The budget may be large, but projected revenue cannot be calculated
Apple reportedly proposed a model that “compensates publishers when content is used.” But public reporting does not tell us whether one use means the moment content is retrieved, incorporated into an answer, or shown as a source. Without disclosed rates and reporting methods, even a large total budget cannot be turned into projected revenue for our company.
Apple was reported to be discussing multiyear deals with multiple publishers to provide current news and information for Siri AI. The compensation model under consideration was variable payment when content is used, rather than a standard fixed license fee, with the total budget reported to be in the nine-figure range.
However, this is reporting on negotiations, not disclosure of signed agreements. The exact budget, term, counterparties, allocation by publisher, rates, and even whether a minimum guarantee exists have not been confirmed. So rather than immediately calling this a “no-minimum-guarantee deal,” the first step is to separate the event that triggers payment from the rights being granted in the contract.
First separate confirmed reporting from the blanks
Only the broad direction of the negotiations is public so far. In particular, a “nine-figure budget” is not an amount promised to a specific publisher, so it should not be converted into Korean won and presented as projected revenue.
| Confirmed in public reporting | Not yet confirmed |
|---|---|
| Multiyear deals discussed with multiple publishers | Whether agreements have actually been signed and with whom |
| Use of current news and information for Siri AI | Scope of rights for training, indexing, retrieval, and answers |
| Variable compensation paid when content is used under consideration | Definition and rate for one use |
| Total budget in the nine-figure range under consideration | Term, allocation by publisher, and minimum guarantee |
| Siri AI beta announced for English-language users in late 2026 | Korean availability timing and how news sources will be displayed |
Apple officially announced that Siri AI will be able to pull current information from the web to answer questions and plans to begin a beta for users with supported devices set to English in late 2026. Because this announcement did not include compensation terms for news licensing, the product launch plan and publisher contract terms should be treated as separate facts.
You cannot reproduce a settlement from a single “usage” number
In usage-based contracts, event-level records matter more than a monthly total. An article may be retrieved as a candidate but not used in an answer, or incorporated into an answer without displaying a source or link.
COUNTER guidance for scholarly publishing distinguishes content activities such as searches, investigations, and requests, and also offers optional fields to indicate AI-system access. It is not a mandatory standard for an Apple contract, but it can inform reporting structures that do not lump distinct events together.
You might phrase the first question like this.
“What constitutes a billable use: content retrieval, access to the full text, incorporation into an answer, source display, or a click to the original? Will you provide counts and applicable rates by content ID, and how are failed responses, retries, duplicate requests, and cache reuse handled?”
If you receive only monthly totals, you cannot verify the payment calculated by the platform. Content ID, usage type, product and region, event count, rate, and reasons for exclusions or adjustments must be visible in the same row. The method for checking sample records, the raw-data retention period, and the dispute process are also negotiation items. The News/Media Alliance likewise calls for transparency into which content is used and for what purpose, along with source attribution in generated results.
Negotiate access for current answers and training rights separately
The right to retrieve and answer questions about today’s articles is not the same as the right to use those articles for model training. The News/Media Alliance distinguishes training, content display, and synthesis/summarization as different forms of use. The Authors Guild’s model clauses also address training rights separately from rights for retrieval-augmented-generation summaries and responses.
Neither resource shows Apple’s actual contract terms. Still, they provide grounds for breaking broad “AI use” language in a proposal into training and testing, indexing, retrieval and RAG, content display, summarization and answer generation, quotation of original text, link display, and model improvement. Each right should also specify term, territory, language, exclusivity, and the scope of permitted third-party reuse.
The Apple News+ example is closer to a due-diligence question than a prediction of failure
At the launch of Apple News+ in 2019, reporting described a gap between expected and actual revenue. While Apple projected that publisher revenue would be 10 times higher than under the previous service, Texture, by the end of the first year, an anonymous publishing executive said about three months after launch that revenue at the time was one-twentieth of Apple’s stated level. Other publishers said it was similar to or lower than Texture.
That number is not a final first-year result. It was an early assessment by an anonymous source, and the sample, methodology, and subsequent performance were not disclosed. It should not be used as proof that “Apple paid only 5% of what it promised” or that the current Siri negotiations will also fail.
It is useful, instead, as a basis for operational due diligence. Five Apple News+ publishers interviewed within a month of launch described the burden of converting article formats and uneven support. Only some received private Slack support, and there were cases where template work was handed to outside vendors and required additional editing. For this proposal, ask before signing about content-conversion costs, the scope of technical support, support for smaller partners, and the cadence for sharing performance data.
Turn one proposal into a settlement-ready table
You do not need to install a separate program or create a platform account. Prepare the original proposal, a list of licensable content, baseline subscription, advertising, and partnership revenue, and a spreadsheet. Having contract/copyright staff and data staff review the same table helps identify gaps in both rights and numbers.
- Separate contract terms from reported information.
If multiyear deal discussions, use of current news, variable compensation, and a nine-figure budget under review are not written into the contract, place them in a “public reporting” column. Put signing status, rates, minimum guarantees, settlement cycle, territory, and language in an unconfirmed column. - Create a rights matrix.
Make rows for training and testing, indexing, retrieval and RAG, content display, summarization and answers, quotation of original text, link display, and model improvement, then mark each as permitted, prohibited, or requiring further negotiation. - Define the billable event in one sentence.
Choose which event generates payment among search, content access, incorporation into an answer, source display, and clicks to the original. Include treatment of failed responses, retries, cache reuse, and duplicate requests. - Check whether a single evaluation row supports a calculation.
The following figures are hypothetical inputs, not Apple’s actual terms. Enter content ID A-1042, 100,000 monthly searches, 12,000 answer incorporations, 3,000 source displays, 800 clicks to the original, a billable event of “answer incorporation,” and an example rate of $0.01 per event. The calculated payment is $120, but also ask whether the output records for the 12,000 events and the exclusion and adjustment details can be rechecked in the raw data. - Compare conservative, base, and optimistic scenarios.
Once you receive the actual proposal’s rates and usage figures, divide them into three ranges and subtract expected traffic declines, content-processing costs, and legal and data-operations costs from the payment. If a minimum guarantee is absent from the document, do not conclude it is “0”; mark it unconfirmed and propose renegotiation or withdrawal terms for low usage.
Review completion criteria: You must be able to recalculate payment using billable event counts and rates by content ID, and training rights must be separated from access rights for current answers. The first comparison is complete when actual contract terms are distinguished from hypothetical evaluation inputs, and dispute, renegotiation, and withdrawal terms for reporting errors or low usage are documented.
If you want to dig deeper
Best Practice on Generative and Agentic AI usage metrics This can help you design settlement questions by separating content activities such as searches, investigations, and requests. countermetrics.org
News/Media Alliance AI Principles Review permissions for training, display, and synthesis use, as well as transparency and source attribution, from a publisher perspective. newsmediaalliance.org
AI-Related Model Publishing Contract Clauses This offers a reference for contract structures that separate training rights from retrieval-augmented response rights. authorsguild.org
.png)
