This episode pairs a fast AI-marketing news rundown with a hands-on tutorial on catching AI slop and hallucinations before they ship.
In the news (June 9-16, 2026): OpenAI shipped user-facing memory controls on June 12, so you can now delete and edit individual memories or turn memory off, useful for keeping brand-voice and client context clean (background on the underlying Dreaming architecture, and the Releasebot log). Google made information agents in AI Mode broadly available to AI Ultra subscribers, a smarter replacement for Google Alerts (PPC Land, 9to5Google). And CallRail added ChatGPT Ads attribution, closing the measurement loop for call and form conversions. Context: the OpenAI Partner Network launched with a reported $150M backing.
The tutorial: What "AI slop" actually is, the vocabulary and structural tells (more signs, overused words), and why hallucination is a brand and legal risk: Air Canada held liable for its chatbot, lawyers sanctioned in Mata v. Avianca and Morgan & Morgan, the FTC's Operation AI Comply, and incidents at Sports Illustrated, CNET, and KPMG. The NP Digital report shows how often marketers ship errors, detectors don't work, and we end with a copyable two-pass edit and fact-check workflow (fact-checking guide).