ChatGPT is down. So are Claude, Gemini, and Grok, all at the same time. If your whole morning routine just hit a wall, you’re not alone.
What actually happened
On September 3, 2026, a wave of outages took down most of the big AI platforms at once. PCMag reported that Downdetector recorded more than 36,000 reports at the peak, which tells you how many people were suddenly staring at error screens.
The timeline is revealing. OpenAI’s problems started around 10:30 a.m. ET, and the company says it applied a fix and is now “monitoring the recovery.” Claude had trouble earlier in the day, with issues starting around 8:30 a.m. ET, first hitting Sonnet 5 and then spreading to other models like Fable 5.1 and Opus 5. Gemini saw its own spike around the same time, with prompts returning errors for some users. Grok went down too. Even Cursor, the coding assistant, caught a piece of it.
Here’s what makes this one weird: it wasn’t a Cloudflare outage. The company explicitly said its services were running normally. So all these AI platforms went down together, without a shared internet plumbing failure to blame. As of this writing, nobody involved has said what caused it. OpenAI’s status page mentioned 24 minutes of downtime across ChatGPT and Codex before the fix rolled out, and Anthropic reported recoveries across its Fable models while Opus 5 and Opus 4.8 were still affected. If you’re wondering whether the outage is over, the short answer is: mostly, for most people, but not everywhere yet.
Why does this keep happening?
Because we built an entire era of work on a handful of chatbots.
Think about your actual day. You draft emails with one model, summarize meetings with another, maybe ask a third to debug a spreadsheet formula. When one of them hiccups, you feel it. When several of them hiccup at once, your whole system turns into a pumpkin.
There’s also a simpler factor: AI services are fragile in ways that regular apps aren’t. A chatbot request is actually a chain of moving parts, model serving, safety checks, memory, plugins, billing. One bad configuration can take out the whole stack. Companies are pushing updates fast right now, all of them, and fast release cycles mean more chances for something to break.
Notice what wasn’t affected, though. Your email still worked. Your video calls still worked. Only the AI layer crumbled. That’s a useful mental model: AI is a layer on top of the internet, not the internet itself. The internet was fine today. The layer wobbled.
What to do when ChatGPT is down
This is the part you actually came for, so let’s make it practical.
Check the status pages first
Before you start rage-clicking, take thirty seconds to confirm it’s not just you. Check OpenAI’s status page, Anthropic’s status page, and Downdetector, which covers everything at once:
- status.openai.com
- status.anthropic.com
- downdetector.com
Downdetector is the fastest read because it shows spikes in real time. If thousands of people are reporting the same issue, it’s not your Wi-Fi, and you can stop restarting your router. A quick search for “is ChatGPT down” also works, but status pages and Downdetector beat random tweets.
One more tip: descriptions matter. If the status page says “monitoring,” the fix is in and they’re watching it. That usually means the service is coming back, slowly. Plan around that instead of refreshing in a panic loop.
Switch to another AI tool
Here’s the trick people miss: the alternatives don’t all go down together. At least, not usually. That’s why a multi-model setup is worth having, even if you’re not a power user.
OpenRouter is a great place to start because one account gives you access to dozens of models from one page. If ChatGPT is down, the next model up might be totally fine. We have a full OpenRouter tutorial for beginners if you want the setup walkthrough.
The free tiers matter here too. If Claude is down but Gemini is up, you can usually get decent work done in the free version of whichever one is standing. It won’t be your usual setup, but it beats staring at an error message.
Keep a running list of what each tool is good at. You don’t want to discover, mid-outage, that you don’t remember how to open anything else. Write it on a sticky note. Future you will be grateful.
Go offline with local AI
When all the big clouds fail, local AI still runs. Models that live on your own computer don’t depend on anyone’s servers, which is exactly what you want during an outage.
Perplexity Hybrid Compute is a good example of this idea: it runs private AI tasks on your own Mac so your work survives cloud hiccups. And if you’ve never tried local AI, even the simplest backup counts. A notebook and a pen, if it comes to that. Draft in a plain document. Thinking doesn’t actually need a model.
Keep a backup workflow
The best time to build an outage plan is when nothing is broken. Keep a folder with your most-used prompts, the ones that took you an hour to get right. Save your custom instructions somewhere you can read them. When ChatGPT is down, the prompts don’t disappear, and neither does the work you did with them. We covered how to train ChatGPT to write like you, including saving those instructions, if you want a head start.
What if it lasts all day?
Most outages are over within hours. Some aren’t. Build for the second case.
For anything urgent, switch to email and plain documents. They’re boring, they never go down, and they’re the universal fallback. For anything creative, use the downtime to outline, sketch, or collect raw material that a model can polish later. For anything repetitive, batch it: write every prompt you’d normally type, in order, and run them when service returns.
The move that saves the most time is deciding, before the outage, which tasks genuinely need AI and which ones only feel like they do. You’ll be surprised how many things survive on a notepad.
What you should NOT do
A few instincts are worth fighting during an outage.
Don’t retry every thirty seconds. It generates noise on the status pages and stresses nobody but you. Check once every ten minutes instead.
Don’t panic-buy anything. Scammers love outage days. If you see ads promising “instant access” or “priority connection” during a crash, that’s a grift, not a solution.
Don’t redo work you already have. If a draft is stuck in a cloud platform, it will come back when the service does. Recreating it from scratch is how you end up with two versions and extra stress.
The takeaway
Outages happen, and they’ll keep happening as long as AI runs on centralized servers. Your job isn’t to prevent them, it’s to have a plan for the twenty minutes, or the two hours, that they last. Check the status page, switch tools, keep your prompts saved, and don’t freak out.
The worst part of an outage is usually the panic, not the downtime. ChatGPT is down today, maybe, but it’ll be back tomorrow. You’ll be fine either way, as long as you’ve got the fallback habit.
