GPT-6 and Intelligent UI: what ChatGPT’s interactive answers mean for UK businesses

On 7 October 2026 OpenAI put GPT-6 into the Chat tab of ChatGPT and gave it a new trick: it can answer with an interface instead of a block of text. In OpenAI’s own demo, a request for a Sunday roast plan comes back with a guest-count stepper that rescales the shopping list, a cooking timeline you can tick through, and a button to copy the list. OpenAI calls this Intelligent UI. Plus, Pro, Business and Enterprise got it first, and the rollout to Free and Go began on 8 October. OpenAI describes the rollout as global, so UK accounts are included.

The model upgrade matters, but nearly all the evidence for it comes from OpenAI’s own tests. The interface change is the bigger story, because it changes what an answer from ChatGPT looks like to the more than 1.2 billion people OpenAI says use it each week, and what they still need your website for.

The answer in brief

Intelligent UI lets GPT-6 decide, question by question, whether to reply in text or to build a small interface: a diagram you can tap through, a side-by-side comparison, a form, a chart or a working calculator. It needs no special prompt, comes with GPT-6 in the Chat tab on every plan, and you can dial it down in settings.

• Who gets it: Plus, Pro, Business and Enterprise from 7 October 2026 on GPT-6 Sol; Free and Go from 8 October on GPT-6 Luna. Enterprise access depends on admin settings.
• Where it works: Chat on the web, in updated mobile apps and in the new desktop app, at any thinking level from Instant to Extra High that your plan includes. It does not work in the older ChatGPT desktop apps for macOS and Windows, the Work tab, Voice, or the Pro reasoning option, which runs on GPT-6 Astra.
• What it costs: it comes with the plan. OpenAI’s models help page says Intelligent UI has no separate usage quota; your existing model, tool and plan allowances apply.
• What it means for businesses: more questions will be answered with a tool inside ChatGPT. If your site’s value is a simple calculator or comparison, expect ChatGPT to build its own version on demand. The facts it builds from still come from the web, and in our parcel test the credit went to a comparison site rather than the carriers selling the service.

What is Intelligent UI in ChatGPT?

Intelligent UI is GPT-6’s ability to compose a reply from text, visuals and interactive elements, choosing the mix to suit the question. In OpenAI’s launch post, responses “can include graphics, tappable buttons, forms, charts, and interactive experiences” used directly in the conversation. When plain text is the most useful answer, the model is meant to give you plain text.

OpenAI groups the examples into three uses. Everyday plans get a visual structure: cooking times beside a recipe, road-trip stops on a map. Hard topics get explanations you can poke at, such as the central limit theorem, GDP or the Monty Hall problem, where you change an input and watch the result move. And you can ask for a tool outright, such as a bill splitter, a retirement savings calculator or a retro game, and use it in the chat.

The Sunday roast comparison on the launch page is worth a look for a UK reader. The GPT-5.6 Instant answer is a long article with pounds and Fahrenheit. The GPT-6 Instant answer is a menu with photos, a stepper that scales the shopping list in kilograms and grams, and an eight-step checklist. Its food-safety note still cites the USDA’s 145°F, so the localisation is partial. Same question, different object.

How does Intelligent UI work?

GPT-6 assembles these answers from ready-made parts. OpenAI built a library of native, streamable components and a compiler that turns the model’s output into an interface while it is still being generated. The library keeps every response looking like ChatGPT; the model decides which pieces to use and how they fit together. Because the compiler works as the text streams, the interface appears progressively instead of arriving all at once at the end.

The design judgement is trained. OpenAI says it expanded its training to cover content, layout, visuals and interaction, and graded the interfaces GPT-6 produced for clarity, usefulness and completeness. It also admits the job is unfinished: the launch post says there is still work ahead on the model’s design judgement and on the range of things it can build.

Two practical details from the help article matter for anyone who uses these tools for real work. Some components, such as checklists, can keep their state when you refresh the chat, but component state does not carry over to another chat. In our test, a mortgage calculator did not even keep its inputs after a page reload, so do not treat these tools as somewhere to store work.

