AI content telemetry tracks what happens when AI systems retrieve, use, cite and present publisher content. We explain SPUR's five-stage standard, its limits and why traditional web analytics may no longer tell the whole story.
AI Is Creating A Measurement Gap After The Click Disappears
AI content telemetry is an emerging way of reporting what happens to online content after an AI system retrieves it. The idea is to go beyond the basic server request and show whether content entered an AI system's generation context, whether it was cited, whether that citation or content actually reached the user and whether the user interacted with it afterwards.
The formal standard currently pushing this idea forward is called Content Telemetry, developed through the SPUR Coalition. We use the phrase AI content telemetry as a clear descriptive framing for the wider concept rather than suggesting that it is already a universally standardised industry term.
The problem it addresses is increasingly obvious. Publishers can often tell when a crawler, agent or other automated system requests a page, but traditional analytics usually lose sight of what happens next. An AI system might retrieve an article, use it to help construct an answer, mention the source, display a citation or satisfy the user's query without sending that person to the publisher's website.
That means the classic analytics model is beginning to stop too early. The publisher sees the machine arrive, but much of the meaningful activity may happen somewhere else.
Traditional Web Analytics Were Built Around The Website Visit
Most web analytics systems make sense because the important activity traditionally happens on the publisher's own property. A person arrives through search, social media, a newsletter or a direct visit, loads the page and begins producing measurable signals.
The publisher can then examine pageviews, referral sources, session duration, conversions, scroll behaviour and other forms of engagement. Those measurements are imperfect, but they still describe an interaction taking place on infrastructure the publisher can observe.
AI systems can break that relationship.
A user may ask an AI assistant a question, receive a detailed response assembled from multiple sources and never visit any of those sources directly. The publisher might see that its content was fetched by an automated system, but the request alone cannot reveal whether the article actually contributed to the answer, whether it received attribution or whether the user ever saw its name.
That missing middle is where content telemetry enters the picture.
SPUR's Content Telemetry Standard Follows Content Through Five Stages
SPUR's current model divides AI content usage into five stages: Retrieved, Grounded, Cited, Presented and Engaged. Together, those stages attempt to describe what happens between an AI system accessing content and a user eventually doing something with the resulting answer.
Retrieved means the AI system fetched the content. Grounded means that content was placed into the context being used to generate a response. Cited means the source was explicitly referenced in the generated output, while Presented means that citation or content actually became visible or otherwise perceivable to the recipient. Engaged describes a later user action such as clicking, copying, sharing or directing an agent to act on what was presented.
The important part is that these are separate events rather than one generic label saying that an AI system "used" a page. Retrieval alone does not prove that the material affected an answer, while grounding does not necessarily mean the publisher was cited. A citation may exist in an output without becoming visible to the user, and a visible citation may still generate no click.
That separation gives publishers a more precise language for describing AI-mediated use.
The Missing Stages May Tell Publishers More Than The Completed Ones
Content telemetry becomes particularly interesting when an interaction stops somewhere in the middle.
A publisher could theoretically see that an article was retrieved but never grounded, showing that the AI system fetched the material without using it in the generation context for that interaction. Another article might be grounded and contribute to the answer while never receiving an explicit citation.
A source could also be cited internally but never presented to the user, depending on how the product handles references. Even when a source is visibly presented, the user may read the answer and move on without clicking through to the original publication.
Those outcomes have very different implications for attribution and value, yet ordinary server analytics may collapse them into little more than a bot request.
This is why content telemetry is more than another traffic counter. It tries to describe the role content plays after the traditional measurement trail has already gone cold.
AI Content Telemetry Is Not A Training Audit
There is an important limit to what the current Content Telemetry standard claims to measure.
Its main focus is inference-time use: what happens when an AI system accesses content while producing an answer or carrying out an interaction. It is not designed as a universal record of every way an AI company might have used a publisher's work.
Activities such as building training datasets, training or fine-tuning models, creating embeddings and constructing retrieval indexes sit outside the main scope of the standard. That distinction should prevent content telemetry from being mistaken for an all-seeing audit trail of everything that has ever happened to an article inside an AI company.
The purpose is narrower and more practical. It is trying to make specific, observable stages of an AI interaction reportable in a common way.
Zero-Click Publishing Makes The Measurement Problem Harder
Publishers were already dealing with reduced visibility before generative AI became a mainstream discovery layer. Traditional zero-click search demonstrated that a platform could answer a user's question directly without producing a visit to the original source.
