Usage reporting: what AI did with your content
The gap after the fetch
A reader asks a chatbot a question. The answer draws on your story and adds a link the reader never clicks. Your logs show that a bot fetched the page. They show nothing about the answer. Provenance tells you whose content the AI read and whether it changed. Usage reporting tells you what the AI did with it.
The Content Telemetry Standard
The SPUR Coalition, a nonprofit group of news publishers, launched in February 2026. Its founding members include the BBC, the Financial Times, the Guardian, Sky News, the Associated Press, the Telegraph and Mediahuis. SPUR maintains the Content Telemetry Standard (CTS), which began as OpenAttribution's specification. Version 1.0 was published on 2 September 2026 under the Apache 2.0 license. It gives AI systems one vocabulary for five events:
- Retrieved: a bot fetched the content.
- Grounded: the content entered the context the AI used to write an answer.
- Cited: the output named the source.
- Presented: the content or a reference to it was shown or read aloud.
- Engaged: a user or agent acted on it, such as with a click.
Who reports what
You can see retrievals at your own CDN or server, and a destination site can confirm a click-out. Only the AI system can report grounding, citation and presentation. A shared Content-Telemetry-ID header lets your log and the AI firm's report be matched to the same fetch, though redirects can drop it.
Every AI report is a claim by the firm that may owe you money. In version 1.0, signing reports is optional. A missing event proves nothing unless the firm declares that it reports every qualifying use, and that declaration is also a claim. Ask for audit rights.
Counts, credit and value
CTS counts events. It does not judge how much a story shaped an answer. Attribution models do that: they turn counts into a value for each story, often to split a payment. Credit is different again. Credit means the answer names your story or links to it. Content that was grounded but not cited was used without credit. The contract decides what to pay for: groundings, citations or answers.
Where C2PA labels fit
A web address can change when a story is syndicated, cached or resold. CTS lets a report carry a stable content ID instead, and it accepts a C2PA manifest hash as one. An AI system can also report that it detected a provenance signal in the content it used, though that report is still a claim.
Put your own story ID in each label and register your ID prefix with the service that collects reports. Reports that carry the ID can then reach you without a web address. Encypher labels also carry your license terms and a usage-reporting request. A label cannot enforce either one. A license contract can.
What reports can and cannot show
Can
- Which content an AI system says it fetched, used, cited and showed.
- How often your work went into answers without credit.
- A shared format, so reports from different firms can be added up.
Cannot
- Prove that a firm reported every use.
- Measure how much your work changed an answer.
- Decide what anyone owes.
- Report model training, which is outside the standard's scope.
What to do now
- 1
Sign content when it goes live. Put your story ID, your license terms and a reporting request in each label, then check that a published copy still has its label.
- 2
Count AI fetches yourself. Your CDN and server logs give you a number to check reports against. SPUR's free code for Cloudflare and Fastly sends fetch reports to OpenAttribution's hosted service.
- 3
Put reporting and audit rights in every AI deal. Name the events, where they go, and your right to audit. An RSL license can require CTS reports, and IAB CoMP can tell AI systems where to send them.