Hasn’t ChatGPT shown interactive answers before?

Yes, in pieces. The ChatGPT release notes show a steady build-up through 2026: interactive maths and science modules covering more than 70 topics in March, interactive bar, line, pie and scatter charts in June, quizzes you answer inside the chat in August, interactive content that appears while it is still being generated, also in August, and flashcards in September. Shopping results already offered side-by-side product comparisons from March.

What changed on 7 October is the scope. Each earlier feature added one capability, such as charts, quizzes or interactive content that streams in. Intelligent UI lets the model compose the layout of any answer in Chat from one shared set of components, and mix them in a single reply. That is why OpenAI calls it a significant shift in how people interact with ChatGPT, and why it closes the launch post with the claim that software will adapt to people instead of the other way round.

Which GPT-6 model am I using in ChatGPT?

The Chat model menu says GPT-6, and OpenAI’s help page points out it is not labelled Sol. Which model sits behind it depends on your plan. The naming is confusing because OpenAI now has GPT-6 models in three places, released across September and October.

Sources: OpenAI launch post, Intelligent UI help article, GPT-6 October system card, GPT-6 Sol and Luna, GPT-6.1 Sol announcements and the GPT-6 and other models help page, checked 10 and 11 October 2026.

The detail most coverage misses is in the system card. OpenAI separates the October versions of Sol and Luna now running in Chat from the September versions that remain in Codex and Work. So “GPT-6 Sol” in Chat and “GPT-6 Sol” in Codex are different releases of the same family. GPT-6.1 Sol, launched on 29 September and priced at $2 per million input tokens and $10 per million output tokens in the API (US list prices), is not yet in Chat. For developers that settles one question early: we found no announced API for Intelligent UI, so switching your model ID will not give your app ChatGPT’s interactive answers. The route for a business that wants its own interface inside ChatGPT is still a plugin; since 29 September, plugins can include interactive panels and forms.

Expect some confusion in OpenAI’s own documentation. On 11 October its GPT-6 and other models help page still said Free and Go users get GPT-5.6 Luna, while the launch post, the release notes, the Intelligent UI help article and the system card all say GPT-6 Luna. Our free account built interactive answers that day, which OpenAI describes only for GPT-6, so the help page looks out of date. The same page confirms that GPT-6 replaces the old Latest option in the menu and that GPT-5.6 Sol stays available as a separate choice. On our Pro account the Chat menu listed exactly that: GPT-6, GPT-5.6 Sol, and GPT-5.5 marked as leaving on 14 October.

Is GPT-6 faster in ChatGPT?

It starts answering sooner, which is a different thing from finishing sooner. GPT-6 can begin replying while it is still thinking or using tools, then add findings in later parts of the same answer without another prompt. OpenAI says it trained the model to weigh the user’s waiting time and to keep the assembled answer as coherent as one written in a single pass.

OpenAI gives two figures, both from its own internal evaluations:

• For questions that need web search, GPT-6 Instant starts answering 44% sooner on average than GPT-5.6 Instant.
• On high-value everyday agentic tasks, GPT-6 at Extra High begins answering in the same time as GPT-5.6 at Medium, and scores better overall than GPT-5.6 at Extra High.

OpenAI gives no method or sample size for either, so treat both as vendor claims until someone measures them independently. OpenAI also says GPT-6 makes better decisions about when to search and more reliably finds information that supports its answer, but it gives no number for that.

What does OpenAI’s safety card say about GPT-6 in ChatGPT?

The October system card, published on launch day, is more mixed than the launch post. On the plus side, OpenAI reports stronger resistance to jailbreaks than GPT-5.6, including attacks that adapt over several turns (though slightly below the September GPT-6 versions, within overlapping margins), fewer factual errors across nearly all its hallucination tests, and better honesty when a tool such as search is broken.

