Social Rocket
The publisher alliance SPUR published version 2 of an open standard on 2026 1.0 October, designed to let AI services report the use of third-party content in a traceable way. The Content Telemetry Standard distinguishes between retrieving a web page, its actual use in an AI response, a source attribution, its presentation to the user, and a subsequent interaction.
For editorial teams and operators of high-quality websites, this solves a fundamental measurement problem – at least on paper. A server log shows that a bot loaded a URL. It does not show whether the content then influenced an answer, whether the source was visibly named, or whether anyone clicked the link. SPUR wants to document these steps with standardised events.
The standard is publicly available. Whether and when the major AI providers will adopt it, however, remains open. A published specification alone does not yet give publishers any new usage data.
Five events instead of a single bot request
The official technical repository describes an event chain:
| Event | What it is meant to capture | What classic server logs reveal about it | |---|---|---| | Retrieved | Content is fetched over HTTP | Often visible, provided the request reaches your own server | | Grounded | Content is loaded into the context of an AI system and used as a basis | Usually not visible | | Cited | The system explicitly names the content as a source | Usually not visible | | Presented | Text, content or a source reference appears on the user interface | Usually not visible | | Engaged | The user clicks, copies, shares or lets the agent act | At most partly measurable via referral and web analytics |
The distinction matters. An article can reach the context of an answer without being cited. A source can be named without a user ever seeing it. And a visible source attribution does not automatically generate a visit.
For such events, the specification provides shared session and reference information. Cached content should also be reportable as usage when it is reused in an AI session. The protocol explicitly measures observable transitions, not the internal reasoning or ranking mechanisms of a model.
Why characters matter more than AI tokens
To describe quantities, version 1.0 uses, among other fields, `chars_ingested` and `excerpt_chars`. This allows the amount of text processed or cited to be expressed in characters.
Tokens remain possible as supplementary information, but they differ depending on the model and tokenizer. A character count can therefore be more practical when comparing different AI providers. However, it says nothing about the economic significance of a use: a short, exclusive report can be worth more than a long standard text.
The specification also covers images, video and audio. SPUR further describes points of connection to provenance standards such as C2PA. For social media publishers and creators, this extensibility may become interesting, even though the current alliance primarily represents journalistic rights holders.
BBC, Financial Times, Guardian and AP want uniform rules
Behind SPUR are, among others, BBC, Financial Times, Guardian Media Group, Sky News, Telegraph Media Group and Associated Press. On the official website, the organisation, founded in spring 2026, describes itself as a non-profit alliance for developing shared technical and business standards.
SPUR does not negotiate blanket licences for all members itself. The rights holders remain in control of their respective contracts. Standardised telemetry data could nevertheless help place different licensing models on a comparable data basis.
The alliance has also announced an AI Licensing Advisory Board. According to Digiday from 2 October, OpenAI, Anthropic, Google, Meta and Microsoft were invited. This does not mean these companies have committed to becoming members or implementing the standard. Google confirmed to Digiday only a general exchange with SPUR and other associations; at the time of publication, there were no corresponding commitments from the other companies.
A technical specification is not yet an independent audit
The central open question is who reports that an AI system has actually used a piece of content, and who checks that report?
The standard is designed so that events from different observers – for example a CDN, the origin server and an AI agent – can be matched using shared identifiers. However, an origin server still does not automatically see what happens inside a closed AI system. If the provider does not report internal events, blind spots remain.
According to Digiday, a technical committee of SPUR is to address verifiability and auditing on 15 October 2026. Such control mechanisms will be decisive in whether the data can later serve as a basis for licence billing.
Data protection, anonymisation and the question of which information about individual user sessions a publisher may receive also belong in practical implementation. The protocol must not be confused with a blanket authorisation to pass on personal data.
What German publishers and marketers can concretely derive from this
The development affects not only publishers with paywalls. Trade media, blogs, comparison portals and brand publishers now also need to distinguish whether content is crawled by search engines, used by AI assistants as a knowledge base, or linked in answers and actually clicked.
This is something different from the established SEO metrics of impressions, rankings and organic sessions. SPUR describes a possible additional measurement layer between content production and AI output. This does not yet yield a new reach KPI as long as the AI providers do not come along.
Remuneration is also unresolved. Our report on Google's tests with pay-per-use for publishers covers a concrete commercial model. The SPUR specification, by contrast, initially provides an open technical vocabulary to describe usage consistently in the first place.
Legal conflicts also arise: US publishers are demanding rules against covert AI crawlers, while the New York Times has formulated criteria for AI licensing partnerships. Technical measurement can support these questions, but it replaces neither consent nor licence rights.
Conclusion: measurability is the first step, not payment
Content Telemetry 1.0 is a concrete, freely accessible protocol proposal with events, a data schema and conformance tests. This distinguishes it from general demands for fair AI remuneration.
Proof of actual use nevertheless remains dependent on implementation and verifiability on the part of the AI services. Whether this ultimately leads to reliable content licences has not yet been decided as of October 2026.