what stands in the way: UMG and Sony’s second suit, filed September 18 over the v6 models and more than 60,000 recordings, with a theoretical statutory maximum of about $9 billion; a fair use ruling not expected before the second half of 2027; claims from Round Hill, Gerencia 360, SOCAN and Koda; a Munich court ruling against Suno in a GEMA case on July 31; and an Italian competition authority probe opened October 6. Under US accounting rules, a probable and estimable loss must be recorded, which a listing filing would have to disclose.
The Operator’s Take: An IPO would put a forecast and a liability estimate on paper that someone has to sign, which is the first time a rights holder could price against a number with consequences. I learned this early in my career, valuing publishing companies for M&A: deal terms are set by whoever controls the forecast.
Commercial · Deal Execution If you license a platform heading toward a listing, write a right to its audited usage figures and a most-favored-nation clause into the deal before a filing sets the numbers.
Meta’s shopping agent must fill its top 5 picks with well-known merchants
Profound, an AI marketing analytics firm, says it pulled the instruction file that governs Meta’s Muse shopping agent on September 25 and shared it with State of Brand. According to that file, which Meta has not commented on and which may have changed, the agent must fill its top five picks with well-known merchants, drops listings priced far above or below the category norm, searches no more than three open-web stores per request, and favors direct-to-consumer brands over resellers.
In-chat checkout works only for products in Meta’s catalog flagged for agentic checkout; open-web items go to browser checkout. Profound’s test of 50 prompts through one account found the catalog supplied about 95% of candidates in the median case, but final picks leaned toward web results. Profound counts more than 2.5 million downloads. Amazon has blocked Muse, saying the agent does not identify itself as a bot. The rules are reported, not independently verified, and the study used a single account.
The Operator’s Take: A ranking rule nobody outside Meta can see is deciding which merchants an AI agent shows first. I learned this building scoring models that enterprises used to judge their suppliers: a score is only accepted when the parties it judges can see how it was built.
Build · Detection & Telemetry Record how often AI shopping agents visit and rank your products, and ask the platform to publish its merchant-ranking rules.
Nicolas Cage Says He ‘Probably’ Won’t Work With Amazon Again After Refusing to Sign an AI Waiver for ‘Spider-Noir’
Actor Nicolas Cage said at New York Comic Con on Saturday that he refused to sign an AI waiver for Amazon’s Prime Video series Spider-Noir and will “probably” not work with the studio again. “I am not an AI-friendly actor,” he said. The article does not describe the waiver’s terms; it says only that it would have let Amazon use AI on the show.
The series was canceled last month, days before it earned 11 Emmy nominations, after 2.6 billion minutes viewed in its first six weeks. Cage also cited a $50 billion Amazon investment in OpenAI, a figure TheWrap reports as his claim, not independently verified. Amazon did not immediately respond to a request for comment, and the article does not mention SAG-AFTRA.
The Operator’s Take: A blanket waiver puts a performer’s consent and a studio’s AI rights into one signature, and nothing public says what the studio may do with it. Work building an oral-history video archive that mixed professional and community voices turned on one point: when people’s voices are the content, consent and attribution are the product.
Legal · Deal Execution Split AI use of a performer’s work and likeness out of the standard waiver into a separate, scoped, priced consent with a record of every use.