The regressions are spelled out too. On OpenAI’s hardest safety prompts, GPT-6 Sol shows a statistically significant regression on self-harm, and GPT-6 Luna, the free-tier model, regresses on self-harm, gore and sexual content. In OpenAI’s evaluations for users under 18, both models regress on age-restricted content, sexual content and emotional reliance, and Luna also on gore. With image inputs, results held level with the August models apart from a regression on the extremism test. OpenAI says it reviewed the failures and judged them generally low in severity, puts part of the emotional-reliance result down to the test penalising friendly nicknames, and applies an extra classifier block for users it believes are under 18 that the scores do not reflect.

OpenAI also says the October models are less likely to refuse harmless requests or add excessive caveats. Read the scores with care: OpenAI runs these evaluations without the safeguards that sit around the model in ChatGPT and says the error rates do not represent normal traffic. For a business deciding whether staff can rely on ChatGPT, the fair summary is that OpenAI reports gains in helpfulness and factual accuracy alongside specific safety regressions, and the scores alone do not tell you how often staff would meet a problem. Organisations with younger users or sensitive subjects should read the card itself.

Can you turn Intelligent UI off?

Not completely. On our free account the control sits under Settings → Personalization as a Layout and visuals menu with two options, Default and Simple. OpenAI’s Intelligent UI help article gives the path as Settings → Personalization → Layout and Visuals → Simple on the web, while its models help page talks of turning Layout and visuals off. Either way, OpenAI says some visual elements may still appear, because the setting does not switch ChatGPT to a text-only mode.

You can also steer it with Custom Instructions, for example “keep answers concise” or “use tables when comparing options”, or ask for a different style in a single conversation. OpenAI is careful to say these preferences guide the response and the exact layout can still vary.

That matters if your team copies ChatGPT output into reports or emails. Interactive components are built to be used in the chat, so check what survives when you copy an answer out. For drafting, the simpler setting will give more predictable text, and the Work tab, where Intelligent UI does not run, is built for finished documents.

What happened when we tested Intelligent UI?

It works on both free and paid plans, and it credits sources inside the interface. On 11 October 2026 we ran three prompts on a UK free account in Chrome and two of them on a UK Pro account in the new ChatGPT desktop app for Windows, with GPT-6 at the Instant setting on the paid account. Each prompt ran once, so treat this as a small exploratory test.

Both calculators returned £1,389.58 a month for £250,000 at 4.5% over 25 years, which matches the standard repayment formula when we worked it out ourselves. On the free account, the £200k preset recalculated the payment to £1,111.66, also correct. When we reloaded the chat, the calculator went back to £250,000. OpenAI says only that some components, such as checklists, can keep their state.

What we saw matters for brands. Sources appear as small citation chips inside the generated components themselves, including inside the heat pump cost card. The credit can go to an intermediary. Our paid parcel answer took its Evri, DPD and UPS prices from Parcel2Go, a comparison site, and credited Parcel2Go for them instead of the carriers. The same prompt also produced different interfaces on the two accounts. The plan, model and app all differed, so we cannot say which made the difference, but a test on one account tells you little about the other.

The desktop test also settles one practical question. The new ChatGPT desktop app on Windows showed Intelligent UI normally; OpenAI’s warning applies to the older desktop apps.

What does Intelligent UI mean for SEO and AEO?

It raises the bar for any page whose whole job is a simple tool. Look at OpenAI’s own examples: a savings calculator, a bill splitter, a side-by-side comparison, a cooking timeline. Those are the kinds of pages many sites use to win search traffic. A user who gets a working mortgage calculator inside ChatGPT has less reason to click through to a generic one.

Google is heading the same way. On 19 August 2026 it said generative UI had launched globally in English in AI Mode and had started rolling out in AI Overviews, so a Google search can now return a custom tool or interactive visual as well. Google’s example is an interactive pH scale inside an AI Overview, and Search Engine Journal reports a mortgage calculator in AI Mode. The announcement gives no timetable for AI Overviews in the UK, so check your own searches before assuming they show it. For a site built on simple tool pages, the pressure now comes from both of the places people search.

