The dream is one thing, the reality is quite another. When we peel back the marketing jargon, we find that a significant portion of AI-generated news isn’t just slightly off; it’s dangerously, demonstrably wrong.
So, What Is This ‘Automated Journalism’ Anyway?
Before we dive into the wreckage, let’s clarify what we’re talking about. At its core, automated journalism uses artificial intelligence, specifically Large Language Models (LLMs) like those powering ChatGPT and Google Gemini, to generate news articles, summaries, and reports. Media organisations, often squeezed by tight budgets and the relentless demand for content, are increasingly experimenting with these tools. The logic is simple: why pay a team of journalists to spend hours summarising a lengthy report when an AI can spit out a version in seconds?
The benefits, on paper, are compelling:
– Speed: AI can produce content almost instantaneously, keeping pace with the lightning-fast 24-hour news cycle.
– Scale: It can cover niche topics, like minor league sports or granular financial data, that would be too costly for human journalists to tackle.
– Data Analysis: It can sift through enormous datasets—think government spending records or scientific papers—and surface patterns a human might miss.
This is the great promise. It’s a vision of journalistic abundance. But as the saying goes, if it sounds too good to be true, it probably is. The shiny exterior of this automated news factory hides a production line with a shocking lack of quality control.
The Problem Isn’t a Bug, It’s a Feature
The study led by the BBC is a sobering wake-up call, a splash of cold water in the face of AI hype. After analysing over 3,000 responses from leading chatbots, the researchers found that a staggering 45% of answers contained at least one serious error. Let that sink in. Nearly half the time you ask a popular AI for a news summary, you’re getting bad information. This isn’t a rounding error; it’s a fundamental breakdown in AI content accuracy.
The issues are not random or isolated. They fall into several disturbing categories:
– Factual Inaccuracies (20%): The AI simply gets the facts wrong, inventing details, misattributing events, or fabricating figures.
– Misleading Citations (31%): The models cite sources that don’t support their claims or, in some cases, don’t even exist. This is particularly insidious as it gives a false veneer of credibility.
– Outright Fabrication: The AI hallucinates entire events or quotes, creating a piece of fiction and presenting it as news.
Think of these LLMs not as knowledgeable researchers, but as incredibly sophisticated plagiarists. They’ve read a vast swathe of the internet but understood none of it in a human sense. They are pattern-matching machines, stitching together words and sentences that are statistically likely to follow one another. This process can produce text that looks coherent and authoritative, but it has no internal concept of truth or falsehood. The result is a high-tech rumour mill, and the consequences for public trust are dire. As Jean Philip De Tender, the EBU’s Media Director, so aptly put it in his response to the study’s findings, “When people don’t know what to trust, they end up trusting nothing at all.” That is the existential cliff edge the media is staring over.
Case Study: Google’s AI Gets an ‘F’ in Journalism
Now, let’s name names. The study didn’t just give a general overview; it broke down the performance of individual models. And the results for one of the world’s biggest tech companies are nothing short of catastrophic. According to the data published in ComputerWorld, Google’s Gemini had problems in 76% of its responses.
This is a strategic disaster for Google. The company is desperately trying to integrate AI into its core search product, positioning it as the future of information discovery. Yet their flagship model is, by a huge margin, the least reliable news source in the test. Whilst Microsoft’s Copilot, OpenAI’s ChatGPT, and Perplexity all had significant error rates, none came close to Gemini’s failure. It’s like an automotive company unveiling its new self-driving car and having it crash in three-quarters of the test drives. Would you get in?
The implications are massive. For years, Ben Thompson’s Aggregation Theory has explained how platforms like Google and Facebook captured value by controlling the distribution of content. Now, Google is trying to move from being the distributor to being the creator, or at least the summariser, of that content. But if the summaries are unreliable, Google isn’t adding value—it’s actively polluting the information ecosystem it once curated. This isn’t just a technical problem; it’s a profound breach of the trust pact Google has held with its users for decades.
Fighting Fiction with Fact-Checking
So, what’s a journalist—or, more importantly, a citizen—to do? We can’t just throw our hands up and surrender to the tide of misinformation. This is where news verification tools become less of a professional utility and more of a basic survival kit for navigating the digital world.
These tools range from simple reverse image search engines to complex platforms that track the origin of a story or video. However, they are not a magical fix. The first and most important verification tool is the one between your ears: scepticism. When you encounter an AI-generated summary, the default assumption should not be “this is probably true,” but “this needs to be checked.”
Here are some practical steps for everyone:
– Go to the Source: Never take an AI summary at face value. Click the link. Read the original article, study, or report. The AI is a shortcut, but it’s a shortcut through a minefield.
– Cross-Reference: Check if reputable news organisations are reporting the same story. A single, unverified AI summary floating in the digital ether is a major red flag.
– Question Everything: Who is the author of the original piece? Does the publication have a history of rigorous journalism? Does the AI’s summary accurately reflect the tone and key findings of the original source?
The EBU and BBC are pushing for institutional change, launching a ‘News Integrity in AI Assistants Toolkit’ to help tech companies improve their models. But waiting for Silicon Valley to voluntarily fix a product they’re rushing to market is a fool’s errand. The responsibility also falls on us.
The Ultimate Antidote: Media Literacy
This brings us to the most powerful, long-term solution: media literacy education. This isn’t about teaching people what to think. It’s about teaching them how to think critically about the information they consume, regardless of its source. It’s about building a populace that is resilient to misinformation, whether it’s generated by a state-sponsored troll farm or a well-meaning but hopelessly flawed algorithm.
Promoting media literacy education means:
– Integrating it into School Curricula: From a young age, students should learn how to identify sources, evaluate evidence, and understand the difference between fact, opinion, and fabrication.
– Public Awareness Campaigns: We need broad campaigns that explain the automated journalism flaws in simple terms, empowering citizens to be more discerning consumers of news.
– Platform Responsibility: Tech companies must do more than just label AI-generated content. They should actively fund and promote media literacy initiatives, treating it as a core part of their responsibility for the information ecosystem they dominate.
This isn’t a “nice-to-have.” It’s an urgent necessity for the health of our democracies. An informed public is the bedrock of a functioning society, and right now, that bedrock is being eroded by technology that prioritises speed and scale over truth.
The age of AI in journalism is here, but it has arrived half-baked and with a user manual full of warnings that no one is reading. The damning findings from the EBU and BBC are not the end of the story, but a crucial chapter in it. They expose the gaping chasm between the utopian promises of AI and its messy, unreliable reality. Ignoring the automated journalism flaws is no longer an option. The challenge now is to build the guardrails—both technological and educational—before the trust that underpins our entire information ecosystem collapses for good.
So, here’s the question I’ll leave you with: Who bears the primary responsibility for cleaning up this mess? Is it the tech giants who released these flawed tools into the wild, the news organisations tempted by their cost-cutting allure, or is it up to each of us to become more vigilant digital detectives?
