Fastest fix: hard-refresh the page (Cmd/Ctrl + Shift + R), then open the same GPT in an incognito window signed into the same account. That clears stale cache and disables extensions in one move and resolves the most common “blank GPT” case. If only the GPT’s action fails (the chat itself works), the problem is the creator’s backend or a known mobile bug — not your setup. Details below.
A Custom GPT isn’t a standalone service. It’s a configuration layer on top of your normal ChatGPT account: system instructions + knowledge files + optional actions (calls to an external API the creator hosts). Any one of these layers can break in a way that looks like “the GPT is broken,” and the symptoms overlap. Correctly identifying which layer failed is the whole job — so this article is organized symptom → cause → fix.
Symptom map
Match what you see to the most likely cause, then jump to the fix step.
| Symptom | Most likely cause | Go to |
|---|---|---|
| GPT card opens blank / spinner never resolves | Browser cached an old manifest, or an extension blocked a resource | Step 1 |
| Chat works, but the action returns “request failed” / “I can’t reach that” | Creator’s action backend is down, or auth expired | Step 4 |
| Action works on desktop but is silently ignored in the phone app | Known mobile-client action regression (since ~May 2026) | Step 2 |
| ”This GPT is no longer available” | Creator deleted it, or OpenAI removed it for policy | Step 3 |
| ”GPT not found” / bounced on open | Set to Private or link-only; your account isn’t allowed | Step 3 |
| ”You’ve reached your limit” / upgrade prompt | Free-tier message cap hit, or the GPT uses a paid capability | Step 3 |
Common causes, in rough order of frequency
1. Browser cached an old GPT manifest
The GPT’s instructions and action schema get cached client-side. After the creator updates them, you can still be served the stale version — out of sync with the backend — so the GPT goes blank or behaves oddly.
How to verify: hard refresh with Cmd/Ctrl + Shift + R. If it loads after that, it was the cache.
2. A browser extension or network is blocking a resource
Ad/script blockers (uBlock Origin, AdGuard, Privacy Badger) sometimes strip a script or iframe the GPT page needs, leaving you with an endless spinner. Corporate firewalls and VPNs can also block either ChatGPT itself or the action’s external API host.
How to verify: open the GPT in an incognito window (extensions are off by default). If it works there, an extension is the culprit. If incognito still fails on office Wi-Fi, try a phone hotspot to rule out the network.
3. Your account doesn’t have access
Browsing and using Custom GPTs in the GPT Store has been free since 2024, so a plain “this GPT needs Plus” wall is now rare. When you can’t open a specific GPT it’s usually because the creator set it to Only me or Anyone with a link and your account isn’t on the list — you’ll see “GPT not found” or get bounced.
How to verify: ask the creator to send you the share link directly. If it still won’t open, your account doesn’t have access. Separately, if you see an upgrade or “you’ve reached your limit” prompt, that’s a Free-tier message cap (roughly 10 messages every 5 hours on the default model as of June 2026 before it drops to a lighter fallback), or the GPT internally invokes a paid feature like advanced reasoning, image generation, or browsing.
4. The action’s backend API is down or auth failed
A Custom GPT’s actions call HTTPS endpoints the creator hosts elsewhere. If their service is offline, the TLS certificate expired, or the API key/OAuth token was revoked, the action returns “request failed” (or the model replies “I couldn’t reach that”) — while the non-action parts of the GPT still work fine.
How to verify: ask the GPT to print the URL it’s calling, then open that URL yourself and check for an HTTP 200. A 4xx/5xx or a TLS warning means it’s the creator’s backend, not you.
5. The action is ignored only in the mobile app (known regression)
As of June 2026 there’s a documented mobile-client bug: starting with ChatGPT Android app build 1.2026.139 (rolled out ~May 25, 2026) and on some Android mobile browsers, configured actions are silently not invoked — the model replies in plain language that it “can’t access the verification endpoint” even though the same GPT and the same endpoint work perfectly on desktop web. This most often hits GPTs that use external auth (OAuth or email/OTP).
How to verify: run the identical prompt on desktop web (chatgpt.com). If it calls the action on desktop but not in the app, you’ve hit this regression — the fix is to use desktop until OpenAI ships an app update.
6. The GPT was deleted or removed
If the creator deleted it, or OpenAI took it down for a policy violation (impersonation, deception, copyright), you get “This GPT is no longer available.”
How to verify: the error explicitly says “no longer available” or “removed.” There’s no client-side fix — see Step 5 for replacements.
Shortest path to fix
Do the 30-second checks first.
Step 1: Hard refresh, then incognito
- On the GPT page, press
Cmd/Ctrl + Shift + Rto bypass cache. - Open an incognito/private window (extensions are disabled there), sign in with the same account, and open the GPT.
