AI assistants misrepresent news content in 45% of responses: EBU and BBC study

Getting your news from AI bots? Might not be the best idea. AI seriously distorts news content in 45% of cases, regardless of language or country, according to a study initiated by the European Broadcasting Union (EBU) and 22 public broadcasters in 18 countries worldwide. Journalists reviewed more than 3,000 answers from #ChatGPT, Copilot, #Gemini, and Perplexity, checking accuracy, sources, fact‑vs‑opinion, and context. They found that 45 percent of answers had at least one serious issue. The most common problem, in 31 percent of cases, was missing or false sources. Twenty percent contained major inaccuracies, such as invented details or outdated information, and 14 percent lacked proper context. Gemini performed the worst, with serious issues in 76 percent of answers. The others were still weak: #Copilot 37 percent, ChatGPT 36 percent, #Perplexity 30 percent. These flaws are amplified by #AI’s confident tone, which makes unreliable answers sound trustworthy. Compared with BBC data from earlier this year, results show a slight improvement, but the error rate remains high across languages. Meanwhile, AI assistants are already replacing search engines for many users. According to the Reuters Institute Digital News Report 2025, 7 percent of online news consumers now get information through AI, rising to 15 percent among people under 25. So I’ll never get tired of saying that critical thinking and fact‑checking are more important than ever.

Study Finds AI Assistants Misrepresent News Nearly Half the Time A major international study by the European Broadcasting Union and the British Broadcasting Corporation reveals that leading AI assistants — including ChatGPT, Copilot, Gemini and Perplexity — produced responses with at least one major error in 45% of news-related queries. Overall, 81% had issues, including sourcing errors in a third of cases. The findings raise concerns about the reliability of AI when used for news. 👉 Read more:https://lnkd.in/dzQQ5M5u #PakistanTechNews #AI #AIaccuracy

"AI assistants are still not a reliable way to access and consume news" - EBU A new study from the European Broadcasting Union (EBU) released this month found that today’s AI assistants — including Gemini, Perplexity, ChatGPT, & Copilot — routinely deliver outdated, biased, or even fabricated news, blurring the line between fact and fiction. The research spanned 22 public service media organizations across 18 countries and 14 languages, underscoring how widespread the issue is. This is why we built Emerging Trajectories: for those who demand systematic analysis, accuracy, transparency, and rigorous evaluation of world events. Our event-based intelligence system transforms unstructured news into verifiable, contextual, and source-transparent data — so every claim can be traced, scored, and trusted. Learn how we’re rethinking AI for media, policy, and global analysis: link in the comments. #AI #DataIntegrity #NewsAnalysis #Monitoring #Analysis #Intelligence #News #CBC #OpenAI #LLMs #Chatbots

I often meet folks who rely on AI assistants (ChatGPT, Copilot, Claude, etc.) to make sense of the day’s news and developments. Yet it’s amazing how often those assistants deliver wrong answers with unabashed confidence. This includes outdated links, fabricated facts, and missing context. And when most users take that information at face value, rarely clicking a link to check sources, it only makes these problems worse. Chatbots might be fine for some use cases, but they fall short when teams need continuous, trustworthy analysis. That’s why we built a systematic approach—an end-to-end workflow designed to structure, score, and validate intelligence with rigor. 👇 #Analysis #Intelligence #Monitoring #Geopolitics #AI #BBC #News #EBU

Have you heard that the EU has already published the first Code of Practice for General-Purpose AI (GPAI)? It describes the main directions and requirements for general AI models, such as GPT chat and others. As a member of the working group that developed this code, I shared my comments on its practical application with AIN: ✔️ Maintain model documentation and make some data open ✔️ Use only legal data sets and take into account the digital signals of rights holders ✔️ Pay attention to systemic risk systems and models, and implement control and reporting procedures 📌 The Code of Practice for General-Purpose AI (GPAI) is a practical tool for demonstrating compliance with the AI Act, minimizing legal risks, and gaining a competitive advantage in the European market. The full article is available at the link in the comments ⬇️ #GPAI #AIRegulation #AIAct #LegalTech #AIlaw