The tool still needs facts, though. A comparison needs prices, specifications and lead times. A savings calculator needs current rates. OpenAI says GPT-6 decides better when to search and finds supporting information more reliably. Our reading is that pages stating those facts plainly are the most likely raw material for the interface. That favours content with real numbers in plain text and tables, the same figures everywhere you publish them, and information ChatGPT cannot invent: your own prices, stock, data and experience.

OpenAI does not document how sources are credited inside a generated interface: as of 11 October 2026, none of its four Intelligent UI documents mentions citations, source links or attribution. Its general search help says only that ChatGPT ranks results on multiple factors and that placement is not guaranteed. Our tests show what can happen in practice: citation chips sit inside the components, and in our parcel test the credit for three services went to a comparison site instead of the carriers selling them. One test cannot show why ChatGPT chose that page. Our advice is to state your own prices and facts at least as plainly as any aggregator, so you are not leaving the citation to someone else.

Advertising does not offer a way in either. Ads began rolling out to UK users on Free and Go plans on 4 June 2026, and OpenAI repeats in its 5 October ads update that advertising does not influence ChatGPT’s answers. Its new visual ad format will be tested during image generation, starting in the US later this month. On our free UK account, a labelled ad for a property adviser appeared beneath the mortgage calculator, with ChatGPT’s note that ads do not influence its answers. The ad sits outside the answer, so visibility inside an Intelligent UI answer still has to be earned through content.

What should UK businesses do next?

1. Run your money prompts. Ask GPT-6 the ten questions your customers ask before they buy, on a free account and on Plus or higher, in the browser, since Free and Go run GPT-6 Luna while the higher plans run GPT-6 Sol, and older desktop apps show no Intelligent UI. Note which answers become interfaces, what facts they use, and whose page each citation chip credits. Not every answer becomes an interface, so also ask for a comparison or calculator outright, as some customers will.
2. Audit your tool pages. List every calculator, comparison and checklist page on your site. Where ChatGPT can now build the same thing in seconds, add what it cannot: your real prices, your data, a quote or booking step. Larger brands with a genuine tool can also look at a ChatGPT plugin.
3. State your facts more plainly than the aggregators. Put prices, specifications, delivery times and eligibility rules in readable text and simple tables, and keep them consistent across your site, listings and profiles. Our ChatGPT guide to SEO, AEO and GEO covers the rest of the groundwork.
4. Set a house rule for staff. Decide whether teams should use the Simple layout setting for drafting, keep client and confidential figures out of generated calculators and forms, and remind them that interactive answers still need checking before anything goes to a client.
5. Watch how citations change. Rerun your prompts monthly and check OpenAI’s help article. The Intelligent UI documentation does not explain how citations work inside these interfaces, so their behaviour could change without notice.
6. Measure what you can. Track visits referred from chatgpt.com in your analytics, and keep dated screenshots of each monthly test with the citations it shows. Analytics only sees clicks, so an answer built from your facts that sends nobody to your site leaves no trace there; the screenshots are your record of it.

Is GPT-6 with Intelligent UI a big change?

For how people use ChatGPT, yes. The model gains are real on OpenAI’s own numbers, but few have been measured independently yet. The interface change is rolling out to every plan, free users included, and can turn an everyday answer into a tool to use as well as text to read.

OpenAI’s own closing line sets out the ambition: people should stop adapting to software and software should adapt to people. Whether that holds depends on design judgement OpenAI admits is still developing. For UK businesses the job this month is simpler. Find out what ChatGPT now builds when your customers ask about your category, and make sure the facts inside it are yours.

AIWIZ can run these checks for you: which answers ChatGPT builds for your category, whose pages it credits, and which of your facts need stating more plainly. Explore our SEO and AEO services, or talk to AIWIZ about an audit.