Gems are disappearing too: why you shouldn’t tie AI assistants to a single provider
After OpenAI, which is replacing its own chatbots (custom GPTs) with plugins, Google is now also discontinuing Gems. On 30 September 2026, Google announced that Gems will be merged into skills, reusable instructions that you use within a regular Gemini conversation. For those using Gemini privately, this is convenient. For schools, municipalities, healthcare organisations and businesses that have built their own assistants on Gems or GPTs, it means yet another transition. And that is precisely why your choice of AI provider matters.
What Google is changing
A skill is a set of instructions and context that you define once and then call up in any conversation by typing a forward slash and the name. Gemini can also apply a skill automatically if it matches your query, and you can combine multiple skills in a single conversation. Google will automatically convert existing Gems into skills.
According to Google’s Help Centre, Gems will disappear in three stages:
November 2026 for personal Google accounts;
March 2027 for Google Workspace for businesses, enterprise and non-profits;
June 2027 for Google Workspace for Education.
For education, Google states that from as early as 1 June 2027, users will no longer be able to create, edit or use Gems. Gems will also disappear from Google Classroom and from learning environments linked via Gemini LTI. Converted Gems will return as draft skills which the creator must check and activate themselves.
What is not yet included
Google is open about what skills cannot do yet. Guided Learning, Canvas, Deep Research, and creating video and music currently do not work with skills. According to Google, sharing will be available “in the coming weeks”, as will adding files from Google Drive. So sharing is not disappearing permanently, but at this moment it is not yet available. Skills also do not yet synchronise between the Gemini app and Workspace apps.
Why this is inconvenient for organisations
A Gem or GPT with a single clear task was easy to explain. Open the link and get started. In education, that might be a reflection coach or a conversation partner to practise a language. In a municipality, an assistant that summarises policy documents or helps answer residents’ questions. In healthcare, a tool for preparing handovers. In a business, a resource for the company’s own collective agreement or staff handbook. With the new approach, everyone has to figure out again how to activate instructions, which features are included, and how colleagues, pupils or customers gain access. And anything that was linked in Classroom, a learning environment or a workflow will have to be set up again.
OpenAI shows just how much is involved. According to its own migration guide, a converted plugin starts off private, must be shared again, and recipients have to install it—unless the administrator handles this. Custom integrations (custom actions) are not transferred automatically and behaviour may change. For Enterprise environments, custom GPTs will stop on 11 December 2026. So converting does not mean everyone can simply continue as before.
Who decides how your AI works?
Two of the largest providers are changing, within a few months, the foundation on which schools, governments, healthcare institutions and businesses have built their AI assistants. That is not a mistake by those companies; it is how their products evolve. But it does show what being dependent on a single provider means. Their product decisions become your migration project, on their timetable and with their choices about what is and is not included.
There is also the question of where data is stored and under which jurisdiction it falls. For education, it concerns pupil data; for government, citizen data; for healthcare, medical information; and for businesses, commercially sensitive information. American providers are subject to US law, even if their servers are in Europe. A European provider with hosting in the EU and an open foundation reduces both risks.
That is also how we have set up our four assistants: EduGPT for education, GovGPT for government, CareGPT for healthcare and OrgGPT for organisations and businesses.
Hosted in Europe. All four run on Scaleway in Paris, with European and open models on Scaleway as standard.
Your assistants remain yours. An assistant created and shared by a teacher, policy officer or healthcare professional with colleagues will continue to work. Instructions and knowledge base are in the organisation’s own environment, always accessible and portable, and the platform runs on an open-source basis. You are therefore not tied to a single supplier.
Open standards. Integrations with sources and systems use the open Model Context Protocol, for example with SLO, Edurep and Edusources in education. What you build also works outside our environment.
No surprises. Changes in an organisation’s environment go through a fixed change process and are coordinated in advance, not at the moment an American provider decides.
What you can do now
If you use Gems or custom GPTs in the classroom, at the town hall, in healthcare or at the office, now record the instructions for your most important assistants in your own document. This way, you are not dependent on how a conversion turns out, and you can take the working method with you to any environment. If you want to see what this looks like in a European environment, visit edugpt.nl, govgpt.nl, caregpt.nl or orggpt.eu, or contact us.
Sources
Google, Let skills in Gemini tackle your most repetitive tasks (30 September 2026)
Google, Gemini Apps Help, About the transition from Gems to skills
Google Workspace Admin Help, About the transition from Gems to skills
Image for illustration, created with AI