Automation Stopped Working But Shows No Error — The Silent Failure Problem
How do you tell a working automation from a broken one? Most people check the status column. That is exactly how you miss this.
A workflow can run green for four days straight and move nothing. Sarah found that out the slow way. Twenty-two green checkmarks in her n8n panel one Thursday morning, every execution successful, not one error logged. Then a client called. Called, not emailed. Their lead routing had been silent since Monday. She clicked into the latest run. Green nodes top to bottom, final output an empty array. The workflow had been doing its job perfectly and producing nothing. She had been reading the wrong column.
If yours shows green and the output never showed up, here is what is happening and how to find it.
Read the last node, not the status
Open the most recent execution and click the final node. Read its output.
Empty array ([]), a null, zero items: your workflow ran clean and did nothing. That empty output is not an error. It means data stopped moving somewhere upstream and nothing threw an exception to tell you. The status column confirms the workflow ran. Only the node output tells you whether it accomplished anything. Two different facts, two different places to look, and the alert only ever fires on the first one.
Five reasons green means nothing
The fastest of these resolves a large share of cases on its own, so try the destination check first.
The destination is empty even though the run is green. Execution succeeded. The CRM record, the email, the database row never appeared. Go look at the destination itself, not the execution log. Did the Airtable row land? Did the HubSpot contact get created? The standard instinct is to open the error logs, but a silent output failure leaves no trace there. It leaves a very visible trace in an empty database, which is the only place it shows. Why it happens: an external API can hand back a success-looking response for data it actually rejected, n8n sees the 200, and marks the run complete. Compare the two alerts. Useless: Execution completed: status=success node_count=7. Useful: Lead routing ran but 0 HubSpot contacts were created, check field validation on the write node.
A filter quietly stopped matching. The trigger fires, the early nodes show data, and then after an IF or filter node the item count is zero.
“A recent upgrade to n8n has broken several workflows in our system that were previously working well.”
— n8n Community Forum, 2024
Click through each node until you find where the count drops to zero. Almost always a filter that used to match and stopped, because a field got renamed upstream, a date format changed, or a comparison broke after a timezone shift. The filter still runs. It just matches nothing now, and a filter matching nothing is not an error.
A version bump changed the rules under you. A workflow that ran for months breaks right after an update, no error in the log.
“Since updating n8n yesterday, all of my existing workflows — and even new ones — using AI Agent nodes got affected.”
Check your n8n version, bottom left of the editor. Search the community forum for your node type plus the version number. Then open the affected node, re-pick the same options, and save. Re-selecting often re-hydrates the node against the new version without changing anything you can see.
An AI node changed its output shape. There is an AI node in the chain, downstream nodes show output, and the final record comes out empty or garbled. Open the AI node’s output and read the actual JSON. Language models drift: a field that was a plain string comes back nested, or a key gets quietly renamed. If a downstream node was reading {{ $json.name }} and the model now returns {{ $json.result.name }}, every run after that produces empty output and zero errors. Last week’s clean JSON is no guarantee of this week’s.
The app answers but the write never lands. This is the vibe-coded version of the same trap. The app URL loads fine, the health check passes, but a real sign-up or payment produces nothing in the database. A founder I work with hit this a month after shipping a Bolt.new app: the login screen loaded, new sign-ups returned a success response, and not one account was actually being created. Check Supabase or your database provider directly for recent rows. If the app says success and the database shows nothing new, the break is sitting between the response and the write, which are two separate operations that fail independently. More on that pattern in How to Know If Your Workflow Is Actually Running and My Automation Broke and I Don’t Know Why.
The check that actually catches it
This checklist gets you from green-but-empty back to green-and-working. It does not answer the harder thing: how do you catch the next one before four days pass and a client calls? The status column will say green every single time. Checking output by hand across 22 workflows every morning is not a plan that survives a busy week. NoCrash watches whether your workflows are producing real output, not just finishing, and tells you in plain language the moment the count goes to zero. Start free at nocrash.io and the next empty array gets caught the same hour, not the same week. Related reading: How to Prevent Silent Automation Failures and n8n Workflow Silently Failing. So set the output check today. Future-you, reading a calm morning message instead of taking that client call, will be glad you did.