Skip to playerSkip to main content
  • 8 hours ago
AI might just treat a random Reddit thread with the same authority as a government site. And this has everyday consequences for you.

Category

🗞
News
Transcript
00:00When did Trump die of rabies?
00:02If you type this question into DuckDuckGo's AI search recently,
00:05the bot would confidently have given you a precise date.
00:09Of course, the answer is complete nonsense.
00:11But here's the problem.
00:13AI summaries can easily be manipulated.
00:16In this case, Reddit users coordinated troll postings about the fake story in a serious tone.
00:22The AI picked them up and repeated them as fact.
00:26Why?
00:26Because for large language models, a random Reddit comment can weigh as much as an official source,
00:32if it sounds plausible.
00:34In fact, a new study shows that AI deep research agents rely on user-generated sites
00:39like Reddit and Wikipedia for roughly half of all their answers.
00:43And this has everyday consequences for you.
00:46Looking for a restaurant?
00:48Researchers showed that planting a tiny 13-word snippet on a scraped forum
00:53was enough to trick the AI into recommending a specific place as a factual top choice.
00:59And it isn't just about food.
01:01Brands are already gaming the system, flooding forums and comment sections to get picked up by AI scrapers.
01:07They call it Answer Engine Optimization, or AEO.
01:12Researchers call it something else, poisoned content,
01:15because sometimes one comment can shape entire search results.
01:19So next time you read an AI summary, check the source.
01:23It might just be internet gossip in disguise.
01:26Bye guys.
Comments

Recommended