Usage Reporting in AI Licensing Deals: Who Audits the Number?

In 2025, Folha de S.Paulo sued OpenAI over unauthorized use of its content. In May 2026, the two sides settled: a commercial licensing partnership, financial terms undisclosed. That sequence is close to the ceiling of what most publishers can achieve today. Litigation and brand power can force a negotiation — but only for publishers who can afford both, and even then, the number that changed hands stays private.

On July 1, 2026, Cloudflare proposed a different lever, aimed at everyone below that scale. It is evolving Pay Per Crawl into Pay Per Use: publishers get paid not when a bot fetches their content, but when that content creates value inside an AI answer. Starting September 15, 2026, crawlers that mix search and training will be blocked by default on ad-supported pages unless publishers change their settings. First partners: Ceramic.ai and You.com.

Microsoft's Publisher Content Marketplace made a similar promise months earlier: usage-based reporting, so publishers can see how their content was valued and where it may create more value in the future.

Different companies. Different mechanisms. Same unresolved question:

who audits the usage number?

In AI licensing, usage reporting is not just a dashboard. It can become the basis for payment, renewal, pricing, and leverage.

That is why usage reporting in AI licensing deals must be treated as a governance issue, not only as a product feature.

In short

Usage reporting in AI licensing deals is the mechanism by which a platform tells a publisher how its content was used and how that usage created value.

It is different from usage evidence.

A report tells the publisher what the platform counted. Evidence gives the publisher a record it can check, compare, audit, and use in negotiation.

That distinction now applies at two different levels. Microsoft's PCM is a single marketplace deciding what counts as a use inside one product. Cloudflare's Pay Per Use, launched July 1, 2026, sits a layer below: it proposes to become the metering and payment infrastructure for the open web itself, on a network Cloudflare says has already helped broker more than 50 major content licensing agreements in the past year. In both cases, the entity that counts the usage is also the entity that benefits from how it is counted.

Folha de S.Paulo's May 2026 settlement with OpenAI shows the alternative path publishers have relied on until now — litigation and negotiation — and its limit: the deal ended a lawsuit, but disclosed no usage number at all.

At Citations Logic, we use the term usage evidence to describe an independent proof layer: event-level records that help publishers verify what was used, when, under which rights, and in what commercial context.

The dashboard is not the proof. The count is the leverage.

Why Microsoft's Publisher Content Marketplace matters

PCM addresses a real market problem.

AI companies need high-quality, licensed content. Publishers need new revenue channels that do not depend only on traffic, clicks, or one-off licensing payments. Microsoft's official framing is clear: PCM is intended to create a transparent economic framework where publishers retain ownership and define licensing terms while AI builders access premium content for specific use cases.

The pilot is not theoretical. Microsoft says it has been co-designing PCM with major U.S. publishers including The Associated Press, Business Insider, Condé Nast, Hearst Magazines, People Inc., USA TODAY, and Vox Media. It started with Copilot grounding scenarios and is beginning to onboard demand partners including Yahoo.

That matters because it moves AI licensing from bespoke agreements toward marketplace infrastructure.

But infrastructure changes the publisher's risk.

A one-off deal can be negotiated case by case.

A marketplace needs standardization.

And once reporting standards become embedded, they become harder to renegotiate.

That is why AI usage evidence for publishers should not be treated as a later compliance feature. It belongs in the commercial architecture from the start.

Platform reporting is useful. It is not neutral evidence.

A platform-generated report can be valuable.

It can show trends. It can help publishers understand demand. It can create operational transparency. It can make usage-based licensing easier to administer.

But platform reporting has a structural limit.

The same environment that enables the use also reports the use.

That does not mean the report is wrong.

It means the report is not independent.

That is not a hypothetical concern. In June 2026, Cambridge University Press & Assessment hosted a workshop with COUNTER and NISO to define what "meaningful" AI use of content actually means — content incorporated into an answer, not merely retrieved. COUNTER's own updated best practice on AI usage metrics, in force since June 30, 2026, confirms the structural limit described above at the standards level: AI usage reporting is explicitly excluded from COUNTER's formal audit process, and is self-reported by the platform generating it. The industry's own standards body is saying, in writing, that a report is not proof.

For publishers, this distinction is not academic. It affects the ability to negotiate future value.

If a publisher wants to renew an AI licensing deal in eighteen months, the central question will not be whether the platform provided a dashboard.

The question will be whether the publisher can challenge, compare, and defend the number behind that dashboard.

That is the difference between proof of AI content usage and ordinary analytics.

Platform-reported usage vs. independent usage evidence

The two layers are not enemies.

The right structure is not platform reporting or independent evidence.

It is platform reporting plus independent evidence.

That is how a marketplace becomes more trustworthy over time.

Cloudflare's Pay Per Use points at the same problem, one layer down

Microsoft's PCM is one marketplace. Cloudflare sits underneath a much larger share of the web — the company says its network sits in front of more than 20% of all web domains.

On July 1, 2026, Cloudflare announced it is evolving Pay Per Crawl into Pay Per Use: instead of charging per fetch, publishers get paid when their content demonstrably contributes to an AI-generated answer. The company named two early partners with different mechanics. Ceramic.ai pays publishers each time their content appears in its AI search results. You.com lets an AI agent pay on demand for a specific piece of premium content it needs. Cloudflare provides the shared metering and payment layer beneath both.

The same announcement sets a deadline: starting September 15, 2026, crawlers that combine search and training — so-called mixed-purpose crawlers — will be blocked by default on pages that carry ads, unless the publisher changes the setting. Cloudflare also says more than half of AI crawler traffic today is spent re-fetching pages that have not changed, and that the largest search engine has roughly twice the content access of competing AI companies because it blends search and training in a single crawler.

Those are real, useful numbers. None of them answer the underlying question.

If a payment lands in a publisher's account because Cloudflare decided the content "created value" in an AI answer, the publisher is in the same position as a publisher reading Microsoft's dashboard: trusting a count produced by the same infrastructure that decides the price. A price tag is not proof of usage. It is a bill — and a bill is only as trustworthy as the ability to check it.

The objection here is fair: Pay Per Use is days old, with two named partners, and its methodology is not yet public at any level of detail. That is exactly the moment for publishers to ask for auditability — before "how Pay Per Use counts a use" hardens into a de facto industry standard nobody remembers agreeing to.

The objection: "The reporting model is still early"

This objection sounds reasonable.

It is also exactly why publishers should act now.

If the pricing, access, and compensation models are still being tested, then the definition of usage is still commercially alive. That is the moment to negotiate auditability.

Not later.

Later, the marketplace may have scaled. Standard terms may have hardened. Reporting categories may already be embedded into contracts, dashboards, and settlement logic.

At that point, the publisher is no longer negotiating the measurement system.

It is negotiating inside it.

That is a weaker position.

The window between "too early to ask" and "too late to change" is often shorter than publishers think.

What publishers should require before signing

Publishers do not need a theoretical debate.

They need contractual clarity.

Before entering a usage-based AI licensing deal, they should ask five questions.

First: what counts as a use?
Retrieval, grounding, citation, summarization, display, answer generation, and reuse are not the same thing.

Second: can usage be tied to a specific work, collection, rights package, or license term?
Aggregated numbers may help executives. They do not always help rights teams.

Third: can the underlying record be exported?
A dashboard that cannot leave the platform is not a negotiation file.

Fourth: what happens if the publisher disputes the count?
A mature licensing model needs reconciliation, not only reporting.

Fifth: will the evidence support renewal?
If the report cannot help the publisher price the next deal, it is incomplete.

This is where recurring AI licensing revenue is won or lost.

Not in the announcement.

At renewal.

Why the issue will spread beyond Microsoft

Microsoft is the visible case. It will not be the only one.

Publishers Weekly reported that Amazon was also exploring an AI content marketplace, while noting that publishers are beginning to push for usage-based fee structures that scale with how much content is consumed by AI systems. Cloudflare's Pay Per Use adds a third model to watch — not a single AI product, but a metering layer several AI products could plug into at once.

That trend creates a second problem.

If every AI platform — or every infrastructure layer — builds its own reporting loop, publishers will face several measurement universes:

one definition of usage here, another there;

one dashboard here, another there;

one valuation method here, another there.

That may work during pilots.

It will not work for a mature licensing market.

Publishers will need evidence that travels across platforms, deals, and renewal cycles.

That shared measurement layer is also what a credible AI content licensing benchmark will depend on. Without common usage units, deals cannot be compared — and a benchmark that cannot be compared is just someone else's number.

This is also why content provenance vs. usage evidence matters. Provenance can help establish where content came from. Usage evidence shows what happened to it inside an AI transaction.

Both are useful.

Only usage evidence supports the commercial count.

The renewal problem

Every AI licensing agreement eventually becomes a renewal conversation.

