Ready for all the ways AI is… not effective?
We will start with the news. A recent study by the European Broadcasting Union reveals that AI assistants misrepresent news content 45% of the time, raising concerns about their reliability. The research involved 22 public service media organizations across 18 countries, evaluating 3,000 responses from popular AI chatbots, including OpenAI’s ChatGPT and Google Gemini. The findings showed that 31% of responses had serious sourcing issues, while 30% were plagued by inaccuracies. Google’s AI, Gemini, performed the worst, with 76% of its outputs flawed.
Or they don’t push back. A recent study from Stanford and Harvard has confirmed that AI chatbots exhibit a strong tendency toward sycophancy, often endorsing user behavior more than humans do. The research involved analyzing responses from 11 chatbots, including ChatGPT and Google Gemini, revealing that these bots validate user actions 50 percent more than human counterparts. In a comparative analysis of advice given on Reddit’s “Am I the Asshole” thread, chatbots provided far more lenient feedback than human respondents, even in cases of irresponsible behavior.
And they are changing the web. The Wikimedia Foundation reports a significant decline in human traffic to Wikipedia due to the rise of generative AI chatbots and search engines that provide information without redirecting users to the site. Data indicates an 8% drop in human pageviews compared to the previous year, highlighting a broader trend where search engines are increasingly providing direct answers rather than linking to original sources. The Pew Research Center found that only 1% of Google searches resulted in users clicking on the link to the summarized page, further illustrating the challenge faced by platforms relying on user engagement.
Why do we care?
Let’s talk about how AI… isn’t that effective.
First, half the time it’s wrong — a new European study found AI assistants misrepresented news in 45% of responses. Then there’s the Stanford and Harvard research: AI loves to agree with you — way more than humans do. And while it’s doing that, it’s quietly killing the open web — Wikipedia traffic’s down 8%, and Google users barely click through anymore.
This is what happens when AI becomes the middleman. It looks confident, acts friendly, but often misleads and never credits its sources. For MSPs, don’t blindly trust AI outputs — validate, verify, and teach your clients to do the same. The future advantage isn’t who uses AI fastest — it’s who uses it wisely.