The launch of Standards for Publisher Usage Rights coalition (SPUR), founded by the BBC, Financial Times, the Guardian, Sky News, and the Telegraph, is the most significant collective response the news industry has mounted to AI’s consumption of journalism.

Alongside Really Simple Licensing (RSL), the US-based open standard now endorsed by more than 1,500 media and technology organisations, it signals that publishers are finally coordinating rather than negotiating in isolation.

Several billion dollars in AI licensing commitments have been discussed, though the majority of deal terms remain undisclosed. There is real money at stake. But individual deals replicate the platform dependency that hollowed out publishers in the search and social cycles.

Collective standards create the leverage that fragmented deals cannot. SPUR is the right move.

However, it addresses only one side of the problem.

The problem licensing cannot solve

Even if every AI company pays fair value for every article it ingests, a deeper structural challenge remains: AI systems that have access to journalism often cannot use it accurately.

A Semrush study of 150,000 citations across 5,000 keywords found that Reddit appeared in around 40% of citations.

Goodie AI’s analysis of 5.7 million B2B SaaS (business-to-business software-as-a-service) citations (responses to questions about B2B software like “what’s the best CRM tool” or “compare project management platforms”) across ChatGPT, Gemini, Claude, and Perplexity confirmed the pattern: User-generated platforms and reference sites consistently outrank professional journalism. Not because journalism is less accurate, but because its content is not structured for machine retrieval.

Reddit’s threaded format and Wikipedia’s consistent meta-data are machine-readable by design. Journalism, written for human nuance, gets bypassed.

This is not a quality problem. It is a compatibility problem. You can negotiate the right to be paid when your content is used. You cannot negotiate your way into being cited if the AI system cannot reliably extract what you reported.

From SEO to SMO to LMO

Search engine opitimisation (SEO) made journalism discoverable on Google. Social media optimisation (SMO) aligned content with Facebook and Twitter. Both shaped how we wrote and structured stories. In both cases, the optimisation served someone else’s platform.

Now, what I have called language model optimisation (LMO) is becoming the next essential discipline. This requires preparing journalism so that AI systems can accurately understand, retrieve, and cite it. But unlike SEO and SMO, LMO’s primary purpose is not to feed external platforms. It is to power our own.

As Shuwei Fang has argued, most recently in her March 2026 Reuters Institute essay, the information ecosystem is being redrawn around machine audiences and liquid content. LMO is the content architecture publishers need to navigate that shift on their own terms, not someone else’s.

Ask The Post at the Washington Post, Hej Aftonbladet in Sweden, and Ask FT all use retrieval-augmented generation (RAG) to ground AI responses in verified journalism. These embody the early architecture of a model where publishers own the AI layer between their content and their audience.

However, most feed these systems with content optimised for human readers alone. The result is potentially inconsistent retrieval, incomplete context, and occasional inaccuracy. That is where LMO comes in.

Machine-ready without compromising quality

LMO does not mean degrading journalism. The LMO-optimised version is not what gets published. Readers continue to receive nuanced, narrative-driven journalism.

The LMO version exists as an internal parallel text that feeds AI systems, adding explicit entities, absolute dates, and semantically clear structures that help machines understand who did what, when, and where.

This dual-text approach improves retrieval accuracy and ensures journalism is represented faithfully in AI responses. The LMO version can be AI-generated from the published article with editorial verification.

It’s about quality control, not authorship. The cost is minimal. The strategic value is substantial.

Licensing funds the transition, not replace

SPUR and RSL address the commercial and technical frameworks for AI licensing. This is necessary infrastructure. But licensing revenue should fund a transition to publisher-owned AI capabilities, not become a substitute for one.

Cloudflare’s data reveals the power imbalance: Google crawls roughly 14 pages for every page of traffic it sends back to publishers. OpenAI has crawled roughly 1,000 to 3,700 pages for every visit it sends back, while Anthropic’s ratios range from tens of thousands to more than 70,000-to-one, depending on the period.

A licensing cheque from the same companies absorbing your traffic starts to look less like a revenue stream and more like a severance payment.

The Guardian shows what the alternative looks like. A SPUR founding member with licensing deals in place with both OpenAI and Google, The Guardian has simultaneously been training AI tools grounded in its own journalism and editorial values. Licensing and infrastructure investment are not alternatives. They are sequential steps.

In what I have described as the AI buddy economy, where audiences increasingly interact with publisher-built AI companions rather than traditional news apps, the organisations more likely to thrive will be those that build those AI products grounded in their own journalism, serving their own editorial mission, and maintaining direct audience relationships.

The architecture for this exists. LMO provides the content layer. RAG systems provide the retrieval layer. Publisher-owned AI companions provide the audience layer. Together, they represent a model where journalism is not raw material for someone else’s product but the foundation of your own.

SPUR is the right collective response to the licensing challenge. Now publishers need to build what comes after. We have ceded that ground twice. We cannot afford a third time.

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .