If your business ever built a custom GPT, whether it’s a customer-facing helper on the website or an internal tool trained on your own documents, OpenAI has just put a firm date on when that stops working. On 11 September, OpenAI announced it is retiring custom GPTs entirely, with creation of new ones ending on 25 September and the feature switching off for good on 11 December 2026. The replacement is Plugins, which combine reusable instructions with connected apps in a different structure.

This has moved quickly. Many small businesses who built a GPT a year or two ago as a cheap, no-code way to wrap ChatGPT around their own knowledge base treated it as a “set and forget” tool. That window has now closed, and the migration isn’t fully automatic.

What actually changes

Under the hood, a custom GPT’s instructions become a Skill inside a Plugin, its uploaded knowledge files are converted into reference files, and any connected apps it used may carry over into the new structure. OpenAI says the migration path exists, but the original GPT becomes read-only once migrated, meaning any updates or fixes need to happen before that point, not after. There’s also a wrinkle worth knowing about: a public GPT built inside an Enterprise workspace follows that workspace’s retirement timeline even when someone opens it from a personal account, so a popular GPT a business built and shared publicly could stop working on 11 December regardless of which plan the end user is on.

For a business that built something simple, an FAQ bot, a document Q&A tool, a lead-qualification assistant, this is a genuine “check it now” moment rather than something to leave until the December deadline arrives.

What to check before December

Three questions are worth answering this week. First, does your business have any live custom GPTs, whether built in-house or by a contractor who may no longer be reachable? Second, has OpenAI’s migration banner appeared for it yet, and if so, has anyone actually reviewed what carried over correctly versus what needs rebuilding? Third, if the GPT is customer-facing or embedded anywhere on your website, is there a fallback plan if it goes dark unexpectedly rather than migrating cleanly?

This kind of quiet infrastructure churn, a vendor deprecating the exact building block a business relied on, is common with fast-moving AI tools, and it rarely comes with much warning built around a small business’s own timeline. If your team built something on custom GPTs without much technical oversight at the time, or nobody currently owns checking on it, this is a reasonable moment to get a proper technical read on what you actually have running and whether Plugins is the right landing spot, or whether it’s worth rebuilding on more stable footing. That’s the kind of practical, unglamorous AI adoption work outfits like BuildApps exist to help UK businesses sort through, rather than guessing and finding out in December.

Why this keeps happening to small businesses specifically

Larger companies tend to have an IT team or a vendor-management process that tracks platform announcements like this one as a matter of routine. Small businesses usually don’t, which means the first sign of trouble is often the tool simply stopping, not an email that got read and actioned months earlier. Custom GPTs were popular precisely because they were quick and cheap to spin up without specialist help, and that same lack of oversight is what makes their retirement easy to miss until it’s urgent. The lesson worth taking from this one, beyond the December deadline itself, is that any AI tool built quickly and cheaply is worth revisiting every few months, not just when a vendor forces the issue.

The takeaway

If your business has a custom GPT doing real work anywhere, internal or public-facing, don’t wait for the 11 December cutoff to find out whether it migrated properly. Check this week, and if nobody currently owns that tool, decide today who does.