Should your business build a ChatGPT app?

At its developer conference on 6 October, OpenAI announced that ChatGPT can now run apps from other companies inside the conversation. Ask it about homes for sale in a US city and Zillow's map appears in the chat. Ask for a poster and Canva opens a design you can edit. Booking.com, Coursera, Expedia, Figma and Spotify were the other launch partners, with more promised by the end of the year.

ChatGPT has around 800 million weekly users, according to OpenAI. So the obvious question for any business is whether it should have one of these apps. For most, not yet. For a few, it is worth starting now.

What a ChatGPT app actually is

Strip away the launch videos and an app has two parts. The first is a set of actions your business exposes to ChatGPT: search our listings, check availability, get a quote, book a slot. The second, optional, is a small interface that appears in the chat to show the results, such as a map, a list of rooms or a booking form.

ChatGPT decides when to use your app based on what the user asks, or the user calls it by name. The conversation carries the context, so someone who has spent five minutes describing what they want does not have to fill in a form from scratch.

The interesting part is what it is built on. OpenAI's Apps SDK uses the Model Context Protocol, MCP, which Anthropic published in November last year as an open way to connect AI assistants to outside tools. Microsoft, OpenAI and Google all adopted it this spring, and most major AI assistants now support it.

Why MCP matters more than the app

MCP is the reason this is less of a gamble than it looks. The actions half of a ChatGPT app is an MCP server, and the same server can be connected to Claude, to Microsoft's Copilot tools and to a growing list of other assistants. You are not writing code that only works in one company's product.

That changes the business case. The real question is whether your core services should be usable by AI assistants. A ChatGPT app is one front end on top of that.

Who should consider building now

The businesses that benefit are ones where a customer's task can be finished in a conversation, and where the answer depends on live data only you have. Some examples:

• Booking and availability: clinics, salons, tutors, rentals, restaurants.
• Searchable inventory: property listings, used cars, product catalogues with real filters.
• Quotes and calculators: insurance, logistics, renovation, anything where the price depends on details the customer can describe.
• Courses and learning, where the app can find the right lesson or quiz someone on it.

In each case, the user would otherwise have to leave the chat, find your website, and start again. The app saves them that trip.

Who should wait

If your website is mostly a brochure, with services, an about page and a contact form, there is nothing for an app to do that ChatGPT cannot already do by reading your site. Spend the money on the site.

There are also practical reasons for everyone else to move carefully:

• It is a preview. OpenAI says app submissions and a directory will open later this year, so for now most businesses cannot publish an app to the public.
• It is not available everywhere. Users in the EU, the UK and Switzerland do not get apps yet. That matters if part of your audience is in Europe; it does not if your customers are in the Gulf or India.
• How users will find apps is unclear. ChatGPT may suggest an app in the conversation, but nobody outside OpenAI knows yet how it will choose between two competing ones.
• Payments inside apps are promised but not detailed.

What we would do today

If your business fits the first list, build the MCP server first. Pick the two or three actions customers ask for most, make them work reliably against your real data, and decide what an assistant should and should not be allowed to do, such as checking availability freely but never cancelling a booking without a confirmation.

That server is useful on day one, inside Claude or your own internal tools, and it is most of the work of a ChatGPT app when submissions open. If the app store idea fades, you still have your services in a form every major assistant can use.

If you are in the second group, keep an eye on it and come back in six months. There is no prize for being first with a thin app.