Normative frameworks, standards and regulation for news diversity and algorithmic curation (with India)

Research date: 4 October 2026. "Binding" = hard law in force; "soft law" = recommendations, guidance, codes. Items I could not verify from a primary or reliable secondary source are listed under Gaps rather than stated as findings.

1. Diversity theory: what "balanced" or "diverse" exposure means, and which kind equal turns per owner implements

Takeaway

The literature separates where diversity is measured (source/ownership, content, exposure; supply vs exposure) from what norm it is measured against: "reflective" diversity (mirror the distribution in society or the market) vs "open" diversity (equal shares for every voice). Democratic theory makes the norm a choice: liberal (user-driven), participatory (reflective/proportional), deliberative (equal representation of perspectives), critical (over-represent marginal voices). A rule that gives every owner one turn before any owner gets a second, regardless of how much each owner publishes, is an "open"/equal-distribution norm applied at the source (ownership) level of supply diversity, close to the deliberative model's "equal representation"; ranking stories by how many outlets carried them is, by contrast, a reflective (volume-proportional) signal.

Cited Findings

• Loecherbach et al. (2020), "The Unified Framework of Media Diversity: A Systematic Literature Review", Digital Journalism 8(5): 605–642. They organise diversity by locus (supply diversity = what is available; exposure diversity = what individuals consume; personalisation blurs the line) and dimensions (entities/sources/actors, topics, viewpoints, structural features such as ownership and outlets, plus language and medium). — Loecherbach et al. 2020 (PDF)
• Same review restates Napoli (1999): source diversity (producers/outlet ownership), content diversity (topics and viewpoints) and exposure diversity (horizontal = spread of audience across options; vertical = spread within an individual's consumption). — Loecherbach et al. 2020
• Same review distinguishes reflective diversity ("mirroring societal heterogeneity") from open diversity ("equal distribution of elements"). — Loecherbach et al. 2020
• Gaps the review identifies: computer-science diversity studies rarely ground themselves in democratic theory; most empirical work measures simple "variety" (counts) and ignores balance and disparity; 142 of 206 articles study supply rather than exposure; automated measures lack validation. — Loecherbach et al. 2020
• Napoli later revisited exposure diversity as a policy goal: "Exposure Diversity Reconsidered", Journal of Information Policy 1 (2011): 246–259. — Scholarly Publishing Collective
• Helberger, Karppinen & D'Acunto, "Exposure diversity as a design principle for recommender systems", Information, Communication & Society 21(2) (online 2016/17, issue 2018), DOI 10.1080/1369118X.2016.1271900. They propose "diversity sensitive design" drawing on democratic theory, computer science and law, and discuss its limits. — IViR
• Operationalisation follow-up: Moe, Hovden & Karppinen, "Operationalizing exposure diversity", European Journal of Communication (2021). — SAGE
• Helberger, "On the Democratic Role of News Recommenders" (Digital Journalism, 2019) sets out the democratic-theory models used for recommender design. — IViR PDF
• Vrijenhoek, Möller, Kaya, Odijk, Metoui & Helberger, "Recommenders with a Mission: Assessing Diversity in News Recommendations" (CHIIR 2021; arXiv 2012.10185) map four democratic models to diversity norms:
  • Liberal: user-driven; diversity follows personal preferences, specialisation acceptable.
  • Participatory: proportional ("reflective") representation of mainstream viewpoints plus editorial steering towards important topics.
  • Deliberative: "equal representation" of perspectives in a debate; impartial tone.
  • Critical: "inverse" representation, prioritising under-represented and marginalised voices.
  — arXiv 2012.10185
• Their five metrics: Calibration (KL divergence between recommendations and user's preferences on topic/style/complexity), Fragmentation (overlap of story chains across users, Rank-Biased Overlap; a shared public agenda vs bubbles), Activation (emotional intensity via sentiment), Representation (KL divergence of viewpoints in recommendations vs a model-specific target), Alternative Voices (share of minority/marginalised group mentions). The authors warn of the "formalism trap": metrics are a starting point, not the concept. — arXiv 2012.10185
• Van Cuilenburg & McQuail, "Media Policy Paradigm Shifts: Towards a New Communications Policy Paradigm", European Journal of Communication 18(2) (2003): 181–207 — diversity "in political, social and cultural terms" as a core policy commitment across paradigms (this article does not itself define reflective/open). — PDF
• Van Cuilenburg, "On Competition, Access and Diversity in Media, Old and New", New Media & Society 1(2) (1999), is the usual cited source for the reflective/open distinction and links diversity to access. — SAGE; a critique of open-diversity statistics is at Radboud repository (could not be fetched, 403).

