Over the past four months, I analysed more than 5,000 news and media domains across 99 countries. I was mapping out how the media landscape responds to artificial intelligence — who is opening up, who is shutting down, and what the consequences are for anyone just trying to understand the world.
I am Danish. I have spent 18 years in Danish digital media.
When I reached the Danish data, I felt something I had not expected.
Unease.
Like many major public service and national publishers globally, DR (the Danish Broadcasting Corporation), Politiken and TV 2 have chosen to block all major AI crawlers. BT and Berlingske block some. Business media outlets in this dataset block in roughly three out of four cases.
This is a defensible choice — and one that many peers in the UK, US and elsewhere have also made. It is legally sound. It is the only credible negotiating position against companies that have built billion-dollar industries on others' content without asking permission, without paying, and without citing the source. The question this report asks is not whether blocking is wrong, but whether it is sufficient.
But their content is not disappearing from AI responses. It is being replaced.
By Reddit threads, small independent blogs, and content farms without editorial oversight. As quality media outlets shut out crawlers, sources like Reddit and other user-generated platforms fill the gap in AI answers. When someone asks ChatGPT or Gemini about Danish topics, the authority no longer belongs to Politiken or DR. It goes to those who have not blocked crawlers — often without journalistic standards, transparency, or accountability.
This is a transfer of authority that was not voted on.
Jyllands-Posten stands out as one of the only major Danish newspapers that is fully open to AI and has implemented llms.txt — the file that tells AI systems what your content policies are. But the more significant Danish finding is not about who is open. It is about who has adopted ai.txt.
Denmark has ai.txt on 25% of its publisher domains. The global average is 1.9%. Of the roughly 75 publisher domains globally that have implemented ai.txt, 60 are Danish — and 42 of those are restrictive. Denmark has not found a way past the block. It has found more ways to formalise it — more precisely and more legally documented than any other Nordic market.
Denmark is not an exception. Across all five markets we analysed, the dominant reflex has been the same: close the door.
| Country | Domains | Block all AI | Allow all | llms.txt |
|---|---|---|---|---|
| Denmark | 240 | 80 (33%) | 119 (50%) | 45 (19%) |
| Norway | 156 | 117 (75%) | 35 (22%) | 4 (3%) |
| Sweden | 130 | 97 (75%) | 29 (22%) | 8 (6%) |
| Finland | 97 | 70 (72%) | 22 (23%) | 4 (4%) |
| Iceland | 29 | 5 (17%) | 18 (62%) | 2 (7%) |
In Norway, VG, Bergens Tidende and Stavanger Aftenblad — three of the country's largest outlets — all block every major AI crawler. The few that remain open, like Filter Nyheter and ITavisen, are smaller independent voices.
In Sweden, TV4, Eskilstuna-Kuriren and Norran block all AI. The pattern holds across the country regardless of what you call yourself.
In Finland, Suomen Kuvalehti and Apu block. MTV Uutiset and Maaseudun Tulevaisuus are among the few that remain open.
Iceland is the exception that proves the rule.
Of the 29 Icelandic domains we were able to analyse, 18 — 62% — allow all AI crawlers. Only 5 block entirely. Morgunblaðið and Fréttablaðið, two of the country's most established outlets, are both fully open. Iceland has not opted out. It has, for now, chosen visibility.
Whether that turns out to be commercially smart or strategically naive, we do not yet know. What we do know is that when an AI system answers a question about Nordic affairs, Icelandic voices are structurally more likely to be heard than Norwegian, Swedish or Finnish ones.
It is about whether this is an active choice — or simply what is happening.
The data clearly shows that those with the best content are not the ones being cited. The ones who are available are cited.
If blocking is an intentional, strategic choice, it should be stated as such. If it is not, then this is precisely the moment to decide what role Nordic journalism should play in the AI answer layer over the next decade.
The analysis that follows presents the full data behind these findings — broken down by country, by metric, and by what the path forward could look like.
