Somewhere around a billion people got a different ChatGPT this week. A month after paying users, GPT-6 is now the default for everyone on the platform, and with it comes the feature OpenAI calls Intelligent UI: answers that assemble themselves out of text, graphics, tappable controls, and interactive elements, mixed according to what your question needs.
Ask how a 7-speed bike works and you don’t get paragraphs describing a derailleur. You get a diagram of the drivetrain you can tap through. Ask for a roast schedule and the timings sit next to the recipe instead of buried in a wall of text. Ask something simple and you still get plain text, because plain text is the right tool for that job.
That last detail is the one worth sitting with. The interface is generated per question. There is no fixed layout anymore.
The bet: software adapts to people
For decades, using software meant learning someone else’s interface. Menus, forms, checkout flows, the particular way one company decided a task should be done. The web softened this but kept the core deal: you visited a page someone designed, and you read it the way they built it.
OpenAI’s closing line in the announcement is the whole strategy in one sentence: “Instead of people adapting to software, software will adapt to people.”
If that works at scale, it’s a bigger deal than any single model release. Describe what you want and get a purpose-built interface on the spot: a form, a chart, a mini-tool generated in the moment for your task, then gone. ChatGPT stops being a text box and becomes whatever the question calls for. It’s the most visible change to the product since voice mode.
Why this lands on a marketer’s desk
This is where it gets awkward for anyone who makes a living from pages.
An entire industry exists around controlling the page. SEO, web design, conversion optimization, analytics heatmaps, all of it assumes there is a page, that people arrive on it, and that what they experience there is largely what you built. Intelligent UI breaks that assumption. When ChatGPT answers a comparison question with an interactive widget, there is no page to rank, no layout to test, no hero image to optimize. The model composed the answer from raw material, and the raw material is what you published.
I’ve been doing this long enough to be suspicious of “the death of the page” takes, so let me be precise: pages are not dying this year. Your site still matters. But the center of gravity is shifting from designing destinations to supplying substance. When the interface is generated, the content is the only thing left that’s yours.
If you’ve been following generative engine optimization, this is the same trend taken to its logical end. GEO was about getting cited inside an AI answer. Intelligent UI is about getting composed into one. The citation becomes a component.
How to write for an interface you don’t control
You can’t design the rendering. You can influence what it’s built from. A few things hold up regardless of how the rendering evolves:
Write in composable units. A claim, a spec, a price, a step, a comparison point. Each one should survive being lifted out of its surrounding paragraph and still make sense. The model picks and arranges; give it parts that hold together under rearrangement.
Structure for extraction. Tables, definition lists, question-and-answer blocks, clearly labeled steps. This is standard GEO advice and it hasn’t changed. What’s changed is the payoff: extracted chunks now feed something rendered, not just quoted.
Answer the renderable questions. “Best CRM for a 5-person agency” is now potentially an interactive comparison. “Our philosophy of customer success” is not. Look at the questions your buyers actually ask and be honest about which ones deserve a widget.
Keep the facts clean and attributable. A generated interface that confidently shows the wrong price is worse than a static page with the wrong price, because it looks more official. Sources, dates, and version numbers inside your content give the model something to anchor to.
Reasons to hold your enthusiasm
Free-tier scale is the untested part. A demo of a tappable drivetrain diagram is one thing; a billion weekly users hitting generated interfaces simultaneously is another. Expect jank in the first months.
There’s also a failure mode nobody has priced in yet: interactive wrongness. A wrong answer in a paragraph reads like an opinion. A wrong answer dressed up as a slick generated widget reads like a product. Confidence goes up whether or not the facts did.
And this shipped alongside October safety updates for GPT-6 Sol and Luna, which is worth noting, though the safety details live in OpenAI’s deployment pages rather than the announcement itself.
None of this breaks the trend line. It just means the transition is messier than the keynote implies.
What to do this week
- Open ChatGPT and ask it the ten questions your buyers actually ask. Screenshot what comes back. That’s your new competitive surface.
- Pick your three highest-traffic pages and restructure the answers on them into composable units: discrete claims, tables, labeled steps.
- Add dates and version markers to anything that changes, prices especially.
- If you publish comparison or buying content, test whether Intelligent UI renders it as an interactive answer, and note what the widget pulled from your structure versus a competitor’s.
The page had a good three-decade run as the unit of attention. The answer is taking its place, and the answer doesn’t have a layout. It has ingredients. Make sure yours are worth composing.
