Updates on our Generative and Agentic AI Best Practice (phase 2)
18 September 2026When we published our Best Practice on Generative and Agentic AI usage metrics in June, we knew it wasn’t finished. Since then, our AI Working Group has been tackling phase two of the project. Phase two focuses on off-platform federated usage and third-party AI tools, with some likely amendments to cover ‘holistic’ usage metrics too.
We’re aiming to release phase two in October 2026. In the meantime, this post provides an update on our recent discussions.
What we’re tackling
Phase one established initial guidance, including the new Access Method: Agent and new AI-specific metrics. Community feedback has been positive. Librarians are happy that we have aligned AI usage metrics with traditional metrics to clearly demonstrate changing usage patterns and budget implications. Publishers and technology providers are actively building out the new metrics.
AI developers told us that some of our choices were too closely tied to specific technologies that are already becoming outdated, such as Chunks. We’re working on updating the guidelines to be more technology-agnostic as part of phase two.
The other aspect of phase two is on bringing in reporting about inference-time usage – tracking content when an AI system retrieves, grounds, or cites publisher material in response to user prompts – outside of publishers’ own platforms. By building on established COUNTER frameworks like our Best Practice on Syndicated Usage, we’re going to be able to deliver timely updates for AI reporting without major changes to the Code or the original phase one guidance.
Syndicated usage
The SPUR telemetry framework
Let’s be real: scholarly publishing is a big industry, but it’s not big enough to hold major AI companies to a small, niche standard like COUNTER. We need to leverage work being done by bigger industries, where they align with our needs. One of the ways we’re doing that is by picking up on the work of the SPUR Coalition. SPUR was born out of a group of news publishers who needed to standardise reporting requirements for AI content usage. It defines a clear telemetry framework for tracking usage across the AI workflow:
- Retrieval: Content is fetched by an AI tool or agent (e.g., via Model Context Protocol / MCP).
- Grounding: Content is loaded directly into the AI agent’s generation context.
- Citation: Content is explicitly referenced or cited in the response (mapping directly to COUNTER’s Investigation and Request concepts).
- Presentation: Content is rendered or presented to the end user.
- Engagement: The user clicks through to the source material on the publisher’s platform.
As of today, we’re planning to reference SPUR telemetry in our phase two guidelines. We’ll be mapping citation and presentation telemetry events to the existing AI COUNTER metrics. We are also going to use retrieval telemetry events for a new holistic “AI retrievals” metric.
To be clear: mapping SPUR telemetry events to our metrics represents a pragmatic approach to collecting consistent usage metrics across different AI tools and services, without having to force them to adopt COUNTER reporting. It is not a formal endorsement of SPUR as an organisation, nor an encouragement to our members to join the SPUR Coalition.
What about MERU?
MERU is Google Scholar’s ‘Mediated Reading’ initiative. It is a form of syndicated usage, and we’ve mapped how MERU messages from Scholar can and should be reported through our AI usage metrics.
Holistic metrics
A challenge with syndicated usage is that it doesn’t usually say what’s been included in an AI response to a user prompt. That means our original requirement – that only materials included in a response can be included as usage in a COUNTER report – will necessarily result in incomplete metrics. To help resolve that, we’re thinking about introducing a new “AI retrievals” metric. This idea has come up a few times through phases one and two, but we only came up with a potential solution this week.
(Side note from Tasha: I don’t care if you think it’s AI, I’ve been using N-dashes for 30 years. You can pry them from my cold, dead keyboard.)
Next steps
We’re working on finalising the phase two best practice. Tasha’s drafted a track-changed version of the AI and Syndicated Usage guidelines for the AI working group to discuss. Once the final questions are resolved we’ll go to the Code Team for a technical check and then to the Executive Committee for approval. Tasha’s goal is to publish before the Frankfurt Book Fair. This is ambitious, given we’re trying to build standards on shifting sands, so please give us grace if we’re a week or so late, and stay tuned for further news.
Thank you to the AI Working Group!
We are very grateful for the enthusiasm and engagement of the AI working group members. Thank you to: Senol Akay, Katie Arabie, Jon Blackburn, Sara Crowley-Signeau, Ed Hallet, Kirsten Fuoti, Rikki Gracia, Sue Hodgson, Athena Hoeppner, Stuart Maxwell, Adam Pulford, Michael Sisolak, Heather Staines, Michelle Urberg, Lisa Walton, Monica Westin, Peter White, Sam Wotring, Lisa Wylie, and Nico Zazworka.