I rely heavily on Custom GPTs for a professional medical-documentation workflow, and I am extremely concerned about their retirement because the current Plugin replacement does not reproduce the capabilities that made Custom GPTs effective for my work.
My Custom GPTs have been used successfully for more than two years to transform clinical evaluation information into highly structured documentation. This workflow saves approximately half of the documentation time associated with my evaluations. It is not a casual use case; it is an important part of my daily professional workflow.
There are four major problems with the current replacement.
1. New conversations need to be truly isolated.
Each new conversation represents a different client. With Custom GPTs, each conversation starts fresh and does not use saved Memory or previous conversations. That behavior is extremely important in a clinical workflow.
I want Memory and Reference Chat History enabled for my normal ChatGPT use, but I need individual professional workflows to be able to opt out completely.
A Plugin should be able to operate in a stateless mode where it has access only to:
-
its permanent instructions;
-
its designated knowledge/reference files;
-
its designated template; and
-
information supplied in the current conversation.
It should not use previous client conversations, saved Memory, unrelated chats, Project context, or other historical information.
2. Plugins do not currently reproduce Custom GPT template fidelity.
My Custom GPTs can treat an uploaded document as an authoritative template and consistently populate it using information from the current evaluation.
The migrated Plugins frequently change headings, reorganize sections, paraphrase required wording, omit information, alter the structure, or introduce information that was not provided for the current client.
I tested this using the same input side-by-side with the original Custom GPT and migrated Plugin. I also supplied OpenAI Support with the original template and both outputs. The difference is reproducible.
If the Plugin cannot reliably reproduce a relatively short scooter evaluation template, I cannot trust it with much longer and more complex power-wheelchair evaluations.
I need a mechanism to designate a file as an authoritative or immutable output template, rather than merely allowing it to be treated as general reference material.
3. The new Plugin editing experience has substantially reduced my ability to control and debug my workflow.
This has become one of my biggest frustrations with the migration.
With Custom GPTs, I could directly inspect and edit the exact instructions controlling the GPT. If a particular sentence caused a problem, I could change that sentence, save it, test the result, and continue refining it.
That level of precise control is essential for advanced workflows.
With Plugins, much of the configuration process is now mediated through the Plugin editor/Plugin Creator conversation. Instead of directly modifying the configuration myself, I am often asking an AI system to modify another AI system.
That introduces another probabilistic layer into something I am trying to make more deterministic.
I want an advanced editing mode that lets me directly inspect and modify the complete Plugin configuration, including its instructions, skills, references, template rules, and other relevant behavior.
Conversational editing can remain available for users who prefer it, but it should not replace direct configuration access.
Ideally, advanced users should be able to work with the Plugin almost like code: inspect the complete configuration, make precise edits, compare revisions, and test them.
4. Migration should preserve the professional workflows that already work.
The original Custom GPT system solved this problem remarkably well:
Permanent instructions + permanent knowledge files + strict templates + a fresh context for every new conversation.
I am not asking for a completely new capability. I am asking OpenAI to preserve a capability that already exists and that I have successfully depended on for years.
Before Custom GPTs are fully retired, I strongly hope OpenAI provides at least one of the following:
-
a true stateless/isolated Plugin mode;
-
per-Plugin control over Memory and Reference Chat History;
-
authoritative/immutable template files;
-
stronger deterministic reference-file handling;
-
direct advanced editing of Plugin configuration;
-
an option to preserve the existing Custom GPT execution model for workflows that require it; or
-
another supported OpenAI product that provides the same isolation, template fidelity, and direct configuration control.
I have already completed troubleshooting with OpenAI Support and provided a synthetic template and side-by-side Custom GPT/Plugin outputs demonstrating the problem.
I genuinely want to continue using OpenAI. Custom GPTs have made a major positive difference in my work and have saved me a significant amount of time. However, if the replacement cannot provide comparable reliability and control before Custom GPTs are removed, I will have to find another platform for a workflow that currently depends on ChatGPT.
I would especially appreciate guidance on whether OpenAI plans to provide an advanced or developer-oriented alternative that preserves the direct control and isolated execution model of Custom GPTs.