The Briefing

OpenAI answers this on its own help pages: a deleted chat is scheduled for permanent deletion within 30 days — unless it has already been de-identified, or must be kept for legal reasons. On August 25 we read seven of those pages and found the sentence that undoes the tidy version: to remove something ChatGPT knows, you must “delete every source where it appears.” Here are all seven stores, which ones the Delete button actually reaches, the order you have to clear them in so memory does not refill from your history, and the block of 2025 conversations no setting of yours can touch.

The advice you remember — count the fingers, watch the lips — is now the worst advice. YouTube started labeling AI videos automatically in May 2026, so on August 25 we took 17 videos that are AI-generated footage by their own title and read the disclosure panel on every one. Seven carried the label. Three showed only an auto-dubbing notice that looks exactly like one. Seven showed nothing at all — including a 4.3-million-view compilation with “Sora Ai” in its title. Here is what the label actually says, why its absence proves nothing, and the four checks that still work.

Amazon’s own help page answers this one: “You can’t turn off these ads.” On August 24 we read the current Echo Show documentation, wrote down the menu path Amazon actually uses now — Display & Appearance, not Home & Clock — and checked every Alexa+ plan for an ad-free tier. There isn’t one, at $139 a year or $19.99 a month. Here are the six switches that genuinely shrink what rotates past you, what each one costs you in weather and reminders, and why the ads grew in the first place.

Almost every guide promises the AI goes away if you follow the taps. On August 24 we read the actual help pages from Apple, Google, Samsung and WhatsApp for twelve places AI now shows up on a normal phone. Exactly one has a master switch that genuinely removes it. Google says outright that AI Overviews “cannot be turned off”; Samsung says some of its AI is built into the system. Here is the scorecard, what breaks when you flip each switch, and a ten-minute pass for a parent’s phone.

Free ChatGPT lost its message cap on August 10, and Plus quietly lost custom GPT creation on August 16. We read OpenAI’s own pricing page and help centre on August 23 and counted: the Free plan card says “Unlimited” once and “Limited” seven times. Here is the $0 vs $8 vs $20 math, including the row nobody prints — Go and Plus have the same context window.

It hit on August 5, 2026, at 6:35 UTC — 2:35 a.m. Eastern — at about 5,400 mph, and left a crater roughly 60 feet wide. NASA published the pictures on August 18. Here is what actually happened, why the viral video of it is fake, and the one number nobody printed.

Nevada approved 7,000 robotaxis for Las Vegas on Aug. 20. None of them can pick you up at Harry Reid, because airport access is a separate authorization nobody has yet. Here is what you can actually ride this weekend, and what it costs.

No — it is a $10-a-month perk, not a free one. We read Verizon’s own terms and found two ways to end up paying twice, plus seven other US AI offers verified on Aug. 22, 2026.

The FTC says social media platforms don’t always vet the ads you see, and Meta says its own ad review is largely automated. Here are five provenance checks that take about 90 seconds, and the number that shows which ads actually take the money.

ChatGPT now guesses your age from four behavioral signals – and when it isn’t sure, it guesses teenager. What the age check actually reads, and the four steps to undo a wrong call.