Inferences

• Equal turns per owner (Huduku's pickBalanced) = open diversity of sources at the ownership level. It equalises each owner's opportunity to appear, independent of output volume, i.e. "equal distribution of elements" rather than "mirroring" the market. In Vrijenhoek et al.'s terms it is closest to the deliberative norm (equal representation), applied to owners rather than viewpoints. It is not reflective diversity, which would allocate room in proportion to output or audience share.
• "Top stories by number of outlets (owners) carrying a story" is a reflective signal of the news agenda (how widely covered), not of popularity; counting owners rather than feeds prevents one group's many titles from counting as many voices, which matches the source-diversity logic behind ownership rules.
• It is supply/source diversity, not content (viewpoint) diversity: one owner per turn does not guarantee different viewpoints, and it is not exposure diversity, which depends on what readers click. A neutral reader can only guarantee diversity of what is offered; the literature treats exposure as the reader's.
• Huduku's design (no personalisation, same page for all) avoids the "fragmentation" risk by construction: every reader sees the same agenda.

Gaps

• Could not retrieve van Cuilenburg (1999) full text to quote his exact definitions of reflective vs open diversity; the definitions above come via Loecherbach et al. 2020.
• Did not retrieve Napoli 1999 ("Deconstructing the Diversity Principle", Journal of Communication 49(4)) or McQuail's Media Performance (1992) directly; content attributed via the 2020 review.
• Helberger 2011 "Diversity by design" (Journal of Information Policy 1) not fetched.

2. Measuring ownership concentration (HHI, CR4, effective number, MPM, India)

Takeaway

Antitrust uses HHI (sum of squared market shares, 0–10,000); the US 2023 Merger Guidelines treat >1,800 as highly concentrated (back to the 1982 level, down from 2,500 in 2010) and add a 30% share presumption. India's TRAI proposed in 2014 to use HHI in 12 language/state news markets with a 1,800 "concentrated" trigger and a 1,000-point per-entity contribution cap, but this was never made law. Per-owner audience caps exist in a few places (Germany's 30% TV audience share; UK's 20% national newspaper share for Channel 3), and are generally confined to broadcast and print. The RSF/DataLEADS Media Ownership Monitor (2019) found Indian print highly concentrated in each language market and no concentration thresholds in Indian law.

Cited Findings

HHI and antitrust thresholds

• HHI = sum of squared market shares; e.g. shares 30/30/20/20 give 2,600; maximum 10,000 for a monopoly. Markets 1,000–1,800 are moderately concentrated, >1,800 highly concentrated; a rise of >100 points in a highly concentrated market is presumed to enhance market power (page updated 17 Jan 2024). — US DOJ, HHI
• The 2023 Merger Guidelines (DOJ/FTC, Dec 2023) lowered the "highly concentrated" threshold from 2,500 (2010 guidelines) to 1,800, presume illegality where ΔHHI >100, and add a presumption for a merged firm with >30% share. — Dechert, Dec 2023; CRS LSB11138
• Noam, Who Owns the World's Media? Media Concentration and Ownership around the World (OUP, 2016): 30 countries, 13 media industries, 10–25 years of data; national concentration trends and transnational shares. — Columbia Business School