Stratechmedia ApS · Technical analysis of 689 publisher domains · May 2026
Over the past four months, Stratechmedia analysed 689 publisher domains across Denmark, Norway, Sweden, Finland and Iceland. We measured whether each domain had the technical infrastructure to operate in an AI-driven information environment — and whether it had declared a policy for how AI agents should treat its content.
The results show a region that has responded to AI with a clear, consistent strategy: blocking. More than half of all Nordic publishers block all AI crawlers. In Norway and Sweden, three in four do. What the data also shows is what that strategy does — and does not — achieve.
A robots.txt block prevents AI crawlers from accessing your content from the moment it is set. That is its function, and it works. What it does not do:
The EU AI Act's transparency provisions, which apply from August 2026, require publishers to document how their content has been used by AI systems. A block alone does not produce that documentation.
| Metric | Nordic Avg | Global Avg | Delta |
|---|---|---|---|
| AI Readiness Score | 40.1/100 | 38.2/100 | +1.9 |
| AI Act Compliance Score | 17.0/100 | 4.4/100 | +12.6 |
| llms.txt implemented | 8.9% | 13.1% | −4.2pp |
| Block-all-AI policy | 52.8% | 21.4% | +31.4pp |
When large language models began scraping the web at scale in 2022 and 2023, Nordic publishers reacted faster than almost any other region. Press organisations issued statements. Trade bodies published guidelines. robots.txt files were updated overnight.
The result: 75% of Norwegian and Swedish publishers now block all AI crawlers. Finland follows at 72%. This is the highest concentration of total AI-blocking in our global dataset of 99 countries.
The intention is legitimate. These publishers are protecting intellectual property that took decades to build. But blocking carries costs that are rarely quantified.
In the AI answer layer — which now handles a growing share of information queries globally — you are either cited or you are invisible. There is no middle position. A competitor with weaker journalism but better infrastructure becomes the default source in a discovery layer your audience is already using.
If blocking is the visible story in Nordic media, the more important story is what sits behind it — or, in many cases, what does not.
A robots.txt block is easy to see. It is a public act of refusal. But what determines whether a publisher can do anything more nuanced than simply say no is a quieter layer of infrastructure: structured metadata, policy files, paywall signals, and machine-readable declarations of what AI systems may and may not do.
That is where the Nordic gap appears most clearly.
AI systems do not just read words. They read signals about what those words are.
JSON-LD structured data is one of the primary ways a publisher tells machines what a piece of content is, who published it, when it was published, and what type of page it is. Across the Nordic region, 50% of Danish publishers have implemented JSON-LD — the highest in the region. Norway sits at 31%, Sweden at 41%, and Finland at 30%.
But the more meaningful signal is not just whether structured data exists. It is whether journalism is identified as journalism.
The NewsArticle schema — which distinguishes a reported article from a generic web page — appears in only 9% of Danish publishers and in none of the Norwegian, Swedish, or Finnish publishers scanned.
That matters because this is how a publisher tells an AI system: this is not just content. This is journalism, with an editor, a publication date, and a rights holder.
Beyond robots.txt, two policy files are beginning to define how publishers communicate with AI systems: ai.txt and llms.txt.
ai.txt allows publishers to declare, in machine-readable form, what AI systems may and may not do with their content. In theory, it makes it possible to say something more nuanced than a blanket yes or no — for example, no training, but citation allowed. Globally, only 1.9% of publisher domains in the dataset have implemented ai.txt. In Denmark, 25% have.
llms.txt serves a related but broader role. It gives AI agents a policy statement covering usage conditions, licensing information, and commercial terms. Across the Nordic region, 8.9% of publishers have implemented it. Denmark leads at 19%, while Norway, Sweden and Finland sit between 3% and 6%.
Together, these files are what separate a block from a policy. robots.txt can say “go away.” ai.txt and llms.txt make it possible to say something more precise: what is prohibited, what is permitted, and where the commercial conversation begins.
There is another layer of missing infrastructure that matters less to lawyers and more to how journalism actually appears in AI products.
When an AI assistant decides whether it can read something aloud, summarise it, or hand users off to the original source, it relies on small technical signals that most readers never see. In the Nordic data, those signals are almost entirely absent.
