By Stuart Kerr, Technology Correspondent, LiveAIWire
In 2026, the Nieman Journalism Lab published a prediction that AI-written content would outpace human-produced content not just in spammy corners of the web but across mainstream channels where people search, scroll, and learn. The prediction describes what is already partially observable. A 2025 Ahrefs study found that 74.2 percent of newly published web pages contain AI-generated material. Large-scale web corpus analyses estimate that 30 to 40 percent of all active web text now originates from AI-generated or AI-edited sources. The rate of AI bots bypassing voluntary access restrictions has quadrupled in six months, from 3.3 percent to 12.9 percent, according to the Open Markets Institute’s April 2026 report on AI content licensing. The collapse in human website traffic is costing publishers billions in lost revenue and leading to ongoing journalism layoffs. Researchers are warning of what one prominent analysis calls sloppification: a slow but accelerating degradation of AI output quality as the human-generated content that AI depends on becomes less available.
The quality collapse is simultaneously an editorial crisis, a professional identity crisis, and a structural economic crisis for the knowledge industries that produce the content AI systems depend on to function at all. Understanding how these dynamics interact requires engaging with evidence that the industry has been slow to confront.
What Sloppification Actually Means
The model collapse hypothesis, documented by Ilia Shumailov and co-authors in Nature in 2024, postulates that AI models trained increasingly on AI-generated output progressively degrade. The mechanism is intuitive: AI-generated content is statistically smoothed relative to the human-generated content it is based on, removing the diversity of expression, the edge cases, the minority-language usage, and the novel turns of phrase that make human language varied and informative. When subsequent AI models are trained on this smoothed content, the smoothing compounds. The outputs become more formulaic, more average, and less capable of handling the specific, the unusual, and the genuinely novel.
The Open Markets Institute’s report was more direct about the immediate economic dimension: AI companies are cannibalising the content they depend on to function. The AI content licensing market that was supposed to address this by compensating publishers for training data use is, according to the report, structured in ways that repeat the mistakes of the social media era — creating value for platforms while extracting it from the content creators whose work makes the platforms valuable. The warning about sloppification is not that AI will immediately produce terrible content. It is that the trajectory leads toward progressively less informative AI training data, producing progressively less capable outputs, in a downward spiral that the current market structure has no mechanism to arrest.
What Is Happening in Newsrooms
The Reuters Institute’s March 2026 AI and the Future of News conference brought together journalists and researchers whose collective assessment was more nuanced than either AI-utopian or AI-catastrophist framings suggest. Muck Rack’s State of Journalism 2026 report, drawn from more than 1,000 newsroom staff surveyed in March 2026, found that 82 percent already use AI tools regularly. ChatGPT leads adoption at 47 percent, followed by Google Gemini at 22 percent. Weekly AI usage among general news audiences nearly doubled in a single year, jumping from 18 to 34 percent per Reuters Institute research.
The uses journalists report as most valuable are those where AI expands rather than replaces human capability: data journalism is accelerating, fact-checking workflows are getting faster, and routine commodity content — earnings summaries, weather updates, sports scores — is increasingly automated, freeing human reporters for work requiring genuinely human judgement. The uses generating documented quality problems are those where AI is used to replace rather than augment human editorial work — producing the appearance of journalism without the processes that make journalism valuable.
In Brazil, fact-checker Aos Fatos found that 16 percent of the 619 claims their team checked in 2025 involved AI-generated content, compared with 7 percent the previous year. The Reuters Institute conference’s fact-checking panel concluded that AI has made the job of fact-checking more demanding, not less — forcing teams to respond to higher volumes of AI-generated misleading content while using AI tools to help them do so. One deepfake detection firm recorded 3,165 deepfake incidents in March 2026 alone, compared to just four in January 2020.
The Professional Standards Dimension
The quality collapse is not primarily a journalism problem. It is a knowledge work problem. The same dynamics playing out in journalism are playing out in legal work, accounting, medical documentation, academic research, marketing, and any professional domain where the deliverable is text. The question in each of these domains is not whether AI will be used — 82 percent of journalists already use it, comparable figures apply across professional services — but whether the professional standards and editorial processes that distinguish valuable professional output from AI-generated volume will be maintained.
The professional identity dimension is acute. Professionals who spent years developing expertise in writing, analysis, and professional judgement are being asked to evaluate, edit, and approve AI-generated content produced without that expertise. When the AI output is good enough — within the tolerance for error of most use cases — the professional’s role shifts from producer to evaluator. When it is not good enough, the professional must identify what is wrong, which requires maintaining the very expertise that would have been needed to produce the original work. Organisations reducing junior hiring in response to AI productivity gains are the ones most likely to face expert evaluator shortages when AI errors require human correction.
The Feedback Loop No One Is Managing
The Red Line Project’s March 2026 analysis of AI and declining news traffic identified a structural circularity the market has no current mechanism to address. Journalism that AI depends on to remain consistent and accurate cannot be produced in a newsroom without reporters. The quality of AI responses will collapse if the system on which it depends — the production of high-quality human-generated content — fails. Yet the economics of AI-generated content creation are eroding the revenue streams that fund human content production, creating exactly the conditions in which that system will fail.