India: TRAI

• TRAI, Recommendations on Issues Relating to Media Ownership, 12 August 2014 (recommendation, never enacted):
  • relevant market = news and current affairs (incl. business news) only, TV and daily newspapers (para 3.13, 3.20);
  • 12 geographic markets by language/state: Hindi (10 states), English (pan-India), and regional languages incl. Kannada (para 3.27);
  • HHI as the measure (para 3.45); a market is concentrated at HHI > 1,800; an entity's contribution threshold 1,000 (≈ 32% share) (para 3.51);
  • "1 out of 2" cross-media rule: where TV and print are both concentrated, an entity contributing >1,000 to one market's HHI may not contribute >1,000 to the other's; applies only if breached two consecutive years (paras 3.52–3.53);
  • "control" at 20% equity or de jure/de facto control (paras 2.13–2.14);
  • annual public disclosure of shareholding, media interests, board, loans; changes within 30 days (para 3.67); 1 year to comply, 3-year review (para 3.61).
  — TRAI 2014 Recommendations (PDF); summary at MediaNama, Aug 2014
• TRAI reopened the question with a Consultation Paper on Issues relating to Media Ownership, 12 April 2022. — TRAI CP 2022 (PDF)

India: Media Ownership Monitor (RSF + DataLEADS, 29 May 2019)

• Four Hindi dailies (Dainik Jagran, Hindustan, Amar Ujala, Dainik Bhaskar) have 76.45% of Hindi readership; in each regional language market the top two papers have >50% of readers; Telugu's Eenadu and Sakshi 71.13%; Tamil top two about two-thirds. — RSF
• Radio news is a state monopoly (All India Radio); private FM cannot broadcast news. TV audience data were unavailable because BARC would not disclose them. — RSF
• The regulatory framework has "considerable gaps", with no concentration thresholds or measurement standards; cross-shareholdings obscure beneficial owners; many owners have political affiliations. — RSF; MOM India indicators

EU: Media Pluralism Monitor (CMPF/EUI)

• Four areas: Fundamental Protection, Market Plurality, Political Independence, Social Inclusiveness. Market Plurality indicators: transparency of media ownership; plurality of media providers; plurality in digital markets (platforms, search, AI intermediaries); media viability; editorial independence from commercial and owner influence. — MPM2025 general report
• MPM2025 moved to six risk bands: very low 0–16%, low 17–33%, medium-low 34–50%, medium-high 51–67%, high 68–83%, very high 84–100% (earlier editions used three: low/medium/high). — MPM2025
• "Market Plurality is the only area that shows an average high-risk level across EU countries"; plurality of media providers and plurality in digital markets are both at very high risk. — MPM2025; European Commission notice
• CMPF notes concentration measurement often relies on estimates and commercial databases. — CMPF blog

Per-owner audience caps

• Germany (binding): under § 60(2) Medienstaatsvertrag, "vorherrschende Meinungsmacht" (dominant power over opinion) is presumed at a 30% TV audience share, or 25% with a strong position in related markets (radio, press, streaming); KEK assesses. Critics: it is limited to TV and barely captures online intermediaries. — bpb; MStV § 60
• UK (binding): the national cross-media ownership rule bars a newspaper operator with ≥20% of national newspaper circulation from holding a Channel 3 licence (and vice versa); plus a Media Public Interest Test for mergers (Enterprise Act 2002) with Ofcom advising on "sufficient plurality". No online-intermediary coverage. — Ofcom, legal framework annex; Ofcom's s.391 review report, 15 Nov 2024: Ofcom (not fetched; URL too long)
• Council of Europe (soft law): CM/Rec(2018)1 says thresholds "may be based on ... capital shares, voting rights, circulation, revenues, audience share or audience reach" without numbers. — CM/Rec(2018)1

