OpenAI Is Killing Custom GPTs. Here’s Your Exit Checklist
On April 22, 2026, OpenAI announced Workspace Agents and buried a sentence in the launch post that stopped a lot of people mid-scroll: Custom GPTs are being deprecated for Business, Enterprise, Edu, and Teachers accounts.
The announcement was framed as an “evolution of GPTs.” The practical reality is blunter. Every new feature OpenAI is building goes into Workspace Agents. Custom GPTs are in maintenance mode. No new GPT Builder investment has been announced, and no one at OpenAI is suggesting otherwise.
If you’re on an individual Plus or Pro plan, you’re not on a forced migration path today. But the direction is clear, and if you built a Custom GPT you actually rely on, for client work, internal processes, or anything with real knowledge files or Actions wired up, the time to audit is now, not when OpenAI sets a hard deadline.
This is that audit.
What’s Actually Happening
OpenAI’s Workspace Agents are more capable than Custom GPTs in most respects. They run persistently in the cloud, connect to Slack, Gmail, Salesforce, and Google Drive, operate on a schedule, and take actions after you close the browser. Custom GPTs wait for you to show up and prompt them. Workspace Agents don’t.
That’s a genuine capability jump. It’s also a deeper lock-in and a more complex setup.
For individual users, Workspace Agents aren’t even available yet. Business-plan users have been on a deprecation path since April, with no forced migration date named and a one-click conversion tool reportedly on the way, but not shipped as of late September 2026.
The problem for AU small businesses isn’t the conversion tool. The problem is that the conversion tool doesn’t handle everything. Live system instructions, uploaded knowledge files, and Actions configurations don’t automatically land somewhere useful. You need to know what you built before you migrate it.
Why AU SMBs Need to Move Now
Most migration guides assume you’re a developer or an enterprise IT team. Most AU SMBs who built Custom GPTs aren’t either of those things.
You built a GPT to handle intake questions. Or to brief a VA. Or to draft responses in your brand voice using a knowledge file full of your own content. That’s real work, and it lives inside a platform you don’t control, on a format that’s being wound down.
Three things don’t migrate cleanly to anything:
Uploaded knowledge files. Files you uploaded to GPT Builder aren’t stored anywhere you control. When the format is retired, those files go with it unless you retrieve them now.
System instructions. Your carefully written prompt, the one that took you three iterations to get right, exists only inside GPT Builder. It won’t appear in Workspace Agents automatically.
Actions configurations. If you wired your Custom GPT to an external service via Actions, that configuration needs to be rebuilt manually in whatever replaces it.
None of that is catastrophic. All of it requires you to act before the deprecation hits.
The Exit Checklist
Work through this in order. It takes 30 to 90 minutes depending on how many GPTs you’ve built.
1. Inventory your Custom GPTs
Go to chat.openai.com → your name → My GPTs. List every GPT you’ve built. Note whether each one is published publicly, shared with a link, or private. The shared and public ones matter most, someone else may be using them.
2. Export all system instructions
Open each GPT in GPT Builder. Go to Configure. Copy the full contents of the Instructions field into a document, Google Doc, Notion, Obsidian, anything you actually use. Don’t skip the Conversation Starters or the Name and Description fields. Those are your brand, not just settings.
3. Download all uploaded knowledge files
In GPT Builder, scroll to the Knowledge section. Download every file. Save them somewhere you control, not another cloud service you don’t pay for. If you used a client’s documents in a knowledge file, this step is also a privacy compliance step (more on that below).
4. Screenshot or export Actions configurations
If you set up Actions, photograph or document the schema, the authentication method, and the endpoint URLs. This is the most technical piece and the one most likely to break on migration. Having the spec somewhere means you can rebuild it.
5. Test your GPT inside a ChatGPT Project
Create a new ChatGPT Project and paste your system instructions into the Custom Instructions field. Run your five most common use cases. Note what breaks. Projects don’t have file-based retrieval yet, so knowledge-file-dependent workflows will behave differently. That’s useful to know now.
6. Rebuild critical workflows as Projects with prompt packs
For most AU SMBs, the lowest-friction path isn’t Workspace Agents. A ChatGPT Project with solid custom instructions and a saved prompt pack covers 80% of what a Custom GPT was doing. It’s less elegant but more portable and less likely to disappear.
7. If you built Custom GPTs for clients, notify them now
If a client is using a Custom GPT you built and shared, they need to know what’s changing. Don’t wait for OpenAI to send that email. You built it, you should be the one explaining what happens next and what you’re doing about it.
What Actually Replaces Custom GPTs
There’s no clean one-to-one replacement, which is the honest answer the OpenAI announcement glossed over.
ChatGPT Projects are the closest replacement for most use cases. No file-based retrieval yet, but custom instructions and conversation history carry across sessions. Available to Plus and Pro users now.
Workspace Agents are more powerful but designed for Business/Enterprise accounts, credit-based pricing, and workflows that need to run without a human triggering them. Overkill for a sole trader. Possibly right for a small team with an IT-literate person to set things up.
Prompt pack + memory combos are the lowest friction option. Save your best prompts as files you paste in when needed. Use ChatGPT’s memory feature to hold persistent context. It’s manual but it’s yours.
For most AU SMBs, start with Projects. Rebuild the instructions properly. Test for two weeks. Then decide whether Workspace Agents are worth the complexity.
The AU Privacy Note
If your Custom GPT’s knowledge files contained client data, invoices, proposals, personal details, anything that would be covered under the Australian Privacy Principles, this migration is also a data handling event.
Before you move those files to a new storage location, check whether that location meets your APP obligations. A Google Doc isn’t automatically compliant just because it’s convenient. If you’re uncertain, the OAIC has guidance on cloud storage and third-party processors. When in doubt, get advice before you copy the data somewhere new.
What to Do This Week
You don’t need to rebuild everything today. What you do need is your instructions and files out of GPT Builder and somewhere you actually control. That’s the one action worth doing before anything else.
The rest, deciding on Projects vs. Workspace Agents, notifying clients, rebuilding Actions, can happen over the next few weeks. The urgent thing is not losing work you’ve already done to a platform transition you didn’t choose.
📖 More deep dives: News Deep Dives
📊 Compare AI tools side by side | 💼 Free resources & AI prompt packs
More hot takes: Hot Takes →
If this interests you
- Your Small Business Website Is Now Free. That’s Not the Point.
- OpenAI Said Sorry. Here’s What That Actually Means for Australian Businesses.
- Obscurity Is Not Security: Why Australia’s Tech Titans Are Calling Out the Medicare AI Panic
More hot takes: Hot Takes →
