Sponsored by edgeful

Happy Monday!

My brain’s been stuck on one question from this whole “who’s winning consumer AI” debate: why, when the models are all getting scary-good, do most people still only really use one assistant?

The data says it all. For most of 2025, fewer than 10% of ChatGPT weekly users even visited another major provider, and only about 9% of consumers pay for more than one subscription across ChatGPT, Gemini, Claude, and Cursor.

That’s not a “people carefully compare features” market. That’s a habit market.

And that’s the part that’s easy to miss if you only look at model quality. The switching cost is not money. It’s mental default. Once one tool becomes the place you go when you’re stuck, curious, writing, planning, or procrastinating, you stop shopping around. You don’t wake up and think, “Which LLM should I use today?” You just open the one that already lives in your muscle memory.

So the real competition in consumer AI isn’t “who has the smartest model.” It’s “who can break the default.” And the only reliable way to break a default is to create a moment where the incumbent feels inconvenient. Not bad. Just inconvenient.

That’s why image and video mattered so much this year. They’re not just features, they’re distribution. Gemini’s recent acceleration is a perfect example: their growth has been helped by viral image models like Nano Banana.

The important detail isn’t the name, it’s what these models do for adoption. A lot of people don’t want to stare at a blank chat box and invent a prompt from scratch. Nano Banana Pro is basically designed to remove that friction: better text rendering inside images, more precise edits, multi-image blending, character consistency, and even the ability to ground to Search for real-world accuracy in certain workflows.

That’s a very different pitch than “chat, but smarter.” It’s “open this, do one obvious fun thing in 20 seconds, and you’ll tell a friend.”

Now here’s the non-obvious part: the viral stuff is acquisition, but the boring stuff is retention. The tool you keep is the one that starts saving you time every week, even when nothing is trending. That’s why the most strategic push isn’t another flashy demo, it’s getting closer to your calendar, your shopping decisions, your docs, your routines. OpenAI’s Pulse is literally framed as proactive daily research delivered as scanable cards, tied to your past chats and connected apps.

And their shopping research feature is the same direction: not “ask me anything,” but “I’ll do the comparison work and bring you a buyer’s guide.”

That’s also where most of the products will fail, even if the models are great. Once you cross the line from “help me think” to “help me act,” reliability becomes the product. If the assistant is wrong in a chat, you shrug. If it’s wrong when it’s pulling “facts,” summarizing your schedule, or nudging you to buy something, you stop trusting it. Trust is a moat, but it’s also fragile.

This is why I’m skeptical about “AI social” as the big next frontier for the main assistant apps. Social platforms run on identity, status, and the feeling that a real human is behind the post. When the content is obviously AI-generated (or feels like it could be), that status loop breaks. The creation tools will win attention, but the consumption and community layer is still going to live where people already have an audience.