OpenAI launched ChatGPT apps to logged-in users outside the European Economic Area, Switzerland, and the United Kingdom, and the first partner set was available only in English and only in markets where each partner already operates. That was the buried constraint in the apps rollout, and it matters more than the DoorDash-Spotify-Uber demo loop getting all the coverage.
For an IT buyer at a 30-person agency running ChatGPT Business, Google Workspace, Figma, and Canva, that footnote changes the purchase decision. If half the team sits in London and the other half is in New York, you are not buying one clean workflow surface. You are buying a split rollout with uneven access, different permission prompts, and support overhead the headline never mentioned.
The useful detail was not the app list
Most coverage treated ChatGPT apps like a broad new app platform. OpenAI's own launch post described something narrower.
Apps were available to logged-in users on Free, Go, Plus, and Pro plans, but only outside the EEA, Switzerland, and the UK. The initial partner group was also small: Booking.com, Canva, Coursera, Expedia, Figma, Spotify, and Zillow. OpenAI said more partners would come later, monetization details would come later, app submissions would open later, and more granular data controls would come later.
That is not a mature app marketplace. It is a preview distribution channel with regional and language gates.
The boring deployment limit doing the real work
The launch post also made a second point most people skipped: ChatGPT apps could be invoked by name in the prompt, and ChatGPT could suggest them when relevant, but the first use still required an account connection and an explicit consent step so the user understands what data may be shared with the developer.
That sounds minor until you put it into a real operating environment.
An agency ops lead trying to standardize a design workflow across ChatGPT, Canva, and Figma now has to document who can connect what, in which region, under which account, with what data exposure. OpenAI said developers must collect only the minimum data they need and publish clear privacy policies, but it also said finer-grained category controls were still coming later. In plain English: the permission model is usable, not finished.
If you have a security review queue, this is the part that slows rollout. Not the prompt quality. Not the UX demo. The consent boundary.
The hidden cost is support, not subscription
OpenAI's pricing page already pushes Business as a secure workspace with 60-plus apps for internal tools and data. The public apps launch, by contrast, opened with a much smaller set of consumer-facing partners. That leaves a gap many teams will misunderstand.
A 25-seat team does not get value just because ChatGPT can spin up a playlist or open a Zillow map in chat. It gets value when the workflow eliminates app switching inside a process someone already owns. If your creative team uses ChatGPT plus Canva and Figma to turn client briefs into first-pass concepts, and each designer avoids four context switches a day at roughly two minutes each, that is about 200 minutes recovered per week across five designers. If ten more people cannot use the same flow because of region or account restrictions, the real cost is the fractured process and the extra internal support time.
That is why this rollout reads less like "ChatGPT got apps" and more like "OpenAI is testing whether conversational distribution can survive enterprise policy reality."
The decision for teams under a compliance or geography constraint
There is a real opportunity here. OpenAI's Apps SDK is built on MCP, is open source, and is clearly meant to become a distribution layer inside ChatGPT. That could become a serious channel for software vendors.
But if you are the buyer for a 20-to-50 person firm, the decision is not "standardize on ChatGPT apps" yet. The decision is whether your team has the right shape for a contained pilot.
Buy if your users are mostly US-based, already live in ChatGPT, and you can isolate one workflow with one owner and one connected app pair.
Wait if your team spans the UK or EU, needs uniform policy enforcement, or cannot tolerate a rollout where some workers get the shiny in-chat flow and others get documentation about why they do not.
Test it the same way you would test any other new integration surface: one region, one use case, one admin owner, one written data-sharing policy.
Verdict: test ChatGPT apps only where geography and permissions are simple; otherwise, wait for the controls to catch up to the demo.