ChatGPT builds the calculator now. Build the one it can't.

The bill splitter just moved into the chat

On October 7, OpenAI put GPT-6 into ChatGPT with something it calls Intelligent UI. Answers can now come back with charts, tappable buttons, forms, and small working tools. OpenAI's own examples are a savings calculator, a bill splitter for dinner with friends, and a game you can play right in the conversation.

Paid plans got it first. Free and Go users start getting it today. So by the end of this week, a lot of the people who might use the calculator on your website can ask ChatGPT for one instead, and it will build it for them on the spot.

Search Engine Journal said it plainly: calculator and converter pages are the easiest things for this to replace. We agree, and we'd go one step further. The question isn't whether your site should still have tools. It's which tools are actually yours.

Here's the short version. If the only things a tool needs are the visitor's own numbers and some public math, the chat can already make it. If it needs something only you know, like your stock, your prices, your rules, or a customer's history, it can't. That second kind is where new budget should go.

So take a stack of your tools and sort them into three piles.

Pile one: let it stop carrying traffic

A mortgage calculator. A unit converter. A square footage estimator. A generic ROI calculator where the visitor types in made-up assumptions and gets a made-up number back.

These were popular because they're cheap to build and people search for them. That's exactly why they're the first to go. Anyone can now get the same thing inside the conversation, adjusted to their question, without opening a new tab.

That doesn't mean you should delete them tomorrow. A visitor who's already on your site and wants a quick number is still helped by one. What changes is what you expect from them. Don't commission a new generic calculator because you hope it will rank and bring in leads. Nobody has published traffic numbers on this yet, including us, so treat it as a strong hunch rather than a measured fact. But a tool that anyone can recreate by asking is a weak bet for a new project.

One quick test: open ChatGPT, describe your tool in a sentence, and ask it to build one. If what comes back is about as good as yours, that tool is in this pile.

Pile two: rebuild it around what you know

Now the interesting pile. These are tools that look generic today, but only because they were built without the one thing that would make them useful.

Take a made-up tile showroom. Its website has a "how many boxes do I need" calculator: room size times a waste factor. ChatGPT can do that in a second. But the store knows things the chat doesn't. It knows that this particular tile in a herringbone pattern wastes more than a straight lay. It knows which location has 14 boxes from the same dye lot, so the floor won't come out two slightly different colors. It knows when the next delivery lands. A calculator that answers with "you need 11 boxes of this one, the downtown store has them in the same lot, and you can hold them for 48 hours" isn't a calculator anymore. It's the first step of an order.

The same move works in other businesses. A shipping estimate built on your carriers, your cutoff times, and the warehouse that actually has the item. A reorder page that starts from an account's last orders and contract prices. An eligibility check that runs your current rules and saves the reason for each answer, so staff can see why someone was told yes.

The value was never the interface. Our founder and team saw this on Wellthie, which turned complex healthcare plan data into comparisons people could use. The comparison screen was the visible part. Keeping the distinctions in the data intact, so a choice still meant what it said, was the work. A chatbot can draw a nice comparison table. It can't draw your data.

Pile three: the tools your team makes for itself

Business and Enterprise plans got Intelligent UI too, though on Enterprise it depends on what your workspace admin allows. So expect people on your team to start asking ChatGPT for quick tools. A pricing calculator for a sales call. A little form to sort returns. A chart for a meeting.

For one-off work, that's fine. It's often great. The trouble starts when a generated tool quietly becomes the place a number comes from. Nobody owns the formula. There's no shared version, so two people get two answers. Nothing is saved, so when a customer asks why their quote was what it was, nobody can show the work.

Watch for two signals. The first is when a second person asks for the same tool. The second is when the tool's output gets pasted into something a customer sees, like a quote, a contract, or an order. Either one means it has stopped being a scratchpad. At that point it needs a home: a real tool with one source for its numbers, someone responsible for it, and a record of what it said.

That's usually a small build, not a big one. But it's much easier to do on purpose than to untangle six months later.

How to sort your own tools

You don't need a workshop for this. List every interactive thing on your site and every spreadsheet your team treats like an app. For each one, ask three questions.

What does it need to know? Write down every input. Mark which ones the visitor types in and which ones come from you.

Where do your inputs live? Inventory system, price list, policy document, someone's head. If the answer is "someone's head," that's useful to know for reasons well beyond this.

What happens after the answer? If the honest answer is "nothing, they leave," the tool is in pile one. If the answer could be a reservation, a quote, a booking, or a handoff to a person who can see what the visitor already entered, it belongs in pile two.

Expect to find one or two tools worth rebuilding and a few you can simply stop investing in. That's a good outcome. The point is to stop paying for things anyone can now get for free, and to put that effort where your business actually knows something.

What we don't know yet

A few honest unknowns. OpenAI's announcement doesn't say how sources will show up when an answer arrives as a chart or a tool rather than a paragraph. That matters if a generated calculator quotes a rate or a rule that came from your site. We also don't know yet how much traffic generic tool pages will lose. Anyone giving you a percentage this week is guessing.

What we do know is what changed on October 7. Simple tools became something people can make by asking. That raises the bar for the ones you build yourself, and it makes the ones built on your own data more valuable, not less.

If you have a tool in pile two and want help scoping it, our custom product build brief is a good place to start writing down what it needs to know. And if you'd like a hand building it, that's the web application work we do. Tell us about it.