Custom GPT No Longer Available: Access and Recovery Checks

A custom GPT that is no longer available can be unavailable for several different reasons: the creator removed it, you lost access to the account it was linked to, the link changed, or the GPT was based on features being retired. The right recovery step depends on which of those applies. Most of what matters — the instruction text, the reference material, the behavior rules — can be preserved outside the platform regardless of what happens to any specific GPT.

Tools like fenbs keep the approved instructions and the context behind them outside the GPT configuration, so losing a particular GPT does not mean losing the work that went into it.

Why a custom GPT may be unavailable

The most common causes, in rough order of frequency:

• The creator unpublished or deleted it. GPTs shared via link rely on the creator keeping them available. If they remove it, the link stops working.
• The GPT is from a shared account you no longer have access to. If a colleague built it on a work account that was deprovisioned, the GPT goes with the account.
• The URL or link is outdated. GPTs accessed through bookmarks or shared links sometimes break after account changes.
• A policy violation triggered removal. OpenAI can remove GPTs that violate its usage policies.
• The underlying feature is being retired. OpenAI announced that custom GPTs will retire on December 11, 2026. Check the OpenAI help center for the current status and any migration options they offer.

What you can recover

If you built the GPT yourself and still have access to your account, you can open the GPT editor and copy the instructions. Do this before the retirement date for any GPT you still use. The instruction text is the most valuable thing; save it somewhere you control.

If someone else built it and shared it with you, the instruction text is not accessible to you through the interface. What you can recover: the behavior you remember it having, the outputs you saved from past sessions, and any documentation the creator provided. Use those to reconstruct the instructions from scratch.

Knowledge files uploaded to a GPT you own are recoverable from the GPT editor as long as the GPT still exists in your account. Download them before any account or configuration changes.

What you cannot recover

You cannot view the instructions of a GPT built by someone else, even if you used it regularly. You cannot recover a deleted GPT from a deprovisioned account through OpenAI support in most cases. Conversation history from sessions with the GPT is in your chat history if it was not deleted, but the configuration that produced those responses is not.

This is why saving the instruction text outside the platform matters before you need it. A GPT you rely on for regular work should have its instructions saved in a place you control, not only in the OpenAI interface.

Rebuilding from what you know

If you need to rebuild a GPT whose instructions you cannot access, start from behavior: what did the GPT always do, what did it never do, and what format did its responses follow? Write those down as rules first. Then add examples of inputs and outputs you remember or saved. That gives you a starting configuration you can test and refine.

The guide on building a custom GPT covers the configuration steps. The guide on Claude Projects vs. custom GPTs is useful if you are considering alternatives for keeping persistent assistant configurations.

Planning ahead for platform changes

The retirement of custom GPTs is a reminder that features tied to a specific platform can change or end. The instructions, constraints, and format rules you develop through testing are your own work, not the platform's. Keeping them in a place you control is a maintenance habit worth building for any AI tool you rely on regularly.

When a GPT is working well, take ten minutes to copy the instruction text, note which knowledge files it uses, and save the test cases that define its expected behavior. Do this before any announced change, not after. A configuration that exists only in one platform interface is one you could lose.

If the creator of a GPT you used regularly has unpublished it, the most practical step is to build a new configuration that achieves the same outcomes. Start by listing the outputs the original produced consistently and the instructions you would give to produce the same outputs manually. Those written-out instructions are the starting configuration for a new assistant on any supported platform.

How to do this with fenbs

1. Create a free account and open a project for the GPT you want to preserve.
2. Copy the current instruction text from the GPT editor and save it to Skills and prompts with the date and what the GPT is for.
3. Note the knowledge files it uses in a comment: names, dates, what each one covers.
4. Record key behavioral decisions in Decisions and rules: "always responds in three sections," "never includes legal interpretations." These survive any platform change.
5. Connect your AI assistant from Integrations. When a GPT becomes unavailable, your assistant can start a new session from the saved configuration rather than starting from nothing.