AI-powered news licensing is beginning to move towards a "pay-per-use" model.

(Image caption) The new generation of Siri's interface demonstrates how the AI assistant can directly answer instant questions, echoing Apple's negotiations with publishers to allow Siri to access and use news content as information infrastructure.

A news article has introduced a second price.

How much is a news story really worth? For the past two decades, this question has generally had a few answers: the price of a newspaper, a subscription, an ad impression, or a database license fee. AI is proposing another algorithm. If a user doesn't open a news website, click on the original article, or subscribe to a newspaper, but simply says to their phone, "What happened at the Federal Reserve today?", Siri reads a report recently published by a media outlet, extracts a few facts, and then answers in its own words—how much should this use of information be worth?

Apple is attempting to answer this question. The Wall Street Journal revealed on the evening of August 12th that Apple has been discussing new multi-year content agreements with news publishers in recent months, hoping to give the upgraded Siri access to real-time news and information. Negotiations are still ongoing, and Apple has not disclosed the final partners or pricing, so it cannot yet be described as an established industry standard. However, one detail from the negotiations may be more noteworthy to the news industry than the agreement itself: Apple is discussing variable compensation, meaning publishers may receive payment only when their content is actually used by Siri, rather than paying a fixed fee upfront for a large amount of content usage rights to the AI company. Related reports indicate that Apple has discussed an overall payment budget in the nine-figure dollar range. (The Wall Street Journal)

If this arrangement is ultimately implemented, the pay-as-you-go pricing model will have the opportunity to be tested in the real world by major consumer AI assistants and mainstream publishers globally. The news industry has spent over two decades learning to calculate page views; the next step may be learning to calculate inference.

Siri's answers start from the news scene.

When Apple unveiled its next-generation Siri AI in June, it already proposed enabling Siri to answer a wider range of questions and perform more operations across profiles, apps, and different tasks. Two months later, The Wall Street Journal revealed that Apple was in talks with publishers about content agreements, hoping to add another crucial capability: enabling Siri to obtain continuously updated news and real-time information. (Apple; The Wall Street Journal) This is a seemingly simple but actually very tricky product requirement. Large language models can answer questions about history, concepts, and general knowledge, but news has a completely different time structure: a company has just released its earnings report, an election is being counted, a central bank has just changed interest rates, and yesterday's correct answer may be outdated this morning. The model itself cannot conjure up what happened this afternoon out of thin air; someone has to be on the scene first, obtain documents, make phone calls to verify, sit in court, check the figures, and then a news organization takes on editorial responsibility to publish the information. The AI answer may end up being only a few lines long before the answer is generated.

This is precisely the economic context of Apple's negotiations with publishers. As Siri gradually becomes the gateway to information, what Apple needs is no longer just a static news database, but a continuously updated, verifiable, time-stamped, sourced, and accountable information supply. News thus acquires a new identity: the real-time information infrastructure of an AI system.

(Image caption) Apple CEO Tim Cook with the Apple logo, symbolizing Apple, as a tech giant, negotiating multi-year content agreements and variable compensation models with news publishers.

The economic unit of news may be changing.

The formula most familiar to traditional digital media is articles, websites, traffic, and then advertising and subscriptions. Generative AI is dismantling these intermediate steps: users ask questions, AI retrieves, understands, summarizes, and cites them, directly providing the answer, even if the reader never visits the media website. The most profound impact on journalism isn't even the decline in traffic, but rather the ineffectiveness of the economic units previously used to calculate news value—an article not being clicked doesn't mean it wasn't used; a media website not being visited doesn't mean its interviews, data collection, and judgments weren't involved in the generation of the AI product.

Therefore, a news article in the future may require a new rights ledger: AI searches for the content, the system retrieves a section of information, the model uses it to generate a summary, the final answer is tagged with its source, Siri converts the facts into a voice response, the content is used for model training or fine-tuning, or it is used in AI products with direct commercial value. The value of these uses is not the same, but if each can be identified, recorded, and priced, news content could potentially generate granular usage rights and revenue rights, just like other digital assets. GFM.News calls this structure "Content as Asset": the value of content is no longer limited to its initial publication but can be repeatedly generated through searching, summarizing, citing, translating, retrieving, training, and commercial use. Apple's current negotiations are not enough to prove that this market has already formed, but they make it more conceivable for the first time.

How does this differ from most AI-powered news licensing today?

Over the past few years, news organizations have signed numerous cooperation agreements with AI companies. Associated Press, Axel Springer, the Financial Times, Le Monde, Prisa Media, and News Corp, among others, have established content partnerships with OpenAI. For example, the Financial Times agreement allows ChatGPT to use source-credited FT content, and also involves model improvement and product collaboration. The specific financial terms of many of these deals have not been disclosed. (Reuters) This generation of agreements confirms at least one previously controversial point: some large AI companies have begun establishing paid licensing relationships for professional news content. Apple's proposed solution takes this a step further: how exactly should payments be made?

