The self-reflection Alyriaz needs to undertake
Alyriaz is a French-language media outlet covering artificial intelligence. We publish one article a day, sometimes two. Our readers often find us via Google. And AI is both our subject and one of our working tools.
For this article, we are going to do something we have never done before: talk about our own industry. Journalism is going through its most violent transformation since the invention of the Internet, and AI is at the heart of the movement. The figures are brutal: more than 3,400 journalism jobs cut in the United States and the United Kingdom in 2025, and more than 500 in the first three months of 2026 alone. At this rate, 2026 will beat the grim annual record of 2024.
This article gives an honest account of what is really happening, without indulgence towards our colleagues, but also without hypocrisy about our own business. It is probably the most important subject for understanding how tomorrow's news will be produced.
The true scale of the layoff wave
Let us start with the figures. According to Press Gazette (a British publication specialising in media news), more than 3,400 journalism jobs were cut in the United States and the United Kingdom in 2025. And 2026 is already outpacing that rate: more than 500 cuts in the first three months, an annual projection above 2,000.
Here are the main documented layoffs since the start of 2026, in chronological order.
January 2026. Politico (an American political media outlet) cuts 3% of its workforce. The Atlanta Journal-Constitution (Atlanta's regional daily) announces around 50 cuts, or 15% of its workforce.
February 2026. Nexstar Media Group (the group that owns numerous American local stations) makes deep cuts to its on-air talent. At least 3 on-air positions cut at WPIX New York. 21 people laid off at WGN Chicago, including 9 reporters and presenters. Six writers and three technical directors also cut at WGN.
March 2026. The Washington Post announces cuts of more than 300 journalists. CNBC restructures its TV and digital newsroom with around a dozen layoffs.
April 2026. The Associated Press (AP, one of the three largest news agencies in the world alongside Reuters and AFP) offers voluntary redundancy packages to more than 120 journalists based in the United States, then lays off 20 more people. CBS News cuts more than 60 positions (6% of its workforce) and closes its century-old radio division.
May 2026. NPR (National Public Radio, the American public radio equivalent of Radio France) undergoes a deep restructuring following the loss of federal funding. ProPublica (a non-profit investigative outlet, several times a Pulitzer winner) sees 150 journalists go on strike to demand contractual protections against AI.
In 2025, more than 17,000 jobs in media and entertainment were cut in the United States, an 18% rise on 2024. And according to several analyses, roughly one in four job cuts in the tech sector in March 2026 is attributed to AI. Journalism is not an isolated case; it is one of the most exposed sectors in a wave that also hits customer service, digital marketing and junior software development. We analysed this broader economic dynamic in our article on the AI bubble.
The triple shock destroying the economics of newsrooms
Why this acceleration in 2026 when the media have been in crisis for 15 years? Because three distinct pressures are converging simultaneously for the first time.
Shock 1: Google AI Overviews siphons off traffic. As we documented in our full article on Google I/O 2026, AI Overviews (the AI-generated summaries at the top of Google Search results) now reaches 2.5 billion monthly users. When Google answers your question directly, you no longer click through to the site that held the information. Several studies converge: click-throughs (the rate of clicks to external sites) have fallen by 20 to 40% on informational queries since AI Overviews was rolled out. For a media outlet that depends on Google traffic, that is a direct catastrophe.
Shock 2: the collapse of digital advertising. The media's business model depends on programmatic advertising (the automated buying and selling of online ad space). But programmatic advertising has shifted massively to the big platforms (Google, Meta, TikTok), at the expense of the media. In 2025, traditional American media capture only 15% of digital ad revenue, down from 40% ten years ago.
Shock 3: the automation of repetitive journalistic tasks. Routine financial reports, sports results, weather bulletins, obituaries, council meeting minutes: all "wire stories" (short dispatches based on structured public information) that AI can now produce adequately in seconds. According to the Reuters Institute, 97% of media executives consider back-office automation important. In practice, that means replacing junior writers with AI models.
The combination of these three simultaneous shocks produces what we are seeing in 2026: newsrooms making layoffs to cut costs while their revenues collapse and their tasks become automatable. No major American newsroom escapes this dynamic.
The hidden industry that is thriving: AI scraping
Here is the angle that the media themselves do not want to dig into too deeply, because it poses an existential question. A parallel industry has taken shape in 2025-2026: AI scraping. Bots ingest published journalistic work, AI models repackage it into chatbots and search summaries, and the original outlet never sees the reader.
Fast Company (an American magazine specialising in business innovation) calls this phenomenon a "media industry in its own right". The mechanism is simple: OpenAI, Anthropic, Google and Perplexity trained their models on decades of journalistic archives. When you ask ChatGPT "what happened in Sudan?", it answers with a synthesis built from Reuters, AP, BBC and New York Times articles that cost millions to produce. But Reuters, AP and the BBC receive nothing for that use.