Inferences

• Effective number of owners = 10,000 / HHI (shares in percent) or 1/HHI (shares as fractions); e.g. HHI 2,500 ≈ four equal owners. TRAI's 1,000-point contribution ≈ √1000 ≈ 31.6% share, matching its "≈32%" gloss. (Arithmetic from the DOJ definition.)
• Huduku's "past 10 cards an owner waits" and check:balance's 15% one-owner alarm are display-level caps far tighter than any legal ownership cap (Germany 30%, UK 20%, TRAI ≈32%); they are design norms, not legal analogues. At 15% max share, worst-case HHI contribution is 225 per owner.
• Computing HHI/CR4 over owners on Huduku's own pages (cards per owner) would be a checkable, count-only disclosure consistent with "Huduku counts; readers judge". CR4 for Hindi print (76.45%) is the kind of number MOM India reports.

Gaps

• Could not confirm from a primary source the 2010 Horizontal Merger Guidelines' 1,500 "unconcentrated" threshold and ΔHHI 200 (only the 2,500→1,800 change is sourced).
• Did not find whether TRAI issued final recommendations after the 2022 consultation; as far as I found, none was enacted. Needs checking against TRAI's site.
• Noam's India chapter figures were not accessible.
• MPM's exact CR4/Top4 thresholds for the "plurality of media providers" indicator not extracted.

3. Regulation of algorithmic curation (binding vs soft law), including India

Takeaway

Binding law on recommender systems is strongest in the EU: the DSA (Art. 27 transparency of "main parameters" for all online platforms; Arts. 34–35 systemic risk incl. "freedom and pluralism of the media" and Art. 38 non-profiling option for very large platforms), and the European Media Freedom Act (applies from 8 August 2025: Art. 18 media privileges on VLOPs, Art. 20 right to customise, Art. 22 pluralism tests for media mergers). The Council of Europe's texts are soft law but are the most explicit about exposure diversity and prominence of public interest content. Australia's bargaining code is about payment, not ranking. India has no aggregator-specific diversity rule; the IT Rules 2021 name "news aggregators" as digital news publishers bound by a Code of Ethics, but the key Rule 9 provisions are stayed by two High Courts.

Cited Findings

EU Digital Services Act (Regulation (EU) 2022/2065) — binding

• Art. 27: online platforms must set out in plain language in their terms "the main parameters used in their recommender systems" and any options for users to modify them, with an easily accessible function to select and change the preferred option at any time. — DSA Observatory, Nov 2024
• Art. 38: very large online platforms must offer at least one recommender option "not based on profiling". — DSA Observatory
• Art. 34(1)(b) lists systemic risks to fundamental rights including "freedom and pluralism of the media"; Art. 35 includes adapting recommender systems as mitigation. DSA Observatory finds platforms took a "minimalist interpretation", explaining systems without giving real control. — DSA Observatory

European Media Freedom Act (Regulation (EU) 2024/1083, 11 April 2024) — binding

• Art. 18: media service providers self-declare editorial independence, regulatory or self-regulatory oversight, and no AI-generated content without human review; VLOPs must give them a statement of reasons and 24 hours to reply before suspending content or restricting its visibility; their complaints get priority. — Art. 18 text; EUR-Lex
• Art. 19: structured dialogue (at least yearly) between VLOPs, media and civil society, organised by the European Board for Media Services. — EUR-Lex
• Art. 20: right of users to easily change default settings of devices/interfaces controlling access to media services (recital 57: customisation to reflect users' interests, "fair access to media services in all their diversity"). — EUR-Lex
• Art. 22: national assessments of media market concentrations must consider impact on media pluralism, safeguards for editorial independence, economic sustainability, rule-of-law findings, and commitments offered. — Art. 22 text
• Art. 6: news and current affairs providers must publish ownership information (incl. beneficial owners) in an easily accessible electronic form. — EUR-Lex
• Art. 29: most provisions apply from 8 August 2025. — EUR-Lex

Council of Europe — soft law (non-binding)