Publishers have responded with content strategies emphasising distinctiveness — original reporting, contextual analysis, and human-centred stories that AI chatbots cannot commoditise. Google search traffic has declined 33 percent globally and 38 percent in the United States, according to Chartbeat data from 2,500 sites. Service journalism and evergreen content have been hit hardest. The publishers best positioned are those who have identified content types where human expertise is irreplaceable and invested in those, rather than those trying to compete on volume with AI systems that can produce indefinitely at near-zero marginal cost.
What Organisations Can Actually Do
The organisations generating consistent quality from AI-augmented workflows share identifiable characteristics. They have made explicit decisions about which tasks AI handles and which human professionals retain ownership of. They have maintained the editorial processes — review, fact-checking, editorial judgement — that separate professional quality from volume production. And they have invested in the human expertise required to evaluate AI output rather than treating AI adoption as a headcount reduction strategy.
Whether the quality gap between human-produced content that maintains professional standards and AI-generated volume will become sufficiently visible to readers and clients to create sustainable market differentiation is the critical question of the next several years. The evidence that it will is most visible in the domains where professional liability, regulatory oversight, and reputational stakes create economic incentives for quality that volume production cannot satisfy. For readers following AI’s impact on professional work, LiveAIWire’s coverage of AI adoption fatigue and developer burnout and our analysis of the attention economy meets AI addresses the parallel dimensions of how AI is changing what professional work actually involves.
The Economic Incentives Driving Volume Over Quality
The economic structure of AI content production creates systematic pressure toward volume over quality in ways that are not self-correcting. Advertising-funded media has always faced the same tension — more content generates more page views, which generates more advertising revenue, regardless of content quality — but AI has dramatically reduced the cost of producing content, which shifts the equilibrium point toward even higher volume production. A media organisation that could previously produce 20 articles per day with a human newsroom can now produce 200 with the same number of people, if it is willing to accept AI-generated content with lighter editorial oversight. The revenue per article falls as supply increases and audiences fragment. The cost per article falls faster, creating a short-term economic logic for volume expansion that the long-term quality degradation does not immediately reverse.
The organisations that are breaking out of this dynamic are those that have made a strategic commitment to distinctiveness that overrides the short-term volume logic. The Reuters Institute’s January 2026 Trends and Predictions report found that news executives are increasingly planning to shift editorial priorities toward original reporting, contextual analysis, and human-centred stories that AI chatbots cannot commoditise. The shift is real in strategy. Whether it is real in practice — whether the economic pressure of declining traffic and advertising revenue allows organisations to invest in the expensive content that distinctive journalism requires — will determine whether the quality collapse is arrested or continues.
The Model Collapse Timeline
The question of how quickly sloppification produces measurable degradation in AI output quality is the subject of active research that does not yet have definitive empirical answers. The theoretical model collapse result demonstrates that degradation is inevitable given sufficient AI-on-AI training. The empirical question is the timescale. AI companies maintain curated training datasets that include high-quality human-generated content from before the AI content flood, which slows the degradation. But the proportion of new high-quality human-generated content available for training is declining as human writers exit markets disrupted by AI, which accelerates it. The researchers most closely watching this dynamic estimate that measurable output degradation at the system level is a matter of years rather than decades — but the uncertainty in that estimate is large enough that it should not be used as grounds for complacency. The economic incentives that are driving AI content volume are creating the conditions for model collapse whether or not any individual actor intends that outcome.
What the Evidence Says About What Works
The organisations that have maintained quality through AI adoption share a set of practices that the research is beginning to identify consistently. They treat AI as a production tool for first drafts and a research assistant for evidence gathering, rather than as an editorial system that can make the judgements about what matters, what is accurate, and what is worth publishing. They have maintained the human editorial roles that apply those judgements — not as a compliance requirement but as a genuine quality control function that the AI cannot perform. And they have been honest with their audiences about what AI contributes to their work and what the human editorial process adds, which builds the trust that distinguishes valuable professional output from the AI-generated volume that surrounds it. LiveAIWire’s coverage of generative AI in journalism and our analysis of why substance beats hype in AI product design provides the wider context of how AI is reshaping the economics and practice of content-dependent professions.
The sloppification warning from the Open Markets Institute is not a counsel of despair. It is a description of a trajectory that is not inevitable, because the market for quality — for content that is accurate, original, and genuinely useful rather than plausible-sounding and volume-filling — has not disappeared. What has changed is that the floor of acceptable quality has risen to a level that AI can meet, which means the ceiling — the quality level that justifies a professional premium — also needs to rise. The organisations and professionals who are investing in the capabilities that sit above the AI ceiling are the ones who will find that the quality collapse creates a market opportunity rather than an existential threat. Those that are competing with AI on volume and cost are discovering, consistently and predictably, that they cannot win that competition on those terms.
About the Author
Stuart Kerr is Technology Correspondent at LiveAIWire, covering artificial intelligence, emerging technology, and their impact on business, society, and everyday life. LiveAIWire publishes original AI journalism every weekday at liveaiwire.com.