Skip to content
All guides

Reporting

Provenance, attribution and measuring AI usage

What an AI answer should carry with it, and how publishers and libraries can count AI usage, including the COUNTER Release 5.1 approach.

Updated 9 November 2026 7 min readOlive Turn

An AI answer is only as trustworthy as what it can show about its sources. For scholarly and professional content, three things should travel with every answer: where the content came from, who it belongs to, and a record that it was used.

Provenance: where did this copy come from?

Provenance is the history of a copy: which service delivered it, from which host, when, and which version it is. For research, version matters: an accepted manuscript, the version of record and a corrected version can differ. A retraction notice, where the source records one, should reach the AI as clearly as the article text.

A missing retraction notice is not proof that an article is sound. It only means none was recorded by the source.

Attribution: can a reader check the claim?

Attribution means the answer points back to the passage, document and collection it relied on. A citation that names the exact passage lets a reader check the claim in seconds, and keeps the rights holder visible wherever the answer travels.

Measuring AI usage

Usage numbers for AI need the same care as any other usage statistics. Useful measures include the number of interactions, the searches made, how often a document was surfaced in results, how often it was actually retrieved, and how many requests were refused because no licence covered them.

Libraries compare usage across publishers with COUNTER, the industry code of practice for usage reports. COUNTER Release 5.1 reports usage by platform, database, title and item. Metrics for AI use are emerging as best practice, with AI usage reported under the access method “Agent”, and are not yet part of the audited standard.

How our products report it

  • Razeta records every interaction by AI tool, organization, collection and outcome, and produces reports in the COUNTER Release 5.1 format, which libraries can harvest automatically through SUSHI.
  • Research Connect produces COUNTER Release 5.1.1 reports of AI usage (platform, title and item) for libraries and publishers.
  • Neither product has yet been audited or listed by COUNTER, and both say so in their reports.