Conversational Custom Screens Open a Market for Field Operations Budgets

An input screen has arrived inside the answer box

On October 7, 2026, OpenAI introduced Intelligent UI alongside GPT-6. Rather than answering only in text, it can create clickable buttons, input forms, selection screens, charts, and simple calculation tools inside a conversation.

The official demonstrations included a meal planner that changes ingredient quantities and cooking steps by group size; a trip-planning screen that maps stops along a route and shows notes about detours; and a calculator for changing savings conditions. The key point is that, instead of reading an explanation and moving to another screen, people can change values and see the result again in the same place.

The screen does not have to appear all at once after the full answer is complete. OpenAI says it is designed to reveal buttons and forms step by step as they are created. When a short response is more appropriate, it displays text only, as usual.

Not every user can access it under the same conditions yet. It is rolling out through the web and supported current apps, and may not appear depending on an account’s subscription plan, an organisation administrator’s settings, and the response mode selected. Older desktop apps and some ways of using the service are also outside the supported scope.

OpenAI says GPT-6 was designed for more than 1.2 billion people who use ChatGPT each week. This is the company’s own figure, but it shows the scale at which this way of generating screens is reaching the public rather than remaining a test tool for a small group of developers. However, official documentation does not currently describe a public feature for attaching these screens directly to external services, so a customer-facing service still needs to be built separately.

Fix one job, not the whole website

Consider a hypothetical window-installation company with 12 employees. Its estimate intake manager receives addresses and window dimensions by phone or text message, while photos arrive separately later. The manager copies the details into a ledger or file, then asks again for missing items.

Each customer needs different questions. The next question changes depending on whether the property is an apartment or a detached house, whether existing windows must be removed, and whether site photos are available. Put every question into a fixed inquiry form and it becomes long. Cut the questions down and someone has to call back.

A custom screen inside a conversation aims for the middle ground. When a customer selects an apartment, it can ask for the building and floor. When they select the number of windows to replace, it can show that many photo-upload fields. At the end, it can organise missing items and what to prepare before the site visit on one screen.

This is different from automatically producing a finished estimate. People still need to confirm actual dimensions, choose materials, assess risky installation conditions, and approve the final price. The screen is less about replacing an expert than collecting information in a state that is easier for the expert to review.

It is also difficult to use a temporary form made in ChatGPT directly as a customer intake window. The appearance and placement of buttons can vary with each request, and connection to an external business system requires separate verification and development. If it collects personal information, the notice for collection and use of personal information and the deletion process must be checked separately against relevant official standards.

The person who can now hold a new budget is not the website owner, but the operations manager measured on estimate-intake results. At a small business, the owner may approve spending while the estimate manager chooses the tool. In an organisation with several branches, the person responsible for customer support or field operations may propose the purchase. What they care about is not a polished screen, but whether abandoned intakes, callback counts, estimate preparation time, and repeat site visits fall.

The sales approach should change accordingly. Rather than proposing a company-wide system replacement, choose one job—such as estimate intake or visit preparation—and let the team run a short trial. In the first explanation, it makes more sense to show how three inquiries missed today can become complete intakes tomorrow than to focus on AI performance.

Abroad, guidance and handoff to people were designed together

GOV.UK Chat, run by the UK government, provides natural-language guidance for government information that can be difficult to find, including tax, welfare benefits, and visas. Across two public trials, more than 10,000 people asked 26,000 questions, and 73% of follow-up survey respondents said it was useful. Accuracy rose from an initial 76% to 90%, but the average response time was 10.7 seconds, and tasks requiring personal-record checks had to be handed to people.

The service draws on more than 700,000 pages on the government website and shows relevant sources and next actions. Its recent app design also addressed larger text and screen-reader support. The useful lesson is not that it automatically creates screens, but that it helps people find what to do next and makes clear where automated guidance ends.

The Driver and Vehicle Licensing Agency (DVLA) chatbot began with repeated questions about vehicle tax and driving licences. It expanded its scope gradually over roughly three months after launch, and handed difficult questions to web support staff.

During operations, it received about 20,000 pieces of feedback each month and adjusted the question flow. It helped with simple inquiries, but short and ambiguous phrases could send people down the wrong path, and it was not suitable for complex inquiries. The case shows that deciding when to hand off to a person matters as much as adding more automated screens.

Four things you could build from here

1. Site-visit estimate intake that does not miss photos

• What it does: Changes the required measurement and photo fields based on the customer’s answers, then organises missing materials before the visit.
• Who uses it: Window installation, wallpapering, and move-in cleaning businesses with around 10 employees that receive estimates by phone and text before going to sites.
• Why now: An operations manager can test industry-specific question screens first, then move only proven flows into a customer-facing service.
• First screen: Three groups of large buttons for job type, building type, and preferred date, plus an expected-time notice.

2. A preparation guide for a shop hiring its first employee

• What it does: Asks about the business type and hiring conditions, then shows documents to prepare and institutions to check in order, while directing final decisions to official guidance.
• Who uses it: Owners of neighbourhood cafes or side-dish shops who have operated alone and are about to hire their first part-time employee.
• Why now: Instead of showing a long explanation all at once, it can reveal only the steps that apply to the selected situation.
• First screen: Fields for business type, planned hiring date, and work arrangement, with a marker for items that require confirmation with an official institution.

3. A comparison helper for venues with different terms

• What it does: Takes attendance, time, and whether outside food is allowed, then shows total costs and unconfirmed conditions side by side for each candidate venue.
• Who uses it: Individuals and small gathering organisers arranging their first class or first-birthday party, who otherwise repeat the same questions to several venue-rental businesses.
• Why now: Conditions raised in conversation can become a comparison screen and follow-up questions immediately, making it easier to filter out quotes with blank fields.
• First screen: Four fields for date, attendance, area, and total budget, plus a button to add candidate venues.

4. A closeout checklist that changes with each job

• What it does: Creates an inspection sequence based on job type and site characteristics, and lets workers record a photo and note for each item.
• Who uses it: Field supervisors at small facilities-management and cleaning businesses where several teams move at once and job-completion reports vary from team to team.
• Why now: Instead of a fixed checklist made in the office, they can test an inexpensive small tool that changes the next item based on field responses.
• First screen: A selector for today’s work site, job-type buttons, a count of incomplete items, and a button to start attaching photos.

Why this matters where you are

Check whether local service businesses repeatedly ask for the same missing information, or ask customers to resend photos through another channel. The relevant rules, official institutions, and handoff options will differ by market. Start with one recurring intake or inspection task, and make the point where a person must take over explicit.

What to check in 30 minutes today

On Naver Map, choose three businesses in one trade and start an estimate inquiry, while making clear that you do not intend to place a real order. Write down the first questions you receive and the information requested again later. If two or more of the three ask the same follow-up question, or require photos to be resent through another channel, it may be worth building a first screen that gathers only those questions.