🔍 AI is fast. But is it accurate? A recent study from the BBC and European Broadcasting Union caught my attention — and raised some serious questions about how we use AI in our daily work. According to the report: - 45% of AI-generated news responses had major issues - Google Gemini showed flaws in 76% of its answers - Sourcing was the biggest problem: missing, misleading, or misattributed citations across platforms like ChatGPT, Copilot, Gemini, and Perplexity As someone who works closely with AI tools, this hit home. I rely on these assistants for research, planning, and content — but this is a clear reminder to double-check everything and build safeguards into our workflows. Accuracy isn’t just a technical detail. It’s a trust issue. And in a world where AI is shaping how we communicate, that trust matters more than ever. 📖 Full story >>https://lnkd.in/dQ4Xm4sd #AITrust #ContentVerification #DigitalOps #AIEthics #MarketingTools #GenerativeAI

Can AI send you up a mountain and leave you there? According to a recent article from the BBC, possibly! http://bit.ly/4o7v3sd I posted a few weeks back about whether AI can be a useful therapist (https://lnkd.in/eu8UTMX7) and the answer was quite rightly: no. Large language models like ChatGPT are just trawling the internet and giving us words that seem to make sense in relation to the questions we ask. But they can't tell the difference between fact and fiction, and they don't know if something posted online is utter garbage or peer-reviewed science (or sometimes both: http://bit.ly/3IWgwk8). We still need to be critical and engaged, or there's a chance we might actually end up trying to find the Eiffel Tower in China (a supposed recommended destination!). Have you got any horror stories of AI mis-direction? (photo credit: ChatGPT)

🛑🛑🛑 📌 𝐀 𝐰𝐨𝐫𝐝 𝐨𝐟 𝐜𝐚𝐮𝐭𝐢𝐨𝐧 𝐚𝐛𝐨𝐮𝐭 𝐀𝐈-𝐠𝐞𝐧𝐞𝐫𝐚𝐭𝐞𝐝 𝐩𝐚𝐩𝐞𝐫𝐬📌🛑🛑🛑 I recently reviewed a paper where several references had 𝐢𝐧𝐜𝐨𝐫𝐫𝐞𝐜𝐭 𝐚𝐮𝐭𝐡𝐨𝐫𝐬, 𝐃𝐎𝐈𝐬, 𝐣𝐨𝐮𝐫𝐧𝐚𝐥𝐬 𝐚𝐧𝐝 𝐞𝐯𝐞𝐧 𝐦𝐢𝐬𝐦𝐚𝐭𝐜𝐡𝐞𝐝 𝐭𝐢𝐭𝐥𝐞𝐬 👉 clear signs they may have been generated using AI tools such as ChatGPT. This is becoming a worrying trend (several colleagues shared the same concern). AI can be a fantastic assistant for drafting and structuring your ideas, but it should never be trusted blindly for generating references or factual citations. These tools often “𝒉𝙖𝒍𝙡𝒖𝙘𝒊𝙣𝒂𝙩𝒆” by creating realistic-looking but FALSE details. As researchers, we have a responsibility to verify every reference and fact before submission. Using AI ethically means combining its strengths with our critical judgment, not replacing it. Let’s protect the integrity of our publications and the credibility of science.

AI assistants like ChatGPT, Copilot, Gemini, and Perplexity have become everyday information sources for many. However, a groundbreaking international study released last week by the European Broadcasting Union (EBU) reveals a concerning trend: these AI tools consistently distort news content regardless of language, country, or platform. The EBU study, which I participated in through Yle, involved professional journalists evaluating over 3,000 AI responses and the results are alarming: 45% contained serious factual errors 31% had significant source problems 20% included substantial deficiencies like outdated information AI assistants are already replacing search engines for information retrieval. According to Reuters Institute's Digital News Report 2025, 7% of online news consumers use them to follow news—rising to 15% among those under 25. What's particularly concerning is the gap between public trust and actual performance. Research by BBC and Ipsos reveals 42% of British adults trust AI summaries, with this figure reaching 50% for those under 35. Many incorrectly believe machines are naturally less prone to errors or bias. We definitely need more discussion and research exploring how AI assistants are failing as reliable news sources despite growing public trust in them. Read more: https://lnkd.in/dVye8QX8

The emergence of Small Language Models (SLMs) marks a significant advancement in AI following ChatGPT. Unlike mere scaled-down replicas of larger models, SLMs represent highly specialized professionals tailored for crucial assignments. When aiming for accuracy, efficiency, and confidentiality, particularly in fields like healthcare and finance, leveraging SLMs proves invaluable.Read More: https://lnkd.in/gPhbHYuD. #AI #GenerativeAI #LLMs #SLMs #TechStrategy #AIinBusiness #DigitalTransformation