• CM/Rec(2018)1 on media pluralism and transparency of media ownership (adopted 7 March 2018): states should encourage measures to "improve users' effective exposure to the broadest possible diversity of media content online" (Appendix 2.5); intermediaries should be transparent about algorithmic processes, assess their impact on diversity, and guide users (2.4–2.5); ownership disclosure above a recommended 5% threshold incl. beneficial owners (4.1, 4.5). — CM/Rec(2018)1; ARTICLE 19 summary
• CM/Rec(2022)11 on principles for media and communication governance (2022): 15 principles across production, dissemination and use; governance should mitigate "the risks posed by algorithmic curation, selection and prioritisation", promote pluralism, transparency/accountability and user empowerment. — CoE edoc
• Guidance Note on the Prioritisation of Public Interest Content Online (CDMSI(2021)009, adopted 1–3 Dec 2021): criteria should be "clear, non-discriminatory, viewpoint neutral, transparent, and objectively justifiable" (¶17), preferably set by independent/self-regulatory standards rather than states; platforms should publish criteria and "how the criteria / standards are used ... and by whom or what" (¶25.i.b); content providers may opt out and users should be able to make informed choices (¶25.vi); states should not oblige carriage of specific content or prioritise their own communications except in emergencies (¶17, ¶21). — CoE guidance note
• Related: CoE IRIS Special 2023 on algorithmic transparency and accountability of digital services. — European Audiovisual Observatory

Australia

• News Media and Digital Platforms Mandatory Bargaining Code, in effect 2 March 2021 (binding statute, but no platform has been designated); the ACCC says the threat of designation drove 30+ deals. — ACCC
• Government announced a "News Bargaining Incentive" (Dec 2024) with consultation in 2025. — Treasury minister release

India

• IT (Intermediary Guidelines and Digital Media Ethics Code) Rules, 2021 (binding subordinate legislation, as amended to 2023): Rule 2(1)(o) defines a "news aggregator" as an entity that, performing a significant role in determining the news and current-affairs content made available, makes available such content aggregated/curated/presented; aggregators fall within "publisher of news and current affairs content" under Part III, which requires a Code of Ethics (Press Council of India Norms of Journalistic Conduct; Cable TV Programme Code) and a three-tier grievance system. "Curation" is undefined. — MeitY consolidated rules (PDF); MediaNama, June 2021
• Bombay High Court, 14 Aug 2021 (AGIJ Promotion of Nineteenonea Media / The Leaflet; Nikhil Wagle) stayed Rules 9(1) and 9(3) (Code of Ethics adherence and the three-tier oversight). — Internet Freedom Foundation
• Madras High Court, 16 Sep 2021 also stayed Rules 9(1) and 9(3), saying government control "may rob the media ... of their independence". — The Tribune
• Draft IT Rules Second Amendment, 30 March 2026 (draft): would extend Part III Rules 14–16 (Inter-Departmental Committee, emergency blocking) to news and current affairs content posted by non-publisher users and intermediaries, and make ministry advisories part of due diligence. — IFEX
• Digital Personal Data Protection Act 2023 / DPDP Rules 2025: Rules notified 13–14 Nov 2025 with staggered commencement (about 18 months for most obligations); verifiable parental consent for children's data. — PIB (PDF); IFF statement
• Press Council of India: Norms of Journalistic Conduct, 2022 edition (self-/co-regulatory; PCI's statutory remit is print). — PCI
• No Indian ownership-concentration threshold exists in law (MOM 2019); TRAI's HHI scheme is a recommendation only. — RSF; TRAI 2014

Inferences

• A small, non-EU, non-personalised aggregator like Huduku is not a VLOP and is outside the DSA/EMFA; but Art. 27's "main parameters + why they matter" is the de facto disclosure benchmark, and the CoE guidance note's "viewpoint neutral, transparent, objectively justifiable" criteria are the closest soft-law description of an owner-balanced, non-engagement ranking.
• In India, Huduku, if seen as "aggregating, curating and presenting" news, could fall within the IT Rules' "news aggregator" definition (Part III); Rule 9(1)/(3) remain stayed, but other Part III obligations (e.g. furnishing information under Rule 18, grievance officer under Rule 10/11) were not stayed as far as these sources show. This needs legal advice; listed in CLAUDE.md as pending legal review.
• No Indian norm requires or forbids owner-balancing; the closest Indian policy precedent for owner-level counting is TRAI's 2014 language-market HHI, which uses the same "per owner, per language market" unit that Huduku's owner counts use.

