Google just told Wall Street where the next shop window is: inside the AI answer. On Wednesday's earnings call, Google said ads in AI Mode, its chat-style search, are going to ten more countries, and that in the US ads will start showing up in the middle of follow-up questions. The stock fell anyway. That tells you how much is riding on this bet.
Also today: Instagram's boss says the backlash against AI content is going to get worse, OpenAI backed down on a price squeeze after its own customers revolted, and Mistral handed out a trillion-parameter model for free. Here's what each one means for you.
Google puts ads inside the AI conversation
Alphabet reported $127.6 billion in quarterly revenue, up 25% on a year ago. Good number. But growth in classic search ads has been slowing all year, and investors know why: people now ask a chatbot instead of clicking ten blue links. Google's answer is to sell ads inside the chat.
Two moves. First, ads in AI Mode are rolling out to ten countries outside the US, including Pakistan, India, Malaysia, Canada and Australia. Existing campaigns join automatically, with no opt-out. Second, in the US, Google is launching multi-turn ads: a sponsored result can appear on the third or fourth question of a conversation, not just the first.
Google also shared two claims worth noting. Questions in AI Mode are about three times longer than normal searches, and sponsored product listings there get click rates similar to organic ones.
Why it matters for you: most of you will never buy a Google ad, and that's fine. Look at the pattern instead. The biggest ad company on earth is betting that buying happens inside a conversation, across several questions, with longer and more specific asks. Your customers already shop that way in your DMs. The seller who answers the third question well, the one about size, delivery or "does it come in black", is the one who gets paid. If you want a refresher on what that conversation should look like, our playbook on how to sell in Instagram DMs walks through it step by step.
Instagram's Mosseri: the AI backlash will grow
Adam Mosseri, who runs Instagram, said on a podcast this week that negativity around AI is going to get worse, and that it's no longer just students and artists who are annoyed. He also admitted something sellers should hear clearly: Instagram doesn't rank AI creators differently on purpose, but he doubts AI-generated posts perform as well as other content.
He pushed back on two things. He doesn't think banning AI content is practical. And he thinks people throw the label "AI slop" around too loosely. He also warned that wrongly tagging a real creator's work as AI would make plenty of people angry, because detection tools make mistakes. Instagram already requires synthetic profiles to identify themselves, a rule that started in August.
Why it matters for you: here's my take. Use AI behind the counter, not in the shop window. Let it draft captions, sort orders and answer the "price?" messages at 2am. But keep your feed real: your hands, your product, your face if you show it. Audiences are getting allergic to posts that look generated, and the head of the platform just told you the algorithm may not love them either. For what actually moves reach, see our breakdown of Instagram's engagement signals.
OpenAI blinks on a price squeeze
OpenAI had planned to cut the usage included in its $200-a-month ChatGPT Pro plan after October 29, while keeping the price the same. Paying customers were furious. On Thursday OpenAI reversed course: no cut, and it doubled the Work and Codex allowances on its Pro 100, Pro 200 and Pro 500 tiers plus Business Premium, effective immediately. It didn't publish exact limits.
Why it matters for you: two lessons. One, AI pricing is still moving under everyone's feet, in both directions, sometimes within a week. Don't build a business process on the assumption that today's plan will look the same next quarter. Two, loud customers still win. When you pick any AI tool, look past the sticker price and ask what you get per month and how often that changes. We wrote an honest guide to what an AI assistant really costs a seller if you're weighing one.
Mistral gives away a trillion-parameter model
French lab Mistral released the open weights of Mistral Large 4, its flagship model with around one trillion parameters. It reads images as well as text, offers low, medium and high reasoning modes, and got extra training on coding and cybersecurity since its public preview earlier this month.
The license is a modified MIT license. Companies can use it, fine-tune it and use it commercially. The one catch: if you resell the model itself as a hosted service, you need a separate commercial license from Mistral.
Why it matters for you: you won't run a trillion-parameter model on your laptop. But every time a big lab gives away something this strong, the price floor for everyone else drops. Cheaper models underneath means cheaper AI tools on top, the ones you actually pay for. That pressure works in your favor.
Quick takes
- Gemini 4 Carbon: Google announced a new model built for long engineering and cybersecurity work. Access is limited to selected security teams for now. Not a seller tool yet, but it shows where Google's top models are heading.
- ChatGPT meets iCloud: ChatGPT can now connect to iCloud Mail and Calendar, so it can search your inbox and update events. Handy if you run your shop from an iPhone. Think twice about what you connect.
- Canada bets on Cohere: the Canadian government committed up to C$1.6 billion to AI company Cohere as part of a much larger funding round. Governments are now buying seats at the AI table directly.
- OpenAI's new rulebook: OpenAI rewrote its Model Spec, the document that sets how its models should behave, around five core values. Expect small shifts in tone and refusals in ChatGPT over the coming weeks.
The thread running through today
Google is selling ads in the middle of conversations. Instagram is warning that people are tired of fake-looking content. OpenAI learned that customers notice when you give them less. Put together, the message is plain: the conversation is where the sale happens, and people can tell when nobody real is behind it. That's the job Vardast was built for: an assistant that answers your Instagram, WhatsApp and Telegram customers fast, in your voice, with your real prices and stock, and hands the tricky ones to you. If your DMs are piling up while you post, try Vardast.