In short. Primary source attribution combines checking support for a factual statement with tracing the evidence to its original producer.
Definition: Primary source attribution traces a factual statement to original evidence and identifies the passage supporting it. A primary source is material produced by someone with direct experience and knowledge of the subject.
Key facts
- The ALCE paper evaluates answer correctness and citation quality separately.
- W3C's PROV-O recommendation defines a primary source through its producer's direct knowledge of the topic.
- As of October 2026, Google's helpful content guidance asks publishers to assess original information and clear sourcing.
How does ALCE assess attribution?
The paper Enabling Large Language Models to Generate Text with Citations, published in December 2023, introduces ALCE, a benchmark for evaluating answers with citations. Its framework distinguishes fluency, answer correctness and citation quality.
Citation recall measures whether a statement has complete support from its cited passages, considered together. ALCE averages this assessment across statements. Citation precision checks individual citations for relevance; the framework allows redundant supporting references.
This supplies a useful starting point for editorial review: compare each factual statement with the passages attached to it. Tracing the evidence's origin adds a separate provenance check. Provenance means the history of how information is produced or derived.
What makes evidence a primary source?
W3C's PROV-O: The PROV Ontology describes primary evidence as material produced through direct experience at the time of the subject's study. Its examples include a sensor reading and a journal written during a historical event. A secondary source derives material from an earlier source; the recommendation explains that referencing the primary material allows its reliability to be investigated.
PROV-O expresses this relationship with prov:hadPrimarySource. The following simplified Turtle example, a text syntax for recording relationships, uses identifiers from the recommendation's example:
@prefix prov: <http://www.w3.org/ns/prov#> .
@prefix : <http://example.com/> .
:post9821v1 prov:hadPrimarySource :crimeData .
In the source's example, the post derives from crime data gathered by a government. The relationship records that origin. Reviewing whether a particular sentence follows from the data remains a separate claim-level task.
How can a website review its factual statements?
Google's helpful content guidance asks whether content supplies original research or analysis, presents clear sourcing and contains easily verified factual errors. Applied alongside ALCE's passage checks and PROV-O's source relationships, these questions suggest an editorial workflow:
- Identify the factual statement and open its cited source. Locate the passage that supports the whole statement, including qualifications.
- Follow references to the original evidence. Record which document supplies the evidence and which document supplies an interpretation.
- Place a descriptive source link beside the relevant statement. Retain the supporting passage in the editorial record so that a reviewer can repeat the comparison.
This is an application of the cited frameworks to website editing. Google's guidance also asks whether material drawing on other sources adds substantial value and originality; attribution accompanies the publisher's own explanation or analysis.
Related terms
FAQ
Can you cite a secondary explanation?
Yes. W3C's PROV-O describes secondary materials deriving from primary sources. When you use an explanation's interpretation, identify that explanation and follow its references for the underlying evidence.
Does a relevant citation cover every claim in a sentence?
ALCE's citation recall requires the cited passages, together, to support the complete statement. If your sentence combines claims, check each part against those passages before retaining the sentence.
How should you handle several supporting sources?
ALCE's citation precision permits redundant supporting citations. You can retain several references when each contributes relevant support, while documenting the original evidence behind them.
Sources
- Association for Computational Linguistics, "Enabling Large Language Models to Generate Text with Citations", December 2023
- W3C, "PROV-O: The PROV Ontology", 30 April 2013
- Google Search Central, "Creating helpful, reliable, people-first content", updated 5 October 2026
Check how a page reads to answer engines. Open the GEO score analyzer