Gaps

• Exact wording of Rule 2(1)(o) not quoted from the gazette (paraphrased from MeitY consolidated text as reported); verify.
• Current status (2026) of the Bombay/Madras stays and of the transferred petitions in the Supreme Court not found.
• Whether the draft 2026 amendment has been notified not found.
• DPDP: exact commencement dates per section and the Section 9 ban on tracking/behavioural monitoring of children not quoted from primary text (Huduku's no-tracking design makes it moot).
• Text of the DSA itself not fetched from EUR-Lex; Art. 27/34/38 wording via DSA Observatory.

4. Professional and ethics guidance for algorithmic news, and transparency of ranking criteria

Takeaway

Professional guidance converges on: disclose what data, model, inference and interface choices a news algorithm makes (Diakopoulos & Koliska 2017); give editorial "business rules" that can override engagement (BBC); measure diversity explicitly (VRT's taste-broadening score); and publish ranking criteria in plain language (DSA Art. 27; CoE 2021 note). Public service media studies show a recurring tension between engagement metrics and diversity goals.

Cited Findings

• Diakopoulos & Koliska, "Algorithmic Transparency in the News Media", Digital Journalism 5(7) (2017): focus-group study proposing disclosure across layers — data, model, inference, interface. — Semantic Scholar; ResearchGate
• Iordache et al. (2025), "Public service algorithms: Balancing the scales between public mission and market pressures at the BBC and VRT", MedieKultur: VRT uses a "taste score" to track and broaden the spread of consumption across categories; the BBC uses editorial "business rules" ("amplifying certain pieces of content, blocking old content, downweighting certain things") and prioritises impartiality; a BBC product chief said the algorithm "really should be focused around finding out what you like"; VRT: "It would be much easier if we were commercial ... but because taste-broadening is included, it's much more difficult." Core tension framed as universality vs diversity. — White Rose eprint; MedieKultur
• RTBF case: embedding public-service values in a news recommender (2024). — ResearchGate
• "Public Service Algorithm" proposal for transparent, editorial-value-based curation (arXiv 2506.22270, 2025). — arXiv
• Journalism Trust Initiative (RSF; CEN Workshop Agreement CWA 17493) — a standard for trustworthiness of newsroom processes, referenced by the CoE 2021 guidance as the kind of independent standard platforms may use for prioritisation. — JTI certification; CoE guidance note
• CoE MSI-RES Guidelines on the responsible implementation of AI systems in journalism (2023). — PDF
• Disclosure benchmarks: DSA Art. 27 "main parameters" and user options (DSA Observatory); CoE 2021 note ¶25.i.b "how the criteria / standards are used ... and by whom or what" (CoE); CM/Rec(2018)1 2.5 transparency of automated processes (CoE).

Inferences

• A disclosure for Huduku that would meet all of these: name each ranking rule (owner turns, 10-card soft ceiling, Top stories by owner count, recency weight owners ÷ (1 + quiet hours ÷ 8)), say why, say there are no hand-set boosts or personal signals, and publish the per-owner card counts (check:balance). Diakopoulos & Koliska's four layers map onto: data (which public_ok feeds, owner facts with sources), model (balance rules), inference (story grouping thresholds 0.50/0.40/0.60), interface (labels say who labelled).

Gaps

• EBU "PEACH" recommender project and any EBU diversity principles: not found in this pass.
• ZDF's published recommender principles: not found.
• Partnership on AI guidance specific to news ranking: not searched/found.
• No code of practice specific to news aggregators (India or elsewhere) was found beyond the IT Rules' Part III.