I want to register strong feedback about OpenAI’s decision to stop allowing new Custom GPT creation and public publishing from personal ChatGPT accounts.
For me, Custom GPTs were not a novelty feature. They became a remarkably effective way to build and share lightweight experimental AI tools with students, researchers, collaborators, and the broader public.
I work in animal-welfare science, and over the last few years I have been building a growing library of public resources around animal-welfare methods and research. Custom GPTs were particularly valuable because they let us take a carefully designed set of scientific instructions, knowledge files, safeguards, and workflows and turn it into something other people could use immediately through a simple link.
This was an unusually powerful combination: very low development overhead, no hosting burden, no API billing for the creator, and a familiar interface for users. It made it possible for a small scientific organization to experiment rapidly with tools that otherwise would have required software development and maintenance.
The timing makes this especially frustrating. Only last week I was presenting at a conference and explicitly praising Custom GPTs to the audience, encouraging researchers and students to use them as a practical way to build and disseminate specialized AI tools. A few days later I discovered that I can no longer create a new GPT and share it with those same people. It is difficult not to feel rather foolish for having recommended a platform feature immediately before OpenAI withdrew the very capability that made it useful for that purpose.
The problem is not simply that one feature disappeared. The larger issue is trust and platform continuity.
People have invested substantial time building libraries of GPTs, scientific resources, teaching tools, workflows, documentation, links, and communities around a product model that OpenAI actively encouraged. Removing the ability to create and publish new ones from personal paid accounts changes the basic proposition after that investment has been made.
Workspace Agents are not an adequate replacement for this use case. They may be excellent for internal organizational workflows, but they do not solve the core problem of building an experimental AI resource and making it available to the broad public through a simple link.
Moving creators to Business, Enterprise, or Edu workspaces is also not a satisfactory answer if public dissemination is not the objective of those products—and especially if Workspace Agents are the long-term successor to GPTs for managed workspaces.
What I would strongly ask OpenAI to consider is one of the following:
- restore public/link sharing for newly created GPTs on paid personal plans;
- preserve Custom GPTs explicitly as the lightweight public-builder layer, even if Workspace Agents become the enterprise successor;
- provide a genuine public successor that supports instructions, knowledge files, branding, a shareable URL, and creator-independent inference costs;
- or, at minimum, provide a clear long-term roadmap and migration/export path before removing capabilities on which creators have built public resources.
Custom GPTs filled a genuinely useful niche: the space between a prompt and a full software product. They allowed researchers, educators, domain experts, and small organizations to turn expertise into usable public AI tools without becoming software companies.
I hope OpenAI reconsiders abandoning that niche.
In our case, we are now actively evaluating Poe, Gemini Gems, Claude Artifacts, and standalone model-agnostic web applications as replacements—not because the underlying OpenAI models are inadequate, but because we can no longer rely on ChatGPT as a stable publishing platform for this kind of work.
I would much rather continue building these resources within the OpenAI ecosystem. But for that to make sense, creators need some confidence that the public distribution mechanism they build around will still exist a few months later.