At signature, publishers negotiate from expected value.

At renewal, they negotiate from recorded value.

If the only record comes from the platform, the publisher starts the discussion from dependency.

Nine Entertainment's new agreement with Microsoft puts a date on this problem.

On July 3, 2026, Nine and Microsoft announced Microsoft's first publisher licensing deal of this kind in the Asia-Pacific region: Copilot can now ground answers in reporting from The Sydney Morning Herald, The Age, The Australian Financial Review and Nine's other mastheads. The financial terms are undisclosed. The term is one year.

A one-year pilot puts a renewal conversation on the calendar before the ink is dry. When that renewal comes, one side will hold usage data by default: Microsoft runs Copilot, so Microsoft counts what Copilot did with Nine's journalism. Nine will hold whatever record Microsoft chooses to share, unless it has agreed — or built — something independent before that date arrives.

That is not a criticism of this specific deal. Attribution and click-through commitments are a genuine, publisher-favorable feature of the agreement as announced. It is a structural observation about pilot-stage AI licensing generally: a twelve-month term compresses the distance between "sign now" and "prove it later" into a single calendar year, and the evidence question either gets answered in that window or gets negotiated from a position of dependency when the window closes.

The same dependency appears in AI display deals, where value is created inside the answer experience itself — and the platform holds the meter by default.

If the publisher has an independent evidence layer, the conversation changes.

It can ask:

which works were used repeatedly;

which content categories drove higher-value use cases;

which rights were activated most often;

which use types were underpriced;

which platform reports align with the publisher's own count;

which discrepancies deserve review.

That is not a compliance luxury.

It is negotiating power.

The future layer of AI licensing is proof

Microsoft's Publisher Content Marketplace and Cloudflare's Pay Per Use are both important signs that AI licensing is becoming more structured, scalable, and operational.

That should be welcomed.

But platform-generated reporting — whichever company runs the platform — should not become the only source of truth.

For publishers, the long-term question is not whether a marketplace or a metering layer provides reporting.

It is whether that reporting can be audited, compared, governed, and used to negotiate from strength.

That is where Citations Logic fits.

Not as a replacement for marketplaces.

As an independent proof layer that helps publishers turn AI content usage into evidence they can trust, use, and defend.

Because in AI licensing, the platform may provide the dashboard, or issue the payment.

But the publisher still needs the proof.

Continue the evidence chain

Read the broader framework: AI usage evidence for publishers

Understand the licensing shift: proof of AI content usage

Clarify the evidence category: content provenance vs. usage evidence

Ready to assess your AI usage evidence layer? Book an AI usage evidence assessment

Sources

Microsoft Advertising, "Building Toward a Sustainable Content Economy for the Agentic Web."

Digiday, "Q&A: Nikhil Kolar, vp Microsoft AI scales its 'click-to-sign' publisher AI content marketplace."

Publishers Weekly, "Microsoft, Amazon Plan AI Content Marketplaces."

Microsoft — "Nine and Microsoft announce Australian-first AI agreement for news media content," July 3, 2026
https://news.microsoft.com/source/asia/2026/07/03/nine-microsoft-copilot-agreement/

The Scholarly Kitchen, "Making AI Use of Scholarly Content Traceable, Measurable, and Trustworthy: A Meeting Report from Cambridge Scholarly AI Workshop," June 25, 2026
https://scholarlykitchen.sspnet.org/2026/06/25/making-ai-use-of-scholarly-content-traceable-measurable-and-trustworthy-a-meeting-report-from-cambridge-scholarly-ai-workshop/

COUNTER Metrics, "Best Practice on Generative and Agentic AI Usage Metrics" (updated June 30, 2026)
https://www.countermetrics.org/code-of-practice/best-practice/bp-ai/

Cloudflare, "Your site, your rules: new AI traffic options for all customers," July 1, 2026
https://blog.cloudflare.com/content-independence-day-ai-options/

Cloudflare, "Cloudflare Allows the Agentic Internet to Flourish with a Simple Philosophy: Your Content, Your Rules," press release, July 1, 2026
https://www.cloudflare.com/press/press-releases/2026/cloudflare-allows-the-agentic-internet-to-flourish-with-a-simple-philosophy-your-content-your-rules/

OpenAI, "OpenAI, Grupo Folha, and Grupo UOL announce strategic content partnership," May 25, 2026
https://openai.com/index/grupo-folha-grupo-uol-partnership/