REVIEW
How Generative AI Redefines Personalization and Diversity in the Newsroom
Computers and Society · Submitted 5 Nov 2026 · 97 references
Pith's verdict: Two-way personalization is a real conceptual contribution and the interview work is solid, but the quantitative validation of within-story consistency is too fragile to carry the weight the paper places on it.
T0 · AI review
Why Pith says this
Highlight any text to comment on it.
Referee report
The paper
The paper claims
Generative AI moves news-diversity decisions where editors cannot see them: into research tools and into the many audience versions of one story.
The reading
What carries the argument
The argument is carried by three linked concepts plus one interpretive lens. 'Two-way personalization' distinguishes tool personalization—practitioners configuring general-purpose AI through model choice, prompts, supplied materials, and routines—from audience personalization, the adaptation of news for different readers. 'AI-mediated source horizon' names the four-layer span of sources that exist, that a system can retrieve, that it presents, and that the journalist selects; this is where pre-publication diversity decisions become invisible. 'Within-story informational consistency' names the preservation of core claims, evidential qualifications, and context across versions, measured separately from survival of the main idea. The interpretive lens is professional jurisdiction and boundary work: journalists retain authority over meaning while delegating form, and that division is what makes them formally responsible for consequences they cannot observe.
Extended reading notes
Core claim
The paper's central claim is that generative AI relocates diversity-relevant decisions in journalism to two sites where conventional evaluation and oversight cannot observe them. Before publication, individually configured AI tools define what the authors call an 'AI-mediated source horizon'—the span of sources that exist, can be retrieved, get presented, and are ultimately selected—and whatever falls outside that horizon leaves no trace in the published report. After publication, generative adaptation can produce multiple audience-specific versions of the same story that preserve its main idea while retaining, on average, only about three-quarters of the information essential to understanding it, with different audiences receiving different subsets of that information. In the most divergent case, versions generated for two audiences had a mean Jaccard distance of 0.323, meaning roughly a third of the essential units retained by each version were not shared with the other. Because practitioners mostly interpret limitations of their own tools as accuracy or capability problems rather than diversity concerns, journalists remain formally responsible for outcomes shaped by decisions they cannot fully see or control, and the paper concludes that evaluating published content and relying on human approval of an original report are insufficient proxies for safeguarding news diversity.
Load-bearing premise
The claim rests on the assumption that the accounts of 16 purposively sampled media professionals describe a genuine and general shift in where diversity-relevant decisions occur, and that the two small validation studies—12 articles, four models, five audiences, and four illustrative sourcing cases—show these mechanisms are real features of AI-assisted news work rather than artifacts of specific prompts, models, or the researchers' evaluation choices; the authors acknowledge in Section 6 and their Ethical Considerations statement that the sample is purposive and that the studies establish mechanisms rather than prevalence.
Editorial extensions
If this is right
- Diversity evaluation must extend from published items to the informational conditions of production: comparing which sources are available through conventional practice, general-purpose models, and individually configured tools on comparable tasks.
- Meaningful editorial control under the EU AI Act's human-review exception requires showing that essential information survives adaptation, since approving an original report does not guarantee consistency across generated versions, and disclosure alone does not establish substantive preservation.
- Where reviewing a thousand versions by hand is infeasible, traceability to a reviewed original plus checks targeting essential information become necessary oversight measures.
- Accountability must reach those who require AI use, configure systems, and set review conditions; otherwise responsibility concentrated on front-line staff risks creating what the paper calls a moral crumple zone.
- Audience categorization and distributive fairness become diversity concerns: automated regional, minority-language, and off-peak services may expand access while distributing editorial attention and verification unevenly across audiences.
Reading between the lines
- If the relocation claim is right, conventional diversity metrics—source counts, viewpoint counts, recommendation diversity—systematically overstate the diversity of what audiences receive, because they cannot see what never entered the journalist's consideration; a natural extension is a 'source-horizon baseline' that measures the considered-versus-cited gap on routine reporting tasks.
- The mechanism implies a testable prediction: the diversity of sources cited in published stories may stay flat even as the diversity of sources considered declines, because journalists cannot report what they never encountered; tracking considered-versus-cited diversity over time would expose this.
- The asymmetry the paper documents suggests that newsroom AI training which frames omissions as craft problems may reproduce the blind spot; an intervention comparing craft-framed with diversity-framed review of AI-assisted sourcing could test whether reframing changes journalists' evaluations.
- The consistency results could be turned into a routine newsroom metric—a cross-version essential-information retention score analogous to existing recommendation-diversity metrics—so that divergence between audience versions becomes visible in production rather than in research studies.
Original abstract
The tension between news personalization and information diversity has traditionally been understood as a matter of different content reaching different people, with diversity evaluated by the range of sources, topics, and perspectives to which audiences are exposed. Generative AI outdates this understanding, since it transforms personalization in two ways: how journalists adapt news to audiences and how the AI tools journalists use personalize content production to their own preferences. Drawing on 16 semi-structured interviews with media professionals across public service, commercial, and state-affiliated news settings, we examine how these practices reshape editorial judgment and the conditions under which news diversity is produced. We find that participants negotiate AI integration by maintaining authority over meaning while delegating aspects of content presentation. Yet the tools they adapt for their own work also mediate which sources and perspectives become available, introducing diversity-relevant decisions that remain largely invisible to organizational oversight. While participants anticipate diversity risks in audience-facing personalization, they predominantly interpret diversity-imposed limitations in their own tools as technical problems. As a result, decisions that shape news diversity increasingly occur before publication and across multiple versions of a story, where neither evaluation of published items nor human approval can readily observe them. Journalists thus remain formally responsible for outcomes shaped by decisions they cannot fully see or control. Our findings highlight that safeguarding news diversity requires looking beyond published content to the professional practices, infrastructures, and distributions of control that shape it.
The compiled record
The paper's own sentences, arranged, kept beside the review and separate from it. Each label is drawn from a closed vocabulary; the number beside it is the probability the classifier gave that label. Numbers in the tables are the authors', checked by code. No sentence in this record was written by a model.
The abstract, sentence by sentence
The tension between news personalization and information diversity has traditionally been understood as a matter of different content reaching different people, with diversity evaluated by the range of sources, topics, and perspectives to which audiences are exposed. Generative AI outdates this understanding, since it transforms personalization in two ways: how journalists adapt news to audiences and how the AI tools journalists use personalize content production to their own preferences. Drawing on 16 semi-structured interviews with media professionals across public service, commercial, and state-affiliated news settings, we examine how these practices reshape editorial judgment and the conditions under which news diversity is produced. We find that participants negotiate AI integration by maintaining authority over meaning while delegating aspects of content presentation. Yet the tools they adapt for their own work also mediate which sources and perspectives become available, introducing diversity-relevant decisions that remain largely invisible to organizational oversight. While participants anticipate diversity risks in audience-facing personalization, they predominantly interpret diversity-imposed limitations in their own tools as technical problems. As a result, decisions that shape news diversity increasingly occur before publication and across multiple versions of a story, where neither evaluation of published items nor human approval can readily observe them. Journalists thus remain formally responsible for outcomes shaped by decisions they cannot fully see or control. Our findings highlight that safeguarding news diversity requires looking beyond published content to the professional practices, infrastructures, and distributions of control that shape it.
background contribution method result interpretation main claim
| # | sentence | role | the authors' own work | states the main conclusion | strength |
|---|---|---|---|---|---|
| s0 | The tension between news personalization and information diversity has traditionally been understood as a matter of different content reaching different people, with diversity evaluated by the range of sources, topics, and perspectives to which audiences are exposed. | background 1.00 | 0.04 | 0.02 | asserted 1.00 |
| s1 | Generative AI outdates this understanding, since it transforms personalization in two ways: how journalists adapt news to audiences and how the AI tools journalists use personalize content production to their own preferences. | gap 0.52 | 0.14 | 0.08 | asserted 1.00 |
| s2 | Drawing on 16 semi-structured interviews with media professionals across public service, commercial, and state-affiliated news settings, we examine how these practices reshape editorial judgment and the conditions under which news diversity is produced. | method 0.96 | 0.96 | 0.03 | asserted 0.91 |
| s3 | We find that participants negotiate AI integration by maintaining authority over meaning while delegating aspects of content presentation. | result 1.00 | 0.91 | 0.09 | asserted 1.00 |
| s4 | Yet the tools they adapt for their own work also mediate which sources and perspectives become available, introducing diversity-relevant decisions that remain largely invisible to organizational oversight. | result 0.98 | 0.82 | 0.12 | asserted 0.99 |
| s5 | While participants anticipate diversity risks in audience-facing personalization, they predominantly interpret diversity-imposed limitations in their own tools as technical problems. | result 0.99 | 0.88 | 0.07 | asserted 1.00 |
| s6 | As a result, decisions that shape news diversity increasingly occur before publication and across multiple versions of a story, where neither evaluation of published items nor human approval can readily observe them. | result 0.70 | 0.81 | 0.17 | asserted 0.99 |
| s7 | Journalists thus remain formally responsible for outcomes shaped by decisions they cannot fully see or control. | interpretation 0.85 | 0.73 | 0.18 | asserted 1.00 |
| s8 | Our findings highlight that safeguarding news diversity requires looking beyond published content to the professional practices, infrastructures, and distributions of control that shape it. | interpretation 0.95 | 0.75 | 0.86 | asserted 0.99 |
What was examined, and what was not
- arXiv listing abstract v1
- full text
- figures and tables
- references and citing sentences
- relations to other papers
- P abstract
- S sections
- T body sentences
- C citations
- R cross-paper relations
Figures
Figures reproduced from arXiv:2611.07931 by the authors.
References
Discussion
Disagree with the review, or know something it missed? Leave a comment, highlight any passage above to comment on it, or challenge this review formally.