Physix FrontierApple Buying Content for Siri: Right Direction, But Catching Up on Debt
The most valuable piece of information in this article isn't the fact that Apple plans to pay publishers, but rather how it plans to pay: settling based on actual content usage volume, instead of a one-time buyout. This detail is more interesting than "Apple finally starts throwing money at its AI shortcomings," as it could determine what the AI content licensing market looks like for the next few years.
Let's start with the conclusion. Apple's move here is directionally correct, but essentially playing catch-up. Since Siri's launch in 2011, it has lagged behind competitors by an entire era in terms of "understanding user intent." Now, trying to feed AI by buying news content addresses the issue of "knowledge freshness," but Siri's real root cause lies in "conversational ability and intent understanding"—these are two different things.
I started heavily using an AI model for information aggregation about a month ago, and looking back, that experience really illustrates the point. When asking "What are the big tech stories today?", this model automatically scrapes, filters, sorts, and finally gives you a summary in natural language, with sources clearly cited. Meanwhile, Siri still suffers from the same old problem when encountering slightly complex instructions: either it doesn't understand, or it answers off-topic. It's not that there isn't enough content; it's that the brain isn't smart enough.
So, seeing this WSJ news, my first reaction was: Apple finally realizes the importance of knowledge bases. But thinking one layer deeper, things aren't that simple.
The variable compensation structure proposed by Apple translates to "I only pay if your content is used." This design is unlike any AI content licensing agreement I've seen before. The traditional approach, like OpenAI's deal with News Corp, involves negotiating a large upfront package, setting an annual fee, and allowing unrestricted use of content. Apple's pay-per-use model offers publishers a higher ceiling because if Siri AI scales up massively, revenue could be several times higher than fixed annual agreements. But the risk is also there: if few people use Siri AI, publisher revenue will look pretty bad.
For Apple, this model is actually more cost-effective. It doesn't need to pre-pay for "content it might not use," keeping costs tied to actual usage and risks controllable. From a negotiation perspective, this structure paints a big picture for publishers, reducing the probability of talks falling apart. Based on my testing, I feel this might be intentional on Apple's part: knowing it entered late, it needs more flexible terms to leverage publishers who have already been signed by OpenAI and Google once around.
However, there is a contradiction here. While Apple is negotiating content licenses with publishers, Bloomberg reported that it plans to spend approximately $1 billion annually renting Google's Gemini model to power the new version of Siri. That model is in the 1.2 trillion parameter class. In other words, the new Siri's "brain" is rented, its "knowledge" is bought, and what Apple contributes is only the shell and ecosystem. This would have been unimaginable ten years ago. Apple's keynote sounds particularly nice, talking about being "profoundly more capable and conversational," but underneath, it's putting someone else's engine into its own car.
It's not fair to say Apple hasn't done anything. Within the Apple Intelligence suite, the engineering capabilities regarding on-device model scheduling, privacy frameworks, and cross-device context continuity are present. But in the AI race, the competitive landscape in 2026 is no longer about "can you use it," but "how much better are you than others." Whether a rented model plus bought content can beat OpenAI and Google's native combinations in terms of experience, I remain skeptical.
Back to the publishers. Apple's willingness to pay is certainly good news. Content creators have been freeloaded upon by AI companies for so long that finally someone is coming with money to discuss licensing. But I want to offer a reminder: don't get too excited yet. Paying by usage sounds fair, but the measurement standards are entirely in Apple's hands. What counts as "one use"? Does Siri quoting one sentence count as one time or one article? If a user reads the summary but doesn't click through to the original text, does it count as usage? If these details aren't hardcoded in the contract, there will be plenty of room for disputes later. I said something similar last week in the post about that AI resume rewriting tool: the tool itself isn't the problem, but if the terms aren't transparent, the trust relationship between users and providers becomes very fragile. The same logic applies to negotiations between publishers and Apple.
That said, Apple's move has positive significance for the industry as a whole: it acknowledges the value of high-quality human content in AI training and inference. Previously, the default practice was "scrape first, apologize later." Apple is now saying "negotiate payment first, then use." This stance itself will drive up market prices for content licensing, which is good for small and medium-sized publishers.
My advice is: if you run a content site or hold copyrighted content, now is a good time to negotiate licensing with platforms. But don't rush to sign exclusive deals. No one knows how much volume Apple's pay-per-use model will actually generate. Those holding content should wait and see the settlement data from the first year before deciding whether to bet everything on a single platform.
Whether Siri can turn things around with bought content and rented models, I remain in a wait-and-see mode. But the fact that Apple is willing to change its inertia of "building everything ourselves in a closed garden" and start spending money to buy external capabilities is, in itself, more noteworthy than Siri's progress.
📌 This article is compiled from Hacker News. Original source: https://www.wsj.com/business/media/apple-in-talks-to-pay-publishers-to-improve-ai-powered-siri-0641f64b
Copyright belongs to the original authors. This is a compilation and independent analysis based on public reports.
