You submit a prompt to Gemini and the spinner runs forever, or you get an empty reply card. This is almost always a network or account problem, not a broken model. Gemini streams its answer over a long-lived SSE (server-sent events) connection; anything that severs that connection mid-stream — corporate proxy, ad blocker, Workspace policy, regional gating — leaves the UI hanging with nothing to render.
Fastest fix: open gemini.google.com in a fresh Chrome Incognito window (extensions off) on your phone’s hotspot. If it responds there, the culprit is your normal network or a browser extension, and the rest of this page tells you which one. If it still hangs everywhere, jump to the status check and the safety-filter test.
Diagnose in this order: account → region → network → browser.
Symptoms this page covers
- Spinner keeps spinning after you hit send
- Reply card appears but stays empty
- Phone works, laptop browser doesn’t (or vice versa)
- Same laptop: Incognito works, your main profile doesn’t
- Mobile app: response stalls partway, or “Something went wrong”
Which bucket are you in?
Use this to skip straight to the fix that matches your situation.
| What you observe | Most likely cause | Go to |
|---|---|---|
| Works on phone hotspot, fails on office/school WiFi | Proxy/firewall blocks Google AI endpoints | Step 1 |
| Incognito works, main profile fails | Browser extension kills SSE | Step 2 |
| Personal account works, work account fails (yellow “managed” banner) | Workspace admin disabled Gemini | Step 3 |
| Fails everywhere, VPN to US/UK fixes it | Regional gating | Step 4 |
| Fails for everyone right now (Twitter/Reddit lit up) | Backend outage | Step 5 |
| One specific prompt returns blank, others work | Safety filter blocked it silently | Step 6 |
| Mobile app only; browser is fine | App cache / outdated app | Step 7 |
| API call hangs or errors | Use the API error table | Step 9 |
Common causes, by frequency
1. Corporate proxy or firewall blocks Google AI endpoints (most common)
Office and school networks frequently block *.googleapis.com or terminate long-lived SSE connections. Because Gemini streams its reply, a killed connection shows up as a blank card.
How to judge: works on a phone hotspot but breaks on office WiFi. In DevTools (F12) under Network, the streaming request to StreamGenerate or *.googleapis.com shows (canceled) or (pending).
2. A browser extension breaks streaming
Content blockers and anti-tracking extensions (uBlock Origin, Privacy Badger, AdGuard, Ghostery, DuckDuckGo Privacy Essentials) can terminate the SSE stream.
How to judge: Incognito (extensions off by default) works; your main profile doesn’t.
3. Workspace account has the Gemini app disabled
If you’re signed in with a work or school Google account, IT may have the Gemini app turned off. You’ll see a blank card, “feature unavailable,” or a yellow “managed by your organization” banner. (The consumer subscription that unlocks Gemini 3.1 Pro is now called Google AI Pro, renamed from “Gemini Advanced” / “Google One AI Premium” in early 2026 — but Workspace gating is a separate admin switch, not a billing problem.)
How to judge: a personal Google account works; the work account doesn’t.
4. Regional gating
Gemini rolls out per region. A handful of countries (for example Russia, Iran, North Korea) remain blocked as of June 2026.
How to judge: connecting through a VPN to the US or UK makes it work.
5. Account flagged for suspicious activity
A recent device, IP, or country change can trigger a temporary lock.
How to judge: there’s a “Security alert” at accounts.google.com.
6. Backend outage (rare)
Occasionally Google’s service really is down. June 2026 has already seen short incidents where content generation failed with 503 errors.
How to judge: check the status pages in Step 5.
7. A prompt silently hits the safety filter
Some sensitive prompts are blocked without an error. The response comes back with an empty parts field and finishReason: SAFETY, which the UI renders as no answer at all.
How to judge: an innocuous prompt (“what’s the weather today”) works fine.
Shortest path to a fix
Ordered by how fast each test verifies.
Step 1: Switch networks (phone hotspot)
Disconnect office/school WiFi
Enable your phone's hotspot
Connect the laptop to the hotspot
Open gemini.google.com and retryIf it works, your network is blocking Gemini. Ask IT to allowlist:
gemini.google.com*.googleapis.comaistudio.google.com*.googleusercontent.com
Step 2: Different browser / Incognito
Open Chrome Incognito (extensions off by default)
Go to gemini.google.com and sign in
RetryIf it works, an extension in your main profile is the culprit. Re-enable extensions one at a time to find it. Usual suspects: uBlock Origin, Privacy Badger, AdGuard, Ghostery, DuckDuckGo Privacy Essentials. Once identified, add gemini.google.com to that extension’s allowlist instead of disabling it globally.
Step 3: Different account
gemini.google.com → avatar (top right) → switch to another Google account
(personal / student / other)
RetryIf a personal account works and the work account doesn’t, it’s a Workspace policy. A Workspace admin enables it at:
Admin Console (admin.google.com) → Menu (≡) → Generative AI → Gemini app
→ Service status → On for everyone (or the right org unit)You can’t self-enable this — only an admin can.
Step 4: Different region
Connect through a VPN to the US or UK and test. If it works, you’re hitting regional gating; switch your region permanently for a long-term fix.
