Schibsted, Axel Springer and a third Iberian group spent late May signalling the same thing to OpenAI and Perplexity. Flat licensing cheques are not enough. They want revenue share on advertising sold inside the answers their journalism trains and grounds. Pilots are being floated for the third quarter. The shape of the next publisher deal is not a content licence at all. It is a joint ad product.
1. Licensing was always a holding pattern
The first wave of publisher deals with AI platforms looked like victory. News Corp at a reported quarter of a billion across five years. Axel Springer locked in. Le Monde, the FT and others taking annual cheques in the low eight figures. Those numbers flattered the moment but disguised the structure. Licensing is a one-time content transaction priced against a platform's training budget, not against the commercial value the platform extracts downstream. Once an answer engine grounds a response in your reporting, the user does not click. The visit does not happen. The ad on your homepage does not render. Publishers traded a recurring revenue line for a finite payment, and the finite payment shrinks in real terms every quarter as platforms diversify their training corpora. The European group now pushing back has done the maths. A licence pays for the past. Revenue share pays for the surface area you keep building.
2. The unit being sold is the answer, not the article
What changes if Schibsted gets its way is the inventory definition. Today an ad impression is a slot on a page rendered in a browser. Tomorrow, inside ChatGPT or Perplexity, the unit is a sponsored module attached to a generated answer. That module needs grounding sources. The sources are the publishers. The proposal on the table treats every grounded answer as a co-produced ad surface, with revenue split between the platform that rendered it and the newsrooms whose reporting made the answer trustworthy enough to monetise. This is closer to a programmatic supply deal than a content licence. It also changes what publishers need to instrument. Article-level attribution against answer-level monetisation. Citation frequency as a billing input. Whether the sponsored unit is visually demarcated from the editorial answer, which the AI Act's transparency obligations will eventually force regardless. The commercial primitive shifts from the URL to the citation.
3. The platforms have a reason to say yes
Perplexity already runs sponsored questions and has been public about wanting publisher revenue share since late 2024. OpenAI's position has been quieter but the direction is the same. Both companies face the same problem. Licensing budgets cannot scale with content needs forever, and litigation from publishers who refused early deals keeps multiplying. A revenue-share model converts an adversarial cost line into a partnership flywheel. Platforms get durable access to premium reporting without renegotiating cheques every eighteen months. Publishers get exposure to the upside if answer-engine ad load grows, which every credible forecast says it will. The risk for the platforms is margin compression on a business that is not yet profitable. The risk for publishers is signing a share of nothing. Both sides know it, which is why the Iberian group is reportedly insisting on minimum guarantees underneath the rev-share structure. That is the negotiation that matters in Q3.
4. The Connect pillar gets a clearer brief
For Broadbrand this is not an abstract policy story. Our Connect work with Newsroom AI and The Brief has been built on the assumption that publisher inventory and AI distribution would converge into joint products rather than parallel pipes. The European move validates the operating model. If sponsored answer units become a standard inventory type, the buy-side question is no longer whether to include them but how to sequence them against open-web display, CTV and retail media in a single plan. The sell-side question is whether your newsroom can supply structured, citation-ready content at the cadence an answer engine demands, and whether your ad ops team can reconcile revenue across surfaces it does not directly render. Both are operational problems, not strategic ones. Both are solvable now if the work starts before the pilots go live. Waiting for the IAB to publish a spec is waiting too long.
5. Generative engine optimisation becomes a revenue function
GEO has been treated, when it has been treated at all, as an SEO descendant. A way to get cited inside AI answers so brand mentions surface. Once answer-level ad inventory exists, GEO becomes something more direct. It becomes the discipline that determines whether your publisher partner gets paid, and whether your client's brand appears in the sponsored module attached to the answer your content helped ground. The skill set sits across editorial structure, schema, prompt-aware copy, and the kind of attention measurement that does not depend on a click. Few agencies have built it. Fewer have built it in-house against live publisher relationships. The European pilots, if they land in Q3 as briefed, will produce the first real performance data on what a citation is worth in advertising terms. Whoever has the instrumentation ready to read that data will write the next playbook. The rest will buy it from them.
The licensing era will not end cleanly. Cheques will keep being written, particularly by platforms that need a quiet news cycle more than they need a sustainable model. But the European group has named the next shape correctly. Premium publishing only monetises generative platforms if it owns part of the ad surface those platforms create, and the unit of that surface is the answer. Build for that now, while the pilots are still being scoped, and you set the terms. Wait, and you take whatever rev-share split lands in your inbox.
One caveat on dates. The Q3 2026 pilot window is what the reporting indicates this week, but the publishers involved have not yet confirmed signed timelines, so treat the quarter as directional until the first pilot is named.