ChatGPT's Intelligent UI: What Interactive Answers Change

On October 7, 2026, OpenAI started rolling GPT-6 out to ChatGPT with a feature it calls Intelligent UI. In OpenAI's description, ChatGPT can now compose an answer from text, visuals and interactive elements, choosing the mix to fit the question: a comparison side by side, an explanation as an interactive diagram, a recipe with its timing alongside, or a small tool such as a calculator built inside the conversation. Plus, Pro, Business and Enterprise accounts get it from October 7 on GPT-6 Sol. Free and Go accounts get it from October 8 on GPT-6 Luna. It lives in the Chat tab, and nowhere else yet.

For a marketer or a small-business owner the question is not how the interface works. It is what changes when the answer to "which of these three should I buy" arrives as a comparison grid rather than four paragraphs, and whether your product is in the grid. OpenAI's post says nothing about how sources are chosen for an interactive answer, and this post does not pretend otherwise. What it does is set out exactly what OpenAI said shipped, who sees which model, where the feature does not appear, and six checks a business can run this week on its own pages and in ChatGPT itself.

The API side of GPT-6 Sol and Luna, with prices and benchmarks, is covered in our September launch guide. This post is about the consumer product, and it stays on the consumer side.

OpenAI's announcement, "GPT-6 and Intelligent UI for everyone", describes two changes. The first is the model: a new GPT-6 model for ChatGPT, after last month's first GPT-6 models for paid customers. The second is Intelligent UI, which OpenAI describes as a trained ability to compose a response from text, visuals and interactive elements and to choose how they fit together based on the question. The named elements are graphics, tappable buttons, forms, charts and what the post calls interactive experiences.

The post gives everyday examples rather than business ones. A Sunday roast plan where the timing sits next to the recipe. A road trip where the stops appear on a map with notes on detours. An explanation you can change an input on and watch the result move. A savings calculator, a bill splitter or a game built on request. OpenAI groups these into three families, and the grouping is a useful map of where a business might meet the feature.

The second half of the release is less visible but may matter more in daily use. OpenAI says GPT-6 in ChatGPT can begin answering while it is still thinking or using tools, then add findings as they arrive. The post describes training the model to account for how long the user has waited and to build the answer across several partial responses without filler. The release notes put it in one line: "You can start reading before the full answer is ready."

This is the part most coverage will blur, and it is the part a business needs exact. GPT-6 in ChatGPT is not one model. OpenAI's availability section says Plus, Pro, Business and Enterprise run on GPT-6 Sol and Free and Go run on GPT-6 Luna, and that both are tuned for everyday conversation. The dates differ by a day. The table is built from the announcement and the ChatGPT release notes entry for October 7, and nothing in it comes from anywhere else.

Why the distinction matters: Sol and Luna are different models, and OpenAI's own API documentation prices and benchmarks them differently. A buyer on a free account and a buyer on Plus can ask the same question and get different answers in different formats. Any check a business runs has to say which tier it ran on, or it is not a record. The failure case is a team that tests on the founder's Pro account on October 7, finds its product in a comparison grid, and concludes that the millions of free users will see the same thing. They are on a different model, and the rollout does not start reaching them until the next day.

OpenAI drew the boundary itself. The update applies to the Chat experience in ChatGPT, across web and mobile. The release notes add that Intelligent UI is available from the Instant setting through Extra High reasoning. It is not in ChatGPT Work and not in Codex; the post says the models behind those are not changing in this release. And the Pro reasoning option, which continues to run GPT-6 Astra, does not support it.

For a business this means the surfaces most likely to carry an interactive answer about your product are the ordinary consumer chat on a phone or a browser, on every plan once the rollout reaches it (on Enterprise, subject to admin settings). The surfaces a company uses for its own work, Work and Codex, are unchanged. If your team uses ChatGPT Work internally and sees no change, that is expected, and it says nothing about what your customers see.

One question the post does not answer is whether a user can switch the visuals off. OpenAI describes the model choosing the format, and says a simple text answer remains available when that is the most useful response. It describes no setting that turns Intelligent UI off, and we have not seen one documented in the release notes. Until OpenAI says otherwise, assume the format decision belongs to the model and that a user influences it only by how the question is asked.

OpenAI's description of the machinery is short, and the three parts it names are worth knowing because they constrain what the feature can show. First, a library of native, streamable components: the model does not draw arbitrary pixels, it assembles from a fixed set of parts, which is why the post says each response keeps a familiar design foundation. Second, a compiler that processes the interface while the model is still generating it, so a chart or a table can appear before the response is finished. Third, training: OpenAI says it expanded its methods to teach the model decisions about content, layout, visuals and interaction, and evaluated the interfaces it produced for clarity, usefulness and completeness.

The constraint that follows is that an interactive answer is made of components OpenAI built, filled with whatever the model has. A comparison component needs rows and columns, so it needs attributes it can line up. A chart component needs numbers. A calculator needs a formula and inputs. If the facts about your product that would populate those components are vague on your own site, the component either omits you or fills the cell with something it found elsewhere. That is the whole practical lesson of section 06, and it holds whatever the source-selection logic turns out to be.

OpenAI is also candid that the work is unfinished. The post says "there's still work ahead to improve the model's design judgment and expand what it can create". A feature described that way on launch day will change, and a business process built on exactly how it looked on October 7 will drift out of date.

Three numbers in the announcement will be repeated without their qualifiers, so here they are with them. All three are OpenAI's internal evaluations, described in the post and not independently replicated at publication.

None of these numbers is about your business. They matter here for one reason: a faster, earlier-starting answer that arrives as a grid is a different reading experience from a paragraph that arrives after a pause, and a buyer who reads the grid may never scroll to the text that follows it. The thing to measure is not OpenAI's latency. It is what the grid says about you.

Each check below is something you control or something you can observe. None depends on knowing how ChatGPT picks sources, which is why the list is short. The first three are about your own pages. The last three are about recording what ChatGPT shows, so that in a month you can tell whether anything changed.

Checks 1 to 3 are the same work that generative engine optimisation has asked for since AI answers started citing pages: clear, consistent, machine-readable facts. Intelligent UI raises the cost of not doing it, because a missing value in a text answer is a vague sentence and a missing value in a comparison grid is an empty cell next to a competitor's full one. Checks 4 to 6 are new work, and the discipline is the same one we use for measuring AI search visibility: fixed questions, fixed accounts, dated records, no inference from a single run.

A worked example. A regional accounting firm lists three packages on its site, each with a "from" price and a bulleted feature list that uses the firm's own names for things. A buyer asks ChatGPT to compare it with two national providers whose pages state flat monthly prices and use the category's standard terms. If a comparison grid is produced, the national providers have values in every cell and the regional firm has "from" and blanks. The fix is on the firm's page, takes an afternoon, and would have been worth doing before October 7.

Launch weeks produce confident advice. The router below pairs the claims you will hear with what OpenAI's documents support, and what to do instead. It is deliberately conservative, because the feature is a day old and its own maker says the design judgment is still being improved.

The right posture for the first month is that of a record keeper. Fix what you control on your own pages this week. Run a fixed set of questions on fixed accounts and date the output. In November, compare. If you want that record built and read for you, with the page fixes prioritised by what the answers actually show, it is the kind of work our agentic SEO team does for clients whose buyers already ask ChatGPT first.