AI systems deepen that problem because they can synthesise information from multiple sources into an entirely new response.
A publisher's work may contribute to an answer without generating a pageview, but that does not mean the contribution had no value. The article may have supplied factual grounding, influenced the response or provided a source that appeared alongside the generated answer.
Traffic still remains commercially important. A reader who reaches the original site can subscribe, buy something, explore other work, view advertising or build a direct relationship with the publication. Content telemetry should not be used to pretend that lost referral traffic suddenly becomes irrelevant.
The point is that referral traffic may no longer describe the entire interaction between publisher content and AI discovery.
Measurement Could Become Part Of AI Licensing Without Becoming The Licence
SPUR's Content Telemetry standard is designed as reporting infrastructure rather than a licensing system.
It does not decide whether an AI company has permission to access an article, determine how much that use should cost or create an automatic right to compensation. Those questions still depend on separate legal, contractual and commercial arrangements.
The specification can, however, connect usage events to an external licensing reference. That creates the possibility of linking permission with evidence about what happened afterwards without forcing the telemetry standard itself to become the contract.
A publisher and AI company could therefore agree separately on the terms of access while using telemetry to provide a clearer record of relevant usage. Over time, that kind of infrastructure could support licensing reports, audits, attribution systems, commercial dashboards and more sophisticated payment models.
The important boundary is that telemetry supplies information. It does not create the commercial terms by itself.
The Biggest Problem Is That The AI System Still Has To Report What It Did
Content telemetry sounds powerful until the obvious question appears: who verifies the deeper events?
Retrieval is comparatively observable because the publisher's own infrastructure may record the incoming request. If an AI system also reports that retrieval using a shared identifier, the two sides can potentially correlate the event.
Grounding, citation, presentation and engagement are harder because much of that activity occurs inside the AI service or its interface. Under the current model, those later stages rely heavily on the agent or platform reporting them.
That creates a genuine trust problem. An event can be cryptographically signed in the future to help establish which party sent it, but proving who sent a claim is not the same as proving that the claim is complete or accurate. A system could theoretically report genuine events while still failing to report every qualifying use.
SPUR acknowledges that stronger verification and auditing remain areas for further development. That is not a small footnote; it is one of the central challenges the standard will need to solve if telemetry becomes commercially significant.
Content telemetry is therefore infrastructure for greater transparency, not an automatic window into every AI model.
Publisher Adoption Gives The Idea More Weight Than A Technical Experiment
SPUR published version 1.0 of the Content Telemetry standard in October 2026, moving the proposal beyond an early concept into a defined technical framework.
The publisher coalition around it has also expanded. More than 40 additional organisations, including The New York Times, Agence France-Presse and The Economist, joined SPUR in October, bringing the coalition to more than 80 members across more than 20 countries.
That level of participation gives the issue considerably more weight. Publishers are not simply debating whether AI systems should access their work; they are beginning to ask for common infrastructure that can describe what happens after access occurs.
It still does not mean the standard has universal AI-platform adoption. Publisher enthusiasm and AI-company implementation are different milestones, and the latter will determine whether Content Telemetry becomes part of real industry infrastructure rather than an agreed specification waiting for everyone else to arrive.
The Publisher-AI Conversation Is Moving Beyond Whether Crawlers Get Through The Door
For several years, one of the loudest questions around AI publishing has been whether bots should be allowed to crawl content at all.
That remains relevant. Publishers increasingly have controls that influence whether their material can participate in AI-powered products, including systems such as Google's AI Search controls for publishers.
But access is only one part of the relationship.
Once an AI system has permission to retrieve something, the next questions become more complicated. Publishers may want to know whether the retrieved material actually contributed to an answer, whether they were attributed, whether that attribution reached the user and whether any measurable interaction followed.
If commercial agreements exist, another question appears: whether the reported usage aligns with whatever access, attribution or licensing terms the parties agreed.
That is a much more mature infrastructure conversation than treating the future of AI publishing as a binary choice between blocking everything and allowing everything.
Publisher Analytics May Eventually Need To Measure Machines As Well As People
The traditional analytics era trained publishers to measure audiences. The next phase may require them to measure intermediaries as well.
Publishers will still want to know whether people visit, read, subscribe, buy and return. Those direct relationships remain extremely valuable. But AI systems increasingly stand between source material and the person seeking information, and that intermediary layer can shape what people learn without always producing a conventional visit.
Content telemetry attempts to make part of that process visible.
