The media industry has long written its AI story. Newsrooms use generative tools for summaries. Publishers experiment with personalised recommendations. At every industry conference between Austin and Cannes, at least one executive shows a slide with the words 'AI strategy' on it — and the audience nods.

The narrative is reassuring: whoever deploys AI tools has kept pace. 43 percent of US workers now actively use AI on the job. In Europe the rate is 25 to 36 percent. The numbers are rising, the dashboards are green, and the press releases almost write themselves.

But what if the entire industry is measuring the wrong indicator?

Look not in press releases but in patent offices, and a different world emerges. Of the five largest Western media houses — Bloomberg, New York Times, News Corp, Dotdash Meredith, Condé Nast — only Dow Jones holds a measurable AI patent portfolio: 30 patents globally, 10 granted. The other four: near-zero entries in IPC classes G06N and G06F.

For comparison: IBM holds 37,407 active patent families. Microsoft holds 29,640. A single tech group owns more AI IP than the entire media industry combined — by a factor of 1,000. That is not a gap. That is a different coordinate system.

The R&D spend looks respectable at first glance. The New York Times invests around USD 132 million in product development against total revenue of USD 2.8 billion. IAC / Dotdash runs at roughly 9 percent of revenue. The issue is not the size of these numbers — it is their destination: mostly platform licences (CMS, CDP, ad-tech stacks). The share that flows into proprietary AI infrastructure is not even reported separately. What is not measured does not exist as a strategic priority.

This is where it gets uncomfortable. The strongest counter-argument is obvious: media companies are not technology companies. They do not need to hold patents — they need to procure and apply technology intelligently. And News Corp demonstrates exactly that path: in early 2026 the group acquired the Symbolic.ai platform rather than building its own AI IP. Problem solved?

Not quite. Acquisitions shift the ownership of the IP — they do not create an in-house capability to extend, recombine and defend it. A company that buys AI capability still needs to staff the team that can integrate it, extend it, and keep pace with the frontier. The Patent-to-Productivity Index is not a count of patents. It is a proxy for the depth of internal capability — the kind of capability that shows up years later as differentiated product velocity.

Bick, Blandin, Deming and colleagues provide the first hard data on the AI adoption gap in an NBER study: 5.2 percent of US working hours are AI-assisted — double the rate in the UK and Sweden, triple that in Germany. The estimated additional productivity gain on the US side since 2022: 3.2 percentage points.

The OECD confirms the mechanism: firms that patent in 4IR technologies show measurably higher productivity. Hall and Ziedonis show that patent quality correlates positively with market capitalisation. Patent activity is not a vanity metric. It is a leading indicator of competitiveness.

The German long-form edition contains the complete argument, 11 sources, the Patent-to-Productivity Index scorecard for the five referenced houses, and the full reverse-acquihire analysis. An English long-form edition is in preparation; it will replace this summary once published.