The advantages of fixed licensing fees are obvious: publishers know how much they can earn in a year, and AI companies can obtain relatively broad usage rights. The disadvantages are equally clear: a financial database used extensively by AI every day, and a small content library used only a few thousand times a year, would have their actual value difficult to accurately reflect if both were priced according to a one-time negotiated price. Usage-based pricing is closer to the logic already familiar to the music industry. Spotify and Apple Music don't need to buy all the songs in the world upfront; when a song is played, the system records it, and the rights holder receives revenue according to the contract and distribution rules. The rights structure of news is far more complex than music, and the streaming model cannot be simply copied, but the basic direction is beginning to converge: usage is recorded, rights are confirmed, and revenue is distributed according to rules. A "royalty market" for the news industry is no longer unimaginable, at least technically.

(Image caption) Reporters and editors process news articles in a busy newsroom, presenting the true cost of news production. This content is becoming a core source of real-time information supply for AI systems.

The real difficulty is who will keep the books?

All pay-per-use models rely on one premise: usage must be measurable. This is precisely the weakest link in today's AI content market. If an AI company claims an article was used over 10,000 times this month, how can the publisher verify this? If the model reads data from ten media outlets and generates a response, how much did each contribute? If one news article only provides background information, while another provides the core facts, should both receive equal compensation? Without transparent records, pay-per-use may ultimately just be a bill issued by the platform itself.

Therefore, the assetization of news content requires more than just contracts; it also needs technical standards. SPUR (Standards for Publisher Usage Rights), established this year, has brought together news organizations such as the BBC, Financial Times, Guardian Media Group, Mediahuis, Sky News, Telegraph Media Group, and Associated Press, aiming to establish a shared framework for content use reporting and rights expression. This framework would allow publishers to control how their content is used and obtain data on "how content is used" within the AI value chain. SPUR is also currently advancing the Telemetry Standard, hoping to ensure that content use itself can be recorded and rewarded. (The SPUR Coalition) Another developing system is Really Simple Licensing (RSL). It attempts to add a machine-readable rights layer between websites and AI systems, allowing publishers to specify whether content is permitted for AI use and what price applies, including subscriptions, pay-per-crawling, or even pay-per-inference. The RSL Collective further aims to create a collective licensing and royalty distribution mechanism. These systems (RSL Standard) are still very young, but they all point to the same problem: the content market in the AI era needs its own unit of measurement.

Why Collective Empowerment is Becoming Important

A large media conglomerate can sit directly at the negotiating table with Apple, OpenAI, or Google, while a local newspaper may not have this capability, and a professional media outlet with only a few dozen reporters will find it even more difficult. If the future AI content market relies entirely on one-on-one negotiations, the largest news organizations will obtain the best prices, and tens of thousands of small and medium-sized media outlets are likely to continue to be excluded from the market. This bears a historical resemblance to the reason why rights management organizations such as ASCAP and BMI emerged in the music industry—if a content market has millions of works, tens of thousands of rights holders, and billions of uses, the transaction costs of one-on-one contracts will eventually become so high that the market cannot function, and standardization will gradually become an integral part of the market itself. SPUR has chosen to establish common rules, while RSL is attempting to establish technical standards and collective licensing. Both paths are still in their early stages, but the direction is clear: to reduce individual negotiation costs while gaining greater bargaining power.

(Image caption) The interfaces of Apple News and Apple News+ across multiple devices demonstrate Apple's established experience in content distribution and publisher partnerships, laying the foundation for extending human reading to AI applications.

Why Apple might become a unique testing ground

Apple's entry into this market has a unique advantage that other large AI companies don't fully possess—it has been operating in content distribution for years. Apple News and Apple News+ have built relationships between publishers, users, content, and revenue sharing over the years. Therefore, Apple's challenge isn't learning how to collaborate with publishers for the first time, but rather extending its content distribution system, originally built on human reading, to AI reading. From this perspective, this isn't Apple's first time dealing with "content use—measurement—distribution"; the real new question is whether the system used to measure "what people read" can be extended to measure "what AI used." Previously, it was about a person reading an article; in the future, it will be about an AI reading an article and answering questions for a person—these two behaviors still lack a fully mature and unified solution within today's copyright, licensing, and commercial systems.

If Apple does adopt a pay-as-you-go model, it could become a significant market experiment. Its success or failure hinges on at least four things: whether usage can be audited, whether sources can be clearly attributed, whether the price is sufficient to support the cost of original news, and whether small and medium-sized publishers can enter the market. The absence of any of these could potentially transform the new licensing market back into a platform-dominated distribution system.

This has become a global issue.

Apple's negotiations with publishers are just one part of this systemic shift. Europe has long addressed the rights of platforms to use news content through copyright and neighboring rights systems; Australia has established a bargaining mechanism for news media; and Canada has re-examined the value distribution between digital platforms and news organizations through the Online News Act. While the design, implementation, and controversies of these systems differ across regions, they all face a common challenge: how should the costs of news production be redistributed with the commercial value gained by large technology platforms as they become information gateways?

Generative AI has taken this issue a step further. In the past, the debate centered on how platforms should pay for displaying, aggregating, or linking news; now another question is emerging: how should we price the news when machines read, understand, summarize, and use it?

There is also a dangerous version of news assetization.

The media shouldn't prematurely declare victory simply because "AI is willing to pay." Apple's own record serves as a reminder: in late 2024, it began testing AI-generated news notification summaries, which subsequently produced numerous headlines that were inconsistent with the original text or misleading, drawing criticism from news organizations such as the BBC. Apple suspended the feature in January 2025. This illustrates that even Apple may not get it right the first time when integrating AI into the news process. The new market may also replicate the platform problems of the past two decades: if a few platforms like Apple, Google, OpenAI, and Meta become the main buyers, and tens of thousands of news organizations compete with each other, the news industry may still be in a weak position.

Another risk lies in editorial incentives. If AI usage eventually becomes a revenue metric, will media outlets start producing content for AI—which headlines are most easily searched, which articles are most likely to get responses? Twenty years ago, SEO transformed the newsroom; AI licensing may one day give rise to Answer Engine Optimization. If economic incentives lack clear institutional boundaries, content assetization could ultimately reverse and change the news itself. Therefore, a healthy AI news market needs a fourth layer—editorial independence: prices can be determined by the market, but news itself cannot be determined by market prices.

(Image caption) Digital Rights and Royalty Management Dashboard: Visualizes content usage measurement, classification, and revenue distribution, corresponding to the core mechanisms of pay-as-you-go and Content as Asset proposed in the article.

GFM.News is not seeing an Apple deal.

Since founding GFM.News, I've been pondering a question: will AI destroy the business model of news, or will it force the news industry to rediscover its asset value? The answer is likely both. AI will take away some search traffic, but it may also create a content usage market that didn't exist before. The key lies in whether news organizations possess their own rights, data, standards, and pricing power. This is why GFM.News incorporates "Content as Asset" into its institutional financial media infrastructure: an article that has undergone research, interviewing, verification, and editing should not only have one URL, but should also have its own source chain, timestamp, author rights, version history, language version, citation relationships, authorization scope, and usage history—search, summarization, citation, translation, training, retrieval, and commercial use can each have their own authorization rules.

For this architecture to work, several things need to happen simultaneously: first, clearly define the boundaries of content rights; second, use technology to identify and track content usage, allowing news to be repeatedly authorized rather than quickly becoming obsolete after a day of publication; and third, the long-term accumulation of data, citations, and usage records may form new research and judgment capabilities. If a Chinese research paper can be reused in English, Spanish, Arabic, or other language markets, its value boundaries are no longer determined by the language of its initial publication. Ultimately, what determines whether anyone is willing to buy this asset is trust—a text without a source, untraceable, and whose authenticity is difficult to verify, even if generated in large quantities, is unlikely to become a high-value information asset. In a sense, the cheaper AI becomes, the more expensive credible content may actually be.

The news industry needs its own "content exchange".

Apple's negotiations are still just negotiations. We don't know which publishers will ultimately sign, how the nine-figure budget will be allocated, how many cents a single Siri use will cost, or even if it will actually be priced this way. So it's too early to declare that "the era of news streaming royalties has arrived." But the market is converging in one direction: Apple is negotiating usage-based compensation, RSL is designing pay-as-you-go pricing, SPUR is pushing for a common standard of rewards for use, and major publishers are still signing AI licensing agreements. Reuters editor-in-chief Alessandra Galloni, when discussing AI and news in July, also put forward a clear principle: news should be licensed, attributed, and fairly compensated. (Reuters) These previously scattered lines are slowly converging.

The next question may no longer be whether AI companies should pay for news, but rather who should build this market—does the news industry need a common set of rights standards, measurement methods, authorization mechanisms, clearing systems, and price discovery methods? The "content exchange" mentioned here doesn't necessarily refer to a traditional stock exchange; it's more likely to be a market infrastructure where rights can be confirmed, content use can be measured, authorizations can be enforced, revenue can be cleared and distributed, and the entire process can be audited by a third party. The music industry took a long time to build its royalty system, and the financial market took centuries to establish trading, clearing, custody, and auditing systems. If news is becoming a crucial raw material in the AI era, it will inevitably need its own market infrastructure sooner or later.

A single Siri response could be the starting point.

Imagine a future person driving in Los Angeles, not opening any news websites or seeing any media homepages, but simply asking, "Why is the market down today?" Within seconds, AI will retrieve breaking news reports from several news organizations, central bank documents, and market data, verifying, comparing, and summarizing them, before providing a thirty-second answer. Those thirty seconds might contain hours of interviews by a journalist, years of professional expertise from an editorial team, and decades of credibility built by a news organization. If AI can accurately calculate how much computing power it uses, how many tokens it consumes, and how much a single inference costs, then sooner or later it should also be able to calculate: whose news it uses, and how much that news is worth.

What Apple is ostensibly negotiating is a Siri content agreement. What journalism really needs to establish is the ledger for the next generation of content economy.

Disclaimer

This article is based on publicly available information as of the time of writing and is intended for news research and institutional analysis only. It does not constitute investment, legal, business, or transaction advice; related negotiations, products, and policies are still subject to change.