A little while back I shared the AI Research Tips video series I’ve been working on at Sinclair, and linked the video I did on topic selection with AI (plus your own ideas).
My next video was a collaboration with Sinclair’s AI Excellence Institute, which I’ve enjoyed getting the chance to work with on a few projects including serving as an external reviewer for some of the AI literacy modules they are creating, and participating in their pilot for paid AI tools.
The video is titled “How to Check AI Sources” and I bring in a popular existing framework for source evaluation to help determine if information provided by AI tools is accurate and trustworthy.
How to Fact-Check an AI Overview: Key Questions to Ask
Rather than treating AI Overviews as a single source, I encourage students to think of its output as a synthesis of many underlying sources, which can be a wide range of sources and viewpoints.
That shift in mindset helps reinforce the need to verify claims by consulting those original sources directly.
In the video I also introduce the concept of lateral reading, a research strategy that involves quickly checking multiple sources to evaluate credibility and context.
AI Overviews do cite sources, but those sources might be they might be news articles, academic studies, advocacy group websites, or even they could be user generated social media content. (In the video I show a Reddit post showing up as an AI citation.)
The question is not just whether they are “true” or “false” but whether they have particular motivations or points of view that might shape the emphasis of their information.
Beyond the individual sources, AI tools also often give you clean bullet points, but there’s no guarantee that they’re actually the most important aspects of the topic.
So look for what might be missing. Are important considerations or trade offs being left out?
Most published research deals with more specific questions, so overviews can compress a complex issue into a simple yes or no or good or bad summary.
Sometimes an overview will be technically accurate, but it leaves out key context needed to interpret it.
So your prompting strategies can help with this too, in addition to the lateral reading strategies.
Watch the full video here: