Agentic Web Today

Letter from the Editor

“AI Accuracy in focus (NYT)”

How often would you accept a wrong answer from google? Its so interesting because for decades we've been ok with clicking blue links and being taken to pages of varying quality. But now that AI distills it all into a single authoritative answer, we expect that answer to be correct despite it just being assembled based on the same crap we used to surf. A new NYT investigation put Google's AI Overviews through the SimpleQA benchmark and found they're accurate about 91% of the time with Gemini 3, up from 85% on Gemini 2. Not bad until you do the math: Google processes over five trillion searches a year, which means tens of millions of wrong answers every single hour. Stat that jumped out at me in the context on brands: 56% of claims are "ungrounded" (sources Google cites don't actually support the claims being made). They're the normal output of a system that guesses answers using probabilistic models, not verified facts. If Google's own AI can misread your website and generate a confidently wrong answer about your product, your pricing, your policies, you're at risk of losing control of the narrative. Risk for brands, but on the other hand as Lily Ray says: "If someone wants to be known as a world expert at something, he or she merely has to write a blog post self-proclaiming that distinction." Something to think about...

— Jake