- Still blank? Skip to Step 3.
If incognito works, it’s cache or an extension:
- Cache: open DevTools → Application → Storage → Clear site data for
chatgpt.com. - Extensions: allowlist
chatgpt.comin uBlock Origin / AdGuard, or disable the blocker on that domain.
Step 2: Try a different device or surface
If desktop web fails, open the GPT in the mobile ChatGPT app with the same account, or vice versa.
- App works, desktop doesn’t → it’s a desktop browser-layer issue, return to Step 1.
- The chat loads everywhere but the action only fails in the phone app → you’ve hit the mobile action regression (cause #5). Use desktop web for now.
- Nothing works anywhere → Step 3.
Step 3: Confirm account access and tier
- Read the actual error text. “GPT not found” or a bounce = a permissions issue; ask the creator for a fresh share link and confirm you’re on the right account.
- An “upgrade” or “you’ve reached your limit” prompt = a Free-tier message cap or a paid capability the GPT uses. Wait for the window to reset, or upgrade (ChatGPT Plus is $20/mo as of June 2026).
- Check your tier under profile menu → Settings → Subscription.
Step 4: Diagnose an action failure
When the main chat works but the action returns “request failed”:
- Ask the GPT: “What is the exact URL of the action you’re calling?” — the model will report it.
- Paste that URL into a new browser tab or a tool like Postman and confirm the backend returns
200. A timeout,401/403,5xx, or a cert warning points at the creator’s backend. - Backend down /
401 unauthorized→ only the creator can fix it. Report it to them with the URL and status code you saw.
If you are the creator, open the GPT in edit mode and go to Configure → Actions:
- Confirm the OpenAPI schema (JSON/YAML) still validates — the editor lists the detected operations when it’s valid and shows errors when it isn’t.
- Check Authentication (API Key / OAuth / None) hasn’t expired; a revoked key or stale OAuth client is the usual cause of
401. - Use the Test control on an individual action row to dry-run the call and read the raw response.
- Verify the privacy-policy URL and the action’s domain are still reachable.
Step 5: Replace it or rebuild your own
If a GPT is reliably broken or gone and you depend on it:
- Search the GPT Store for an alternative — entries from verified/official builders tend to be more stable.
- Copy the GPT’s instructions (if you saved them) and paste them as a system prompt in a normal chat.
- Build your own Custom GPT from My GPTs → Create, internalizing the core prompt so you control the lifecycle.
How to confirm it’s fixed
- The GPT card opens to a working chat with no spinner.
- A test message that exercises an action returns real data (not “request failed” or a vague natural-language apology).
- The same flow works in a clean incognito window, confirming it wasn’t a one-off cache state.
Prevention
- Pin only GPTs you use long-term from trusted creators — third-party GPTs can disappear without notice.
- Save the instructions/prompt of any critical-workflow GPT outside ChatGPT (Notion, Apple Notes) so you can rebuild it if it breaks.
- For your own GPTs, version the action schema URL (
/v1/manifest.json) and bump the version on schema changes instead of overwriting in place. - Export important GPT configurations periodically; OpenAI occasionally corrupts or loses GPT data.
- For business-critical actions, don’t rely solely on the GPT call — monitor the backend independently so you learn about an outage before your users do.
FAQ
Why is my Custom GPT a blank screen but regular ChatGPT works?
Almost always cache or an extension. Hard refresh with Cmd/Ctrl + Shift + R, then try incognito. If incognito loads it, clear site data for chatgpt.com or allowlist it in your ad blocker.
My GPT chats fine but says it can’t reach its action — is that my fault?
No. Actions call the creator’s own external API. If it’s down, has an expired certificate, or revoked credentials, the action fails while the rest of the GPT works. Verify by opening the action URL yourself and checking for a 200.
The action works on my laptop but not in the ChatGPT app — bug?
Yes, as of June 2026 there’s a known mobile regression (since app build 1.2026.139, ~May 2026) where actions are silently skipped on Android clients with the model saying it “can’t access the verification endpoint.” Use desktop web until OpenAI ships an app fix.
Do I need ChatGPT Plus to use Custom GPTs? No — using GPTs from the Store has been free since 2024. You’ll only see an upgrade prompt if you hit the Free-tier message cap (around 10 messages per 5 hours on the default model as of June 2026) or if that specific GPT invokes a paid capability like advanced reasoning, image generation, or browsing.
It says “This GPT is no longer available” — can I get it back? No. That means the creator deleted it or OpenAI removed it for policy reasons. Find an alternative in the Store, or rebuild it yourself from any instructions you saved.
Related
- ChatGPT projects guide
- ChatGPT slow response
- ChatGPT beginner guide
- ChatGPT prompt improvement
- ChatGPT model selection guide
Tags: #ChatGPT #Debug #Troubleshooting