August 10, 2026: OpenAI cuts ChatGPT’s factual error rate by roughly 60 percent with a new model update

Welcome to the AI Industry Report, the daily brief built by AI, about AI, from the Agentic Marketing Academy. It’s Monday, August 10th. Over the weekend, OpenAI quietly rewired the model answering millions of ChatGPT conversations, and it came with a number attached. Three stories, and proof that the AI recommending your business just got measurably more honest. Let’s go.

Let’s talk about what actually changed inside ChatGPT this weekend, because it’s bigger than a routine update. Starting Thursday, OpenAI began swapping out the model behind ChatGPT itself. Plus and Pro users got an updated GPT-5.6 Sol. Free and Go users got a new default model called GPT-5.6 Luna. Both replace the older GPT-5.5 Instant, and the rollout reached everyone by Saturday. Now, here’s the number that actually matters to you. OpenAI ran its own tests on financial, medical, and legal questions, the kind where a wrong answer costs someone something real, and found the updated Sol model produced 68 percent fewer responses containing a factual error. Across its broader testing, the drop landed around 60 percent. That’s not a cosmetic tweak. That’s OpenAI’s own data admitting the previous model was hallucinating a lot, and doing something serious about it. Why does that land on this show? Because when somebody asks ChatGPT whether your shop is open on Sundays, what your starting price is, or whether you do a specific repair, that answer used to have a real chance of just being wrong. It’s still not flawless. But it just got meaningfully more trustworthy, for free, without you lifting a finger.

Sticking with OpenAI for a second, because they weren’t finished this week. ChatGPT Ads Manager picked up a batch of features aimed squarely at anyone selling something online. There’s a new carousel ad format that shows several of your products in one unit inside a ChatGPT conversation, live right now. There’s a beta version of automatic bidding that optimizes for actual sales instead of just clicks. And there’s a deadline worth circling on your calendar. On August 17th, a feature called Automatic Advanced Matching becomes the default on every existing ChatGPT ads account. It uses hashed customer data from your own website’s forms to match more conversions back to your ads. If that’s not something you want switched on automatically, go find the opt-out under Tools, Conversions, Data Source, before the 17th, and don’t worry, it’s a regular setting, so you can flip it off later too if you miss the date or change your mind.

And one from Google that explains something you may have already noticed and shrugged off. Google’s John Mueller confirmed this week that when somebody answers a follow-up question inside AI Mode, even a one-word reply like “yes,” Google logs that as a brand-new search query. It shows up right in your regular Search Console performance report, complete with its own clicks and impressions, sitting next to your normal keyword data. So if you’ve spotted strange one-word entries like “yes” or “yes, pricing” cluttering your reports lately, that’s not a glitch you can ignore. That’s a customer’s AI Mode conversation about your business, showing up in pieces.

So here’s your one thing for today. Open ChatGPT and ask it three plain questions about your own business, your hours, your pricing, and one thing you specialize in, and see if the newly updated model gets all three right.

That’s today’s report. Transcript and every source linked at A I industry report dot com. If you want your business to be the one AI recommends, the free checklist at the leveling window dot com is where everybody starts.

And full transparency, because we practice what we teach: this show is researched, written, and voiced by AI, built by our founder. We teach businesses what AI can do. This show is the proof. Hit follow, and tomorrow’s report finds you on its own.