Step 5: Check Google service status
Two different pages cover two different surfaces — check the one that matches how you use Gemini:
Consumer Gemini app → workspace.google.com/status (look for "Gemini")
Vertex AI / Gemini API → status.cloud.google.com (look for "Vertex AI" /
"Generative Language API")Green means the service is healthy and the problem is on your side. Orange or red means wait it out — nothing you do locally will help.
Step 6: Innocuous prompt to test the safety filter
Ask Gemini: "what's the weather today"
Works → your previous prompt was blocked by the safety filter
Still blank → not a prompt-content issue; go back to Steps 1-5If the filter is over-triggering on a legitimate prompt, rephrase it, remove quoted material that looks unsafe out of context, or split a long prompt into smaller turns.
Step 7: Mobile app — clear cache and update
If only the phone app fails (browser is fine), the app’s local cache or an outdated build is usually to blame.
Android: Settings → Apps → Gemini → Storage & cache → Clear cache
Repeat for the Google app
iOS: delete and reinstall the Gemini app
Both: update Gemini AND the Google app to the latest versionOn Android, if the assistant still won’t respond, reset the digital assistant: Settings → Google → Settings for Google apps → Search, Assistant & Voice → Digital assistants from Google — switch away from Gemini, wait ~10 seconds, switch back.
Step 8: Clear browser cookies and cache
Chrome → Settings → Privacy and security → Clear browsing data
Scope to sites: [*.]google.com
Restart the browser and sign back inA stale auth cookie can leave the stream unable to authenticate, which also reads as a blank reply.
Step 9: API users — call the API directly
If the Web UI keeps failing but you need an answer now, the API bypasses the browser’s blocking chains entirely. The current Python SDK is google-genai:
# pip install google-genai
from google import genai
client = genai.Client(api_key="YOUR_KEY")
response = client.models.generate_content(
model="gemini-3.1-pro",
contents="hello",
)
print(response.text)If the API itself errors, match the code:
| Code / message | Meaning | What to do |
|---|---|---|
429 RESOURCE_EXHAUSTED | Rate limit hit (common on the free tier) | Back off and retry; raise your quota or add billing |
503 UNAVAILABLE (“overloaded”) | Service temporarily out of capacity | Retry with exponential backoff |
504 DEADLINE_EXCEEDED | Prompt/context too large to finish in time | Shorten the prompt or trim context |
403 ... Generative Language API has not been used in project ... or it is disabled | API not enabled for your Cloud project | Enable the Generative Language API in the Cloud Console |
If response.text is empty with no error, check response.candidates[0].finish_reason — SAFETY means the safety filter blocked it (see Step 6).
How to confirm it’s actually fixed
Run one clean test that exercises streaming end to end:
- In the window/profile/network you just changed, send: “Write three short bullet points about coffee.”
- You should see tokens stream in within a second or two and a complete, non-empty reply.
- Send one follow-up (“now make it two bullets”) to confirm the connection stays alive across turns.
If both turns return full answers, the blocking chain is gone.
When it isn’t on you
- Workspace-managed accounts have Gemini gated by IT policy; you can’t self-enable it.
- Regional gating is a Google rollout decision; a VPN is the only workaround.
- Backend outages can only be waited out.
Prevention
- Know which Google account is personal vs managed, and use the personal one for AI work.
- Bookmark both workspace.google.com/status (consumer app) and status.cloud.google.com (API).
- Keep a clean Chrome profile for AI with no content blockers, or allowlist
gemini.google.comin the ones you keep. - On locked-down networks, try aistudio.google.com; it’s often more reliable than
gemini.google.com. - API users: keep the API path as a fallback for when the Web UI gets flaky.
FAQ
Why does Gemini show a blank reply card with no error?
The most common reason is a severed SSE stream — a proxy, firewall, or extension killed the streaming connection, so the UI has nothing to render. The second most common is a safety-filter block, where the response returns with empty parts and finishReason: SAFETY. Test with a phone hotspot first, then an innocuous prompt.
Gemini works on my phone but not my laptop. Why? Your laptop is on a network or browser profile that’s interfering. Phones on cellular data dodge corporate proxies, and the mobile app has no desktop extensions. Run Steps 1 and 2 on the laptop: hotspot test, then Incognito.
My work account says “managed by your organization” and won’t respond. Can I fix it myself? No. A Workspace admin has to turn on the Gemini app at Admin Console → Menu (≡) → Generative AI → Gemini app → Service status → On for everyone. As a workaround, use a personal Google account for AI tasks.
Is Gemini down right now, or is it just me?
Check the status page that matches your usage: workspace.google.com/status for the consumer Gemini app, status.cloud.google.com for the Vertex AI / Generative Language API. If both are green, the problem is local — work through Steps 1 to 8.
The API returns an empty response.text but no error. What’s wrong?
Inspect response.candidates[0].finish_reason. SAFETY means the safety filter blocked the output; MAX_TOKENS means it ran out of room (raise the output limit); RECITATION means it matched protected content. Each has a different fix, so always read the finish reason before retrying.
Related
Tags: #Gemini #Debug #Troubleshooting