No publishers in the Nordic sample use the tags that explicitly mark text as suitable to be read aloud by assistants. Only around 6% use clear machine-readable indicators showing whether an article is free to access or behind a paywall. And almost none combine their robots.txt rules with page-level signals that explain restrictions on individual articles.
This means that even where publishers are trying to control AI access, they often provide very little guidance about how their content should be handled once it is seen.
The result is a region that is relatively good at blocking, but much weaker at declaration.
Once those gaps are visible, the country patterns become easier to read. The Nordic region is not behaving as one market. It is behaving as five variations of the same dilemma.
| Metric | Denmark | Norway | Sweden | Finland | Iceland |
|---|---|---|---|---|---|
| AI Readiness Score | 44.8 | 40.4 | 48.5 | 43.9 | — |
| AI Act Compliance | 21.8 | 36.2 | 6.0 | 0.4 | — |
| llms.txt | 19% | 3% | 6% | 4% | 7% |
| Block-all-AI | 33% | 75% | 75% | 72% | 17% |
| Allow-all | 50% | 22% | 22% | 23% | 62% |
| NewsArticle schema | 9% | 0% | 0% | 0% | — |
| Score = 0 on compliance | 48% | 4% | 75% | 98% | — |
Denmark is the clearest outlier in the Nordic data — and the reason is ai.txt. While the global average for ai.txt adoption is 1.9%, Denmark sits at 25%. Of the roughly 75 publisher domains globally that have implemented ai.txt, 60 are Danish.
In most of those cases, ai.txt is not being used to open up but to refine an already restrictive position. 42 of the Danish ai.txt domains combine it with blocking rules in robots.txt, using ai.txt to spell out in more detail what is not allowed.
So Denmark does not stand out because it has embraced AI more openly than its neighbours. It stands out because it has gone further in turning refusal into explicit, machine-readable policy. It is not the most open market in the region. It is the most precise in how it says no.
These findings are not a verdict on any individual publisher, but an invitation to revisit whether the current stance matches the role they want in the AI answer layer.
Norway has the highest average AI Act compliance score in the region — 36.2 — but that score is being earned primarily through blocking. 75% of Norwegian publishers block all AI crawlers, and only 3% have implemented llms.txt. VG, Bergens Tidende and Stavanger Aftenblad all block every major AI crawler.
What Norway shows is that a market can look disciplined from a compliance perspective while still building very little infrastructure for negotiation, licensing, or selective access. It has built a legal shield, but not much beyond it.
Sweden has the highest average AI Readiness Score in the region — 48.5 — yet 75% of Swedish publishers block all AI crawlers, and 75% score zero on EU AI Act compliance. TV4, Eskilstuna-Kuriren and Norran are all blocking.
Sweden’s strength lies in general technical maturity: decent metadata, structured pages, and a modern publishing stack. But that does not translate into AI-specific preparedness. Sweden looks ready in the old sense of the web. It looks much less ready for the answer layer that is replacing it.
Finland is the starkest case in the dataset. 70 of 97 publishers block all AI crawlers, and 98% score zero on EU AI Act compliance. None of the Finnish publishers scanned have editorial policy pages, none use NewsArticle schema, and only 4% have llms.txt. Suomen Kuvalehti and Apu block all AI; MTV Uutiset and Maaseudun Tulevaisuus are among the few still open.
Finland has built a wall and then largely stopped. There is almost no machine-readable infrastructure behind it.
Iceland presents a strikingly different picture. 62% of Icelandic publishers allow all AI crawlers, and only 17% block entirely. Morgunblaðið and Fréttablaðið, two of the country’s most established outlets, are both fully open. Iceland has not opted out. It has, for now, chosen visibility.
Whether that proves commercially smart or strategically naïve remains to be seen. What is already clear is that when an AI system answers a question about Nordic affairs, Icelandic voices are structurally more likely to be present than Norwegian, Swedish or Finnish ones.
The question is not whether Nordic publishers should protect their content. They should.
The question is whether protection, on its own, is enough.
