In this article
Define what “stopped working” means in your account
Write down the first date the account changed and the first metric that changed. Did spend stop? Did impressions fall at the same spend? Did outbound traffic continue while recorded conversions disappeared? Or did the sales system show fewer customers despite normal traffic?
Use complete days in the account’s timezone. Comparing half of today with all of yesterday can manufacture a crisis. A seven-day view should also use comparable weekdays and enough time for the account’s normal conversion delay. Retain the raw counts alongside percentages: a decline from two purchases to zero has a different evidential weight from a sustained fall across thousands of orders.
Check outages without treating status as a diagnosis
Look at Meta’s official Ads Manager status page for the relevant incident period, and save any incident that overlaps your account’s change. A current green indicator cannot establish that nothing happened earlier. Equally, an incident affecting campaign creation does not by itself explain a fall in website sales from ads already delivering. [1]
If spend stopped, check billing, account restrictions, campaign schedules, delivery notices and review states in the native account. Keep account-specific screenshots and timestamps. A complaint thread can suggest an issue worth checking, but it cannot tell you how common the issue is or whether your account shares the same cause.
Separate the auction from the website
Build a simple chain: spend, impressions, outbound visits where available, measured website sessions, completed business actions and paid or qualified outcomes. Find where the ratio first breaks. Rising cost per impression suggests a different investigation from stable traffic with a collapsing checkout completion rate.
Do not expect the platforms to match perfectly. Attribution windows, consent, repeat visits and event definitions differ. Instead, ask whether each source changed relative to its own comparable baseline. If both backend purchases and landing-page conversion rate fell after a site deployment, investigate the site before rebuilding every ad set.
Audit changes before blaming an algorithm
List campaign edits, tracking releases, consent changes, offers, price changes, stock problems and new landing pages on one timeline. Then compare affected and unaffected segments. A decline limited to one country or one mobile destination is weaker evidence for a global delivery failure than a synchronized change across otherwise stable independent accounts.
Meta published an engineering update about GEM on October 13. That publication is evidence of its recommendation-system development, not proof that a particular advertiser’s decline was caused by a rollout on that date. [2] Do not turn a plausible explanation into a confirmed cause without account evidence.
A worked diagnosis: the same CPA, different problems
| Review item | Question or evidence | How to use it |
|---|---|---|
| Delivery loss | Spend or impressions stop | Inspect billing, restrictions, schedule and incident timing. |
| More expensive traffic | CPM or outbound visit cost rises | Compare markets, allocation and auction context. |
| Website conversion loss | Visits hold, completed actions fall | Reproduce the destination and inspect release timing. |
| Reporting loss | Backend outcomes hold, reported actions fall | Trace event definitions, collection and attribution. |
In a hypothetical campaign, $500 buys 1,000 visits and 20 completed actions: $25 per action. Later, the same spend buys 1,000 visits but only 10 completed actions, so cost rises to $50. That pattern directs attention toward conversion or measurement. If the later spend instead buys only 500 visits at the original 2% completion rate, there are also 10 actions and the same $50 cost—but the break happened before the visit. The account’s final CPA alone cannot distinguish these explanations.
Run one test that can distinguish explanations
If the page is suspected, reproduce the conversion journey on the device and browser that lost the most users. If tracking is suspected, reconcile a known completed action with the event trail and backend record. If creative fatigue is suspected, prepare a materially different concept while retaining a stable comparison.
Changing targeting, budget, bid strategy, creative and destination together makes the next result difficult to interpret. Document the hypothesis, intervention, unchanged conditions and review date. An urgent business problem can justify a fast action, but the report should still distinguish a protective action from a controlled experiment.
Ask for evidence rather than an instant verdict
In 1ClickReport, select the client and connect Meta plus the relevant analytics or revenue source. Ask the Analyst to break the decline into cost, traffic and conversion-rate changes. Request an explicit list of unavailable evidence, including billing notices, checkout logs or native status history that the connected data cannot supply.
The useful output is a ranked investigation with numbers and owners. “Meta is broken” provides no next action. “Mobile visits to one destination held steady while successful registrations fell after Tuesday’s release” gives the team a concrete place to look. A causal conclusion should wait for the supporting test.
A first task to try
Investigate why this client’s Meta results fell. Compare the latest complete seven days with the previous matching weekdays, and the preceding 28-day baseline. Separate spend, delivery, traffic, measured conversion rate and backend outcomes. State data limits and the next test for each plausible cause. Do not edit campaigns.Questions readers ask
Is there proof Meta ads stopped working for everyone?
This review does not establish a platform-wide failure. Check official incidents and your own account’s funnel before drawing that conclusion.
Should I immediately duplicate every campaign?
First identify the failing stage. Rebuilding campaigns can obscure the original cause and does not repair a broken website or event.
Sources and editorial method
Research checked October 26, 2026. Platform facts link to primary sources below. The worksheets and operating examples are our editorial recommendations; hypothetical examples are not customer results. This article is published by 1ClickReport.