AI-generated news needs provenance “nutrition labels” — and fair licensing that protects local journalism

Executive summary

• As AI becomes many people’s front door to news, two urgent fixes are required: clear provenance “nutrition labels” and a licensing regime that won’t squeeze out smaller publishers.
• IPPR tested ChatGPT, Google AI overviews, Google Gemini and Perplexity across 100 news queries (~2,500 links) and found skewed sourcing, reduced clickthroughs, and concentration risks.
• Practical next steps for business leaders: demand provenance, push for collective bargaining, insist on audit rights and algorithmic discoverability, and diversify revenue away from platform dependency.

The problem, boiled down

As AI becomes people’s front door to news, two faults are already clear: answers often hide where they came from, and monetisation favours the publishers already partnered with tech giants. Both problems threaten trust and the economic health of local and investigative journalism.

Provenance labels (short, machine-readable badges that show which outlets informed an AI answer) are the transparency fix. A licensing regime that lets publishers collectively bargain with platforms is the payment fix. Neither is magic, but both are practical—and urgent.

Picture a reader asking ChatGPT “what happened with the council budget vote” and getting a confident paragraph with no cited sources. That answer can shape public understanding—and it can cut traffic to the local paper that did the reporting. That’s the harm IPPR measured.

What IPPR actually tested and found

IPPR ran 100 news-focused queries through four AI tools and analysed roughly 2,500 links those tools surfaced. Key findings:

• Source concentration: ChatGPT cited The Guardian in about 60% of its answers, likely reflecting licensing arrangements; some widely used publishers (Telegraph, GB News, The Sun, Daily Mail) appeared in fewer than 4% of ChatGPT outputs.
• BBC content: ChatGPT and Google Gemini rarely cited BBC journalism because the BBC blocks many scraping bots, while Google AI overviews and Perplexity still surfaced BBC content despite BBC objections.
• Reach and behaviour: Google’s AI overviews reach roughly 2 billion users per month and IPPR observed that AI overviews can reduce click-through traffic to publishers.
• Public use: Reuters Institute data suggests about 25% of people now use AI to get information—so the stakes are large for public knowledge and for publisher revenue.

“If AI firms profit from journalism and shape public information, they should be required to pay fairly for news and operate under rules that protect plurality, trust and independent journalism.” — Roa Powell, IPPR

Three policy levers that must work together

1) Provenance “nutrition” labels

What they are: a short badge and metadata attached to an AI answer showing which outlets informed the response, ideally with a direct link and a trust signal (publisher name, timestamp, degree of reliance).

How they could be enforced: platform UI standards combined with machine-readable metadata (HTML microdata or schema.org), audit logs publishers can inspect, and regulatory requirements backed by the Competition and Markets Authority (CMA) or an equivalent code of practice.

Mockup (text-only):

[Provenance] Sources: The Guardian (45%), Local Gazette (25%), BBC report (15%) · Updated: 2026-01-15 · View sources

Design principles: concise for humans, machine-readable for audits, and linked to a verifiable audit trail so publishers can see how their content was used.

2) A licensing regime for publishers

IPPR recommends a UK licensing framework that enables collective bargaining so publishers can negotiate with platforms instead of taking one-off deals. The Competition and Markets Authority (CMA) can force platforms to the table and start structured negotiations—similar in spirit to Australia’s News Media Bargaining Code, though tailored to UK competition law and media pluralism concerns.

Why copyright matters: keeping copyright intact creates the legal basis for a functioning licensing market. But licensing without guardrails risks cementing the position of the few publishers already partnered with platforms.

3) Public support and safeguards for plurality

Public funding for local and investigative journalism, incentives for non-profit models, and supporting public broadcasters (like the BBC) to experiment with AI on their terms are all part of the safety net. Guardrails should include mandatory discoverability rules, revenue-sharing floors, and publisher audit rights.

How licensing can go wrong — and how to prevent it

Licensing can replace some lost ad revenue, but it can also create dependencies that evaporate if platform priorities change. Worse, prominence in AI answers could be conditioned on licensing deals, locking out smaller outlets from discoverability. Practical guardrails include:

• Mandatory algorithmic discoverability — platforms must allocate a portion of AI answers to diverse sources, not just licensees.
• Transparency and audit rights — publishers can see how models use their content and how revenue is calculated.
• Revenue floors or scaling rules — prevent a winner-takes-most scenario that hollow out local media.

Why provenance is technically harder than it sounds

Large models synthesize information. When answers are generated via retrieval-augmented generation (RAG), a provenance trail is easier to record because the system explicitly pulls discrete sources. When models are trained on large swathes of web text, provenance is murkier. That’s why provenance standards must include both training-source declarations and runtime source attribution.

Regulatory comparators

Australia already forced tech platforms into bargaining with publishers under its News Media Bargaining Code, with mixed effects. The European Union’s Digital Markets Act and the upcoming layers around AI governance also create a regulatory backdrop for provenance and platform accountability. These models offer playbooks and cautionary tails—useful templates for the UK’s CMA and policymakers.

Mini case study: a local paper’s upside and downside

Upside: Under a collective licensing deal with discovery protections, a local paper could receive a steady revenue share and see its investigative pieces surfaced by AI answers, increasing subscriptions.

Downside: If prominence is reserved for a small group of licensed national outlets, the same local paper could be invisible to the millions who rely on AI overviews—losing both audience and ad revenue.

Playbook for C-suite and media owners

Five immediate steps:

• Demand provenance: insist any platform partnership includes machine-readable provenance labels and measurable attribution.
• Join collective bargaining: work with peers to avoid one-off, unequal deals.
• Secure audit rights: negotiate the ability to inspect how your content is used and how revenue is calculated.
• Diversify revenue: invest in membership, events, branded content, and other direct-to-audience models that AI cannot easily replicate.
• Pilot defensible content: develop reporting that is unique, data-rich, or regulatory in nature—stories that retain value beyond summary answers.

Questions to ask platform partners:

• How will you display provenance for AI-generated answers that rely on our journalism?
• Will we get access to audit logs showing how often our content informed answers?
• How is revenue shared and on what basis is it calculated?
• Can you commit to discoverability safeguards for non-licensed and local outlets?
• What governance mechanisms allow us to contest misuse or misattribution of our content?

Key takeaways & questions

• Who should pay when AI firms use journalism?
  

  Platforms that profit from summarising and redistributing journalism should share revenue with content creators through fair, collective licensing arrangements.

• Will provenance labels confuse users?
  

  Well-designed labels can be concise and actionable—like calorie counts for news—improving trust without cluttering the UI.

• Can licensing replace lost ad revenue?
  

  Licensing can offset some losses but is unlikely to fully replace advertising; diversified income and public funding for public-interest journalism remain essential.

• Are regulators ready to act?
  

  The Competition and Markets Authority (CMA) and similar bodies have new powers; there is a narrow window to establish rules before norms harden.

Policy decisions now will shape whether AI amplifies a plural news ecosystem or centralises control in a handful of platform-publisher relationships. For executives, the tactical priorities are simple and urgent: insist on provenance, organise for collective bargaining, secure audit and discoverability rights, and continue investing in journalism that cannot be cheaply replicated.

If helpful, a practical starter pack can be prepared: a one-page provenance-label mockup, a sample collective-negotiation term sheet, and a quick risk matrix showing exposure to different platform scenarios. These materials make meetings with platform partners and regulators concrete, focused, and hard to ignore.