It could also add another layer to questions around brand visibility in AI Search. Knowing that a brand or publisher appears in an AI answer is useful, but a mature measurement system could go further by showing what material entered the interaction, whether attribution appeared and what happened once that information reached the user.
SPUR's current standard may or may not become the final industry answer. Technical standards evolve, competing approaches can emerge and adoption can take years.
The underlying problem is more durable: publishers need better ways to understand the contribution of their work when AI systems increasingly mediate the relationship between content and audience.
Tanizzle Says: The Web Needs Better Receipts
AI-powered search and assistants can be genuinely useful. We do not believe every answer should force a pointless visit to a website merely to preserve an old measurement model.
But an information economy cannot mature if one side can observe almost everything happening inside the interaction while the source publisher sees little more than a crawler request.
Better infrastructure should make permission, usage, attribution and commercial relationships easier to understand without turning AI innovation into an impossible compliance exercise.
Content telemetry is an early attempt to build that infrastructure. Its biggest challenge will be proving that the reports are trustworthy enough to become commercially useful.
The next generation of publisher analytics may therefore need to answer a question that pageviews were never designed for: what happened when our content created value somewhere else?
From Tanizzle: For You
AI content telemetry becomes easier to understand when placed inside the wider shift from links towards generated answers. Our look at AI Search and the changing publisher relationship explores what happens when discovery platforms increasingly answer questions instead of simply directing people elsewhere.
Publishers are also receiving more controls over whether their work participates in those systems. We explain the current direction in Google's AI Search controls for publishers, including the growing separation between ordinary web indexing and AI-mediated use.
Measurement then adds another layer. Our guide to brand visibility in AI Search looks at how publishers and brands become visible inside AI discovery, while content telemetry asks what happened to the source material behind that visibility.
Tanizzle FAQs: AI Content Telemetry
What is AI content telemetry?
AI content telemetry is the reporting of how content is used during an AI interaction after it has been retrieved. It can describe whether the material entered an AI system's generation context, was cited, was presented to the user and generated an engagement event.
What is the SPUR Content Telemetry Standard?
The SPUR Content Telemetry Standard is an open technical specification for reporting AI content usage. Its current model separates usage into retrieval, grounding, citation, presentation and engagement so publishers and AI systems can describe different stages of an interaction more precisely.
What are the five stages of Content Telemetry?
The five stages are Retrieved, Grounded, Cited, Presented and Engaged. They describe a progression from an AI system fetching content through to that content or its source potentially reaching and being acted upon by a user.
What does grounded mean in AI content telemetry?
Grounded means the content was placed into the AI system's generation context for the interaction. This is different from retrieval because an AI system can fetch a page without ultimately using that page as part of the context for its response.
Is grounding the same as citation?
No. Content can be used as grounding without receiving an explicit citation in the generated response. The standard therefore treats grounding and citation as separate events.
What does presented mean in Content Telemetry?
Presented means the content or source reference became perceivable to the recipient through the user-facing experience. This is separate from citation because a citation can exist within an output without necessarily becoming visible to the eventual user.
What does engaged mean in Content Telemetry?
Engaged describes a user action involving what was presented. Depending on the experience, that could include clicking a source, copying material, sharing it or instructing an agent to take a related action.
Can content telemetry show when AI uses an article without sending traffic?
Potentially, if the AI system implements the telemetry standard. Grounding, citation and presentation events could reveal forms of content use that occur even when the user never clicks through to the publisher's website.
Does Content Telemetry track AI model training?
No. The current standard focuses on inference-time interactions. Activities such as model training, fine-tuning, embedding creation and construction of retrieval indexes sit outside its main scope.
Does Content Telemetry make AI companies pay publishers?
No. Content Telemetry is a reporting system rather than a licensing or payment system. The information it produces could support commercial agreements, but permission, compensation and pricing have to be established separately.
Can publishers independently verify AI content telemetry?
Only partly under the current model. Retrieval can sometimes be compared with records held by the publisher, while later events such as grounding, citation, presentation and engagement depend more heavily on reporting from the AI system. Stronger auditing and verification remain unresolved parts of the wider implementation challenge.
Has every major AI platform adopted Content Telemetry?
No. SPUR is working on implementation and industry adoption, but the existence of the standard and growing publisher participation should not be confused with universal deployment across AI platforms.
Why could AI content telemetry become useful to publishers?
It could give publishers information about forms of AI-mediated content use that pageviews and referral analytics cannot capture. As more answers are generated without a traditional website visit, that additional measurement layer could become useful for attribution, analytics and future commercial relationships.