An AI wrote a cold email congratulating a prospect on their $40M Series B. Real raise. Wrong company. The prospect was Northwind Logistics, a freight company. The $40M belonged to Northwind Inc., an Austin cybersecurity startup that happens to share the name. Every word was true. It was just true about somebody else. Send it, and you open by congratulating a stranger on a raise that isn't theirs. The part that should worry anyone running outreach at scale: that line sails through a normal accuracy check. The words match a real funding announcement, exactly. What it fails is a question almost no personalization tool asks: is this source about THIS company, or a different one that shares the name? So we built a demo, the Veracity Engine, to ask it. An AI still writes the email. Then a separate layer of plain code, not a second AI grading the first, re-checks every claim against its source. The Series B line failed the entity check and got stripped before the email could send. What ships is only what a current, correctly-matched source backs. (Northwind is a synthetic lead built for the demo, so you can watch a claim get caught without anyone's real inbox on the line.) Honest question if you run AI outreach: after the model drafts, does anything re-check the finished claim before it sends, or does it go out on the model's word? Curious how teams handle that step. #RevOps #SalesAI