Everyone now owns a printing press. Almost nobody has an editor.
This summer, for the first time, social media and video platforms overtook news organisations’ own websites and apps as the most widely used source of news across all 48 markets in the Reuters Institute’s Digital News Report 2026: 54% of people used them for news in the previous week, against 51% for publishers. In the same report, trust in news fell to 37%, the lowest since the Institute started measuring it in 2015. Weekly use of AI chatbots for news rose from 7% to 10% in a single year, and among 18- to 24-year-olds, more than half now say social media, video or AI is their main source. The newsroom has not been abolished. It has simply been made to look unnecessary.
That is the illusion worth taking seriously, because it is now spreading from the news into the boardroom. When information flows instantly and answers arrive on request, the middleman starts to look like friction. Why wait for a journalist to check the sources when the feed already has the story? Why wait for a data team to build the model when the chatbot already has the number?
The answer, in both cases, is that the middleman was never in the business of delivering information. Delivery was the easy part. The job was deciding what was true, what was relevant and what was missing, which is precisely the part that speed makes harder, not easier.
The evidence on the news side is not subtle. In October 2025, the European Broadcasting Union and the BBC coordinated the largest study of its kind into how AI assistants handle news, involving 22 public broadcasters and more than 3,000 answers. Forty-five percent of those answers contained at least one significant issue, from fabricated details to sources attributed to the wrong outlet. The assistants refused to answer only 0.5% of the time. They are confidently wrong far more often than they are honestly silent. And the BBC’s parallel audience research found that many people simply assume AI summaries of the news are accurate, and when they discover an error, blame the news organisation as much as the machine.
Now move the same pattern indoors. Workiva’s 2026 Midyear Executive Benchmark Survey, of 2,272 finance, risk and sustainability professionals including 847 C-level executives, found that 84% of executives are at least somewhat confident in the accuracy of AI output without human review. In the same survey, only 11% believe their data quality is good enough for AI, and one in four say internal audits have already caught AI errors that reached external audiences or the board. AlphaSense’s 2026 State of AI for Business and Finance, a survey of more than 1,000 AI decision-makers published in September, found that 62% have been negatively affected by at least one bad AI output. Confidence is rising. So is the bill.
The technical picture explains why. On classroom-style benchmarks, language models now translate business questions into SQL with accuracy above 90%. On Spider 2.0, a benchmark built from real enterprise databases (often over a thousand columns, spread across cloud warehouses, with queries running past a hundred lines), the best agent its authors tested solved only around one task in five. The demo is the clean benchmark. Your company is the messy one.
And yet the perception keeps travelling in the opposite direction. McKinsey’s State of AI 2026 found that 39% of organisations expect AI to reduce their headcount over the next year, even though only 14% reported any actual reduction the year before. Expectation is running well ahead of evidence, and the functions that look most like “answering questions” are the first to feel it. If the chatbot can tell the CFO last quarter’s margin by region, what exactly is the analytics team for?
It is for the same three things a good newsroom is for. First, is the source true? A journalist checks who said it, when, and whether they were in a position to know. A data professional checks where the number came from, which system it was extracted from, and whether that system is the one Finance actually trusts. Second, was it handled properly? A journalist compares accounts and notices when a quote has been stripped of context. A data professional checks the joins, the filters, the currency conversion, the definition of “active customer” that Sales and Finance quietly disagree about. Third, and most important, is this even the right question? An editor decides what the story is. An analyst notices that the margin question is really a pricing question, or that the dashboard everyone is debating measures the wrong thing entirely.
None of those three steps is visible in the answer. That is the whole problem. A fluent, well-formatted response looks identical whether it rests on the governed figure or on a stale extract someone left in a shared folder in 2023. The AI does not show you the transformations it guessed at, and it will rarely tell you that you asked the wrong question. It will simply answer the one you asked, beautifully.
I’ve spent the better part of two decades watching numbers travel from the system that produced them to the slide that announced them, and the damage almost never happened in the system. It happened in transit: a filter nobody documented, a definition that changed meaning between departments, a question answered precisely when it should have been challenged. The reason organisations built data teams was never that querying a database is hard. It was that knowing which answer to believe is.
The executives who are tempted to treat AI as a replacement for their data function are making the same mistake as the reader who stopped buying a newspaper because the feed was free. The information did not get cheaper. The verification got outsourced to nobody. In a world where every question gets an answer in seconds, the scarce thing is no longer answers. It is the judgement to tell signal from noise, and the people whose job is to exercise it.
Cut the editors, and you don’t get faster news. You get more noise, delivered with better grammar.
Sources: Reuters Institute for the Study of Journalism, “Digital News Report 2026” (June 2026) for news sources, trust and AI chatbot use; European Broadcasting Union and BBC, “News Integrity in AI Assistants: An International PSM Study” (22 October 2025) for the 45% significant-issue rate and refusal rate; BBC, “Audience Use and Perceptions of AI Assistants for News” (October 2025); Workiva, “2026 Midyear Executive Benchmark Survey” (August 2026) for executive confidence, data quality and audit findings; AlphaSense, “The 2026 State of AI for Business and Finance” (9 September 2026); Lei et al., “Spider 2.0: Evaluating Language Models on Real-World Enterprise Text-to-SQL Workflows,” ICLR 2025 (November 2024, revised March 2025); McKinsey, “The State of AI in 2026: On the Road to ROI” (25 August 2026) for headcount expectations.
Related: The Shadow Function, Part II on capability without accountability, and The Plural of AI on who owns the bin.
Leave a Reply