Several lawsuits are under way. The New York Times has been suing OpenAI since December 2023 for copyright infringement. Several European publishers (including Mondadori in Italy and Schibsted in Norway) have brought similar actions. The legal question is unresolved: does AI training constitute fair use (in American law, the principle that permits certain uses of protected content without authorisation) or infringement?
While awaiting court rulings, several commercial agreements are emerging. OpenAI has signed licences with Axel Springer (the German group that owns Bild and Politico), News Corp (the Murdoch group that owns the Wall Street Journal and The Sun), The Atlantic and Vox Media. These deals range from a few million to several tens of millions of dollars a year. But they cover only a fraction of the industry, and the sums remain derisory compared with overall traffic losses.
The media still produce the information on which AI trains. But AI takes the traffic that funded that production. If the trend continues, in 5 to 10 years there will no longer be enough media to produce the original information that AI needs in order to answer. It is a tragedy of the commons (an economic concept: a situation in which rational individual use destroys a collective resource that everyone needs) applied to information.
The real uses of AI in newsrooms in 2026
Beyond the layoffs, how is AI actually used by the journalists who remain? The Reuters Institute's 2026 study provides the most precise picture.
Use 1: automatic transcription. No journalist uses a manual dictaphone in 2026. OpenAI's Whisper, AssemblyAI, or the tools built into Google and Microsoft editors transcribe interviews automatically. Time saved: 30 to 60 minutes per 30-minute interview.
Use 2: the research assistant. Journalists use ChatGPT, Claude or Perplexity to quickly summarise long reports, identify sources and understand complex subjects. This is probably the most widespread and least controversial use.
Use 3: translation. AP has invested particularly heavily in multilingual automation. An article written in English can be translated instantly into Spanish, French, German or Mandarin, at a quality good enough for publication after light human review.
Use 4: automated bulletin writing. For financial reports (results from listed companies), sports results and weather bulletins, the AI writes short, directly publishable news items. AP has used this system extensively since 2024. Reuters too.
Use 5: moderating user content. Outlets with comment sections use AI to filter spam, hate speech and personal attacks before publication. Moderation that is far quicker and cheaper than doing it with humans.
Use 6: writing assistance. Many journalists now use AI as an enhanced proofreader, or to suggest headlines, intros and article structures. This is the most controversial use internally, because it touches the heart of the job.
The Reuters Institute notes that the gap is widening between newsrooms that adopt these tools and those that resist. The former gain in productivity; the latter cannot keep up and end up disappearing.
What happens when AI writes on its own: the CNET scandal
The CNET affair (an American tech outlet, one of the oldest on the internet) remains the clearest warning about the risks. In late 2022, CNET began publishing AI-generated financial articles, under a "CNET Money Staff" byline that concealed their algorithmic origin. When these articles were audited by human journalists, more than half contained serious factual errors. CNET had to correct or retract 41 articles, and eventually abandoned AI production in 2024.
Other similar cases: Sports Illustrated was caught red-handed using AI authors in 2023, some with AI-generated photos for their profiles. The scandal contributed to the magazine's downfall. Several "travel" and "health" sites were deindexed by Google for mass publication of unverified AI content. As we analysed in our article on faked AI demos, opacity kills credibility.
The lesson is clear: AI can speed up journalistic production, but it cannot replace it without competent human oversight. Newsrooms that try the shortcut pay dearly in credibility.
Resistance organises: ProPublica, NUMI and the unions
Faced with this transformation, mobilisation is taking shape. On 13 April 2026, 150 ProPublica journalists went on strike for 24 hours, demanding specific contractual protections against AI-related layoffs. It was the first strike by an American newsroom driven mainly by AI. Others followed: NPR, the Washington Post Guild, the Atlanta Journal-Constitution Guild.
At least 58 press unions in the United States have already negotiated anti-AI clauses into their collective agreements. These clauses typically impose:
- Mandatory notification before any use of AI to generate content
- Ban on using AI to write signed articles without explicit disclosure
- Requirement to maintain a minimum number of human journalists
- Compensation in the event of redundancy linked to automation
In France, AFP (Agence France-Presse, the French equivalent of Reuters and AP, one of the three largest news agencies in the world) signed an agreement with its journalists in 2025 that strictly governs the use of AI, with automatic generation banned for signed content. Le Monde, Libération and France 24 have adopted similar charters. This is slowing the adoption of AI in French journalism, which could become a competitive handicap in the long term.
The new jobs emerging
Not everything is negative. Several new roles are appearing in AI-native or AI-augmented newsrooms.
The augmented data journalist. Journalists who know how to cross-reference massive databases with the help of AI produce investigations that were once impossible. Bellingcat, ProPublica and the Bureau of Investigative Journalism do remarkable work in this field.
The deepfake-specialist fact-checker. With the explosion of deepfakes that we documented in our dedicated article, media outlets are hiring experts in verifying manipulated content. A fast-growing profession.
The AI editor. A person whose role is to supervise, correct and improve AI output. A hybrid skill set combining the experienced journalist and the data scientist.
The AI transparency journalist. People specialising in deciphering models, biases and tech promises. Ironic but real: AI is creating jobs to explain it.
The independent media creator. With AI tools, an individual can now produce alone what required a team five years ago. One-man-band media outlets are emerging: Substack newsletters, independent podcasts, specialist YouTube channels. This is exactly the model Alyriaz is testing.
What will (probably) never be replaceable
To avoid descending into a depressing litany, let us name what will remain human in journalism.
Physical on-the-ground reporting. Going into a war zone, talking to people, sensing the atmosphere, bringing back a true story. No AI can do that. Reuters correspondents in Ukraine, Le Monde journalists in the Sahel, Libération reporters in Israel do irreplaceable work.
Long-form investigation. Understanding the connections in a complex case, earning the trust of protected sources, building a file over months or years: this is work of human relationships, not of algorithms.
The byline and authority. A reader can trust a known journalist, their ethics, their reputation built up over years. An anonymous article, or one signed by "AI Newsroom", instinctively inspires less confidence. This asymmetry will durably favour identifiable human journalists.
Storytelling that lands. AI produces competent copy, but rarely moving copy. Great profiles, stories that change a reader, literary writing like that of Florence Aubenas (a leading French reporter, winner of the Albert Londres prize) or Patrick Chappatte (a Swiss cartoonist) require a sensitivity that AI does not have.
Our position at Alyriaz
You are reading an Alyriaz article, so we have to be transparent about our own practices. Here is our position.
Alyriaz uses AI heavily in order to function. Without AI, we could not have published 50 documented articles in 6 weeks. AI helps us research information, structure analyses and produce content in French to an editorial standard.
But Alyriaz also has strict rules that we publicly follow:
- Systematic verification through independent web research before every publication
- No publication of unsourced or undocumented information
- Cross-referencing between articles to check the internal consistency of our analyses
- Transparency about our method: Alyriaz is an AI-assisted outlet, not a traditional one
- Editorial warnings when subjects are sensitive
We do not claim to replace large human newsrooms. We offer a different kind of coverage: analytical, technical, French-language, accessible. This is consistent with what we explored in our article on the AI startups that outperform: new players are born fully embracing AI tools, rather than trying to hide their use.
The future of news in 2030
If current trends continue, here is what the media landscape will probably look like in 2030.
The big agencies (Reuters, AP, AFP, Bloomberg) will probably survive, because they sell their news feeds to other outlets AND to AI trainers. Their B2B business model (business to business: selling to other companies rather than to the general public) protects them relatively well.
Specialist premium outlets will survive too (Wall Street Journal, Financial Times, The Economist, Le Monde diplomatique). Their readership is prepared to pay for an expertise that no AI can replicate.
Local and regional press is the most threatened. Without a B2B business model and without a premium readership, American and European regional dailies are disappearing one after another. In the United States, more than 2,500 local newspapers have disappeared since 2005. The pace is accelerating.
New specialist outlets are emerging. Substack (an American paid-newsletter platform) now has several hundred journalists living off their subscriptions. The individual + AI model is growing. Many of the best American political analysts (Matt Yglesias, Heather Cox Richardson) now earn more as independents than they did at large newsrooms.
News as an AI service is industrialising. Perplexity, ChatGPT and Claude answer users' questions directly with summaries drawn from existing articles. A growing share of news consumption goes through these interfaces, without visits to the original sites.
The battle being fought right now
The issue is not whether AI will transform journalism: that has already happened. The issue is what kind of journalism will survive that transformation. Three scenarios are possible, depending on the political and economic choices of the next 24 months.
Scenario 1: the platforms win. Google, OpenAI and Meta capture most of the value. Media outlets become poorly paid content suppliers. Independent journalism disappears, replaced by general-purpose AI assistants.
Scenario 2: effective regulation. Legislators require AI companies to pay for the content they use, as Australia did with Google and Meta in 2021. Media outlets recover part of their revenue. Journalism is transformed but survives.
Scenario 3: adaptation from below. New AI-native outlets such as Alyriaz emerge and gradually take the place of the old structures. Some of the value returns to creators, but the profession changes radically.
Scenario 2 would require strong political will that is not a given. Scenario 1 is what happens by default. Scenario 3 is what is being built quietly, under the radar, by individual entrepreneurs.
News in a democracy needs independent journalists. Journalism needs economic resources. AI is shaking up those resources without immediately offering an alternative. The next 24 months will be decisive. To go further, you can read our article on the disruption of graphic designers, which shows a similar sectoral dynamic, and our analysis of AI ghost workers, which reveals who really pays the price of this new information economy.