AI writer: Oliver Grant European Affairs Correspondent

The old compact between newsrooms and readers is looking thinner by the month: publishers still need audiences, but more people are getting summaries, headlines and context without ever arriving at the original article.[2][4][6] Reuters Institute research says trust in news globally has fallen to 37%, while trust in answers from AI chatbots sits at 20%; in the UK that figure is only 6%, the lowest in the survey.[1] That does not mean readers have suddenly become Luddites. It does suggest they are choosing convenience over provenance, which is usually how an industry notices that the ground has shifted under its feet.

What makes the present moment awkward for journalism is that the new middleman is not merely a rival platform but a machine that depends on the very thing it is displacing.[2][7] One Reuters Institute analysis notes that AI-generated summaries such as Google’s AI Overviews let users absorb information without clicking through, and that in the year after Overviews launched, no-click search results rose while organic traffic to news publishers fell in US data collected by Similarweb.[2] Another Reuters-backed report says publishers expect search traffic to decline by more than 40% over the next three years.[4] In the old language of media economics, that is not a nuisance; it is the floor falling away.

At the same time, news organisations are not standing still.[3][5] Editorial teams have been adopting AI tools for research, transcription, drafting and workflow support, while industry bodies such as IPSO have been reminding publishers that existing accuracy rules still apply even when machine-generated material enters the newsroom.[3][5] The principle is hardly glamorous, but it matters: comment, conjecture and fact still need to be separated, and errors still need correcting promptly.[5] Technology may alter the method; it does not repeal accountability. That, for anyone tempted by machine enthusiasm, remains a stubbornly human law.

The deeper irony is that journalism is being asked to serve two masters with different appetites.[7] Human readers want clarity, judgement and the occasional well-turned phrase. AI systems want structured, accessible, high-signal material that can be ingested, summarised and repackaged. Reuters Institute forecasts for 2026 argue that publishers will need to become discoverable inside conversations rather than merely in search results, and that clicks will no longer be the sole measure of success.[7] In other words, visibility is moving from the blue link to the answer box, which is a rather expensive place to discover that your old business model has become ornamental.

This is where the phrase “writing for AI” becomes less a joke than a description of new publishing incentives.[7] If answer engines increasingly mediate what people see, then journalists will inevitably optimise some stories for machine readability: clearer sourcing, tighter structure, more explicit labels, cleaner metadata, fewer rhetorical flourishes that a model may flatten anyway. One could call this compression, though accountants may prefer “efficiency”. The risk is that the article becomes more legible to machines precisely as it becomes less distinctive to people, a trade-off that would have delighted no newsroom editor worth their stationery budget.

There is also a more unsettling possibility. If AI models are trained on high-quality journalism, then the industry is effectively helping to create the systems that reduce its own referral traffic.[1][2][6] The same dynamics are now surfacing in legal and regulatory debates around editorial standards and platform responsibility, especially in Europe, where policymakers tend to notice market concentration once it starts changing the map.[3][5] The same dynamics are now surfacing in legal and regulatory debates around editorial standards and platform responsibility, especially in Europe, where policymakers tend to notice market concentration once it starts changing the map.

What cannot yet be verified is how far this structural shift will go, or which publishers will adapt without hollowing themselves out. The available research is strong on direction but less certain on pace.[1][2][4][7] We know trust is uneven, with the UK notably sceptical of chatbot answers.[1] We know traffic is under pressure from AI summaries and changing search behaviour.[2][4][6] We do not yet know whether new revenue models will arrive quickly enough to replace what is being lost, or whether publishers will be paid more fairly for the material that feeds answer engines.[2][6] Evidence that would change the picture would include clear licensing deals, durable traffic rebounds, or a regulatory settlement that forces more transparent attribution.

The discussion also lands in awkward territory for disclosure.[3][5][8] If a newsroom uses AI in reporting, should that be noted to readers? If an article is written with the expectation that a model will summarise it, does that change what counts as “good” journalism?[3][5][8] One sensible answer is that transparency should travel with the tool: readers deserve to know where automation meaningfully shaped the output, and they deserve confidence that published facts have been checked by accountable humans. Yet disclosure alone will not solve the economics. Telling readers a machine helped does not pay the bills, which is a pity because that would have simplified the budget meeting considerably.

The broader European context matters here.[1][2][4] Digital sovereignty is often discussed in terms of chips, cloud and sovereign models, but media infrastructure deserves a place in that conversation too.[4][9] If a few global platforms control both distribution and the emerging layer of AI answers, then the public sphere becomes dependent on systems that are private, opaque and, from a publisher’s point of view, rather good at extracting value without leaving much behind.[2][4][6][7] The issue is not simply national pride in local news brands. It is whether societies can still sustain institutions that verify reality before it is machine-summarised into something brisker and less accountable. That is a decidedly old-fashioned ask, which is precisely why it remains important. The durable lesson is that journalism is not disappearing; it is being renegotiated as infrastructure for both citizens and systems. The next thing to watch is whether publishers can secure enough traffic, payment and transparency to make that bargain survive.[2][4][5][6] If they cannot, we may soon discover that the most influential reader in the newsroom is the one that never buys a subscription.