Right now, the Nordic response to AI is defined by refusal. In many cases that refusal is justified, legally coherent, and commercially understandable. But a defensive position is not the same as a long-term strategy.
Blocking tells AI systems what they may not do. It does not, on its own, define what publishers want to happen instead.
That is the gap between fortress and ownership.
The first shift is from a binary block to an explicit policy.
robots.txt can only say yes or no. It can tell a crawler to enter or leave. But it cannot express nuance: no training, yes to citation; no scraping, yes to summaries with attribution; commercial use only under licence.
That is where ai.txt and llms.txt become important. They allow a publisher to move from a blunt refusal to a structured position. Not openness without limits, but terms. Not access without control, but machine-readable conditions.
For the majority of Norwegian, Swedish and Finnish publishers currently operating with a full block, this does not mean opening the door. It means stating, clearly, what sits behind it and under which conditions anyone may engage with it.
The second shift is from content that is merely published to content that is clearly identifiable as journalism.
Without NewsArticle schema, editorial metadata, and clear authorship signals, many news pages appear to AI systems as little more than generic web documents. A strong brand may be obvious to a human reader. It is much less obvious to a machine if the page does not say so in a language machines understand.
This is why the structured data gap matters. It is not just a technical oversight. It affects whether journalism can be recognised, attributed, ranked, and cited as journalism.
If Nordic publishers want their authority to survive in AI interfaces, they need to make that authority legible.
The third shift is from passive blocking to active oversight.
A robots.txt file tells a crawler what it is allowed to do from this point forward. It does not tell a publisher what happened before, who has already accessed the content, or whether any actor is respecting the declared rules in practice.
That is where logging, monitoring, and documentation become essential.
As AI regulation matures — including the EU AI Act's transparency requirements — publishers will need more than principles. They will need records. Who came. What they accessed. Whether the declared rules were followed. Without that, a policy remains a statement of intent rather than an enforceable position.
Ownership does not mean giving AI systems free access to journalism.
It means being able to decide, in a technically and commercially credible way, what can be used, what cannot, what must be licensed, and how those rules are communicated. It means replacing improvised refusal with policy, metadata, and monitoring.
That is what much of the Nordic market is still missing.
The strongest publishers in the region already own the journalism. The next challenge is owning the terms under which that journalism appears — or does not appear — in the AI answer layer.
This report is not meant to be read as a verdict. It is meant to be used as a plan.
For a Nordic publisher, the next 12–18 months can be surprisingly concrete. The task is not to “solve AI”, but to decide a stance and encode it in the infrastructure that already exists.
Before touching any files, write down the actual stance:
Without this sentence, every technical change will be inconsistent.
Ask (or your vendor), very specifically:
This is a diagnostic exercise, not a transformation project.
Make sure the file matches what you decided in step 1. If you block, block clearly and consistently. If you allow some access, make that explicit and remove accidental contradictions.
For most medium-to-large publishers, a minimum viable stack over the next year is:
This does not commit you to any specific deal. It simply makes your position legible.
Extend your existing structured data so that all articles carry:
This is a small change for a developer and a large change for how AI systems understand your site.
Set up logging of AI-related user-agents and access patterns. Keep that data. It will be needed for any serious conversation about compliance, and it will be your evidence base in future negotiations or disputes under the EU AI Act.
As the AI Act's transparency provisions come into force, revisit your stance and signals. The goal is simple: no surprises — for regulators, for partners, or for your own newsroom.
The data from 689 Nordic domains tells a coherent story: a region that reacted to AI with speed and decisiveness, and built a legal defence without completing the commercial position.
Nordic publishers already own the journalism. The next question is whether they also want to own the terms under which that journalism appears — or does not appear — in the AI answer layer. That choice will not be made by regulation alone. It will be made by the signals that sit in robots.txt, ai.txt, llms.txt and the metadata on every article.
The August 2026 deadline is the moment when regulatory pressure and product reality converge. Publishers who arrive there with a clear stance, a basic policy stack, and monitoring in place will at least be at the table. Those who arrive with only a block will still be affected by AI — just with less visibility and less leverage.