L-05 · Measurement

Citation persistence in AI answers

Primary intent
Measure citation recurrence without equating persistence with source quality or ranking causation.
Evidence state
Source-grounded reference
Review owner
Matthias Ramahi · independent review not claimed
Last reviewed
2026-08-22
Direct answer

Citation persistence

Citation persistence is the recurrence of a normalized visible source for the same defined observation unit across repeated runs or windows. It is measured against eligible opportunities and must distinguish URL-level persistence, domain-level persistence and source-position changes.

Use this measure when the research question is whether the same visible sources continue to appear, not merely how many unique sources exist.

Define an eligible opportunity

The most defensible unit is usually a stable question observed on a declared surface, locale and route in a defined window. A source has an opportunity to persist only when both comparison observations are eligible. Missing answers and interfaces without visible citations need explicit treatment.

Do not silently remove failed or uncited observations. Report the eligible pair count, the missing pair count and the rule used for each.

Preserve visible and normalized identities

Store the URL exactly as shown, then derive a normalized URL and registered domain. This permits several views: exact document recurrence, canonical-page recurrence and publisher recurrence. Each view answers a different question.

Redirect resolution can change over time and may require network access. Record when and how it was resolved. Never rewrite the original captured value.

Report recurrence as a set of measures

For each control question, report intersection and union across windows, then aggregate with a declared weighting rule. A question with ten source slots should not silently dominate a question with one unless that weighting is intentional.

Separate persistence from position. A source may recur while moving from a prominent citation to an expandable list. If visibility tiers matter, preregister them and capture the interface evidence.

  • Exact-URL persistence rate.
  • Normalized-page persistence rate.
  • Registered-domain persistence rate.
  • New, returning and disappeared source counts.
  • Visibility-tier or position change, when reliably observable.

Persistence is not endorsement or quality

A persistent citation is a recurring visible source in the measured outputs. It is not proof that the source caused the answer, is correct, is preferred globally, or will appear for another user. Those claims require different evidence.

Persistence becomes useful when paired with source relevance, question class and comparability notes. Alone, it is a descriptive recurrence measure.

S

Source notes

These sources support the definitions, standards or project boundaries named in this reference. They do not prove that a public observation dataset exists.

  1. portfolio-dossier
    Canonical ai-fanout.com domain dossier

    Confirmed ownership, accepted public Evidence Lab purpose, named Research Owner, indexable website launch and separately gated provider research.

    Owner record
  2. w3c-prov-o
    PROV-O: The PROV Ontology

    Provides provenance concepts for entities, activities, agents, derivations, sources and versions.

    Open
  3. rfc-3339
    RFC 3339: Date and Time on the Internet

    Supports an interoperable timestamp representation tied to UTC.

    Open
  4. fair-principles
    The FAIR Data Principles

    Supports reusable research data with metadata, provenance and clear usage licenses.

    Open