Quick Answer: If your conversion tracking was down or degraded during a server-side tagging migration, Smart Bidding lost the signal it uses to decide when you're worth serving, and it pulled back delivery. Confirm it by checking whether your conversion volume fell 2 to 5 days before impressions did. If the timeline lines up, apply a data exclusion over the outage window, verify tracking parity, then leave the campaign alone for 2 to 3 weeks.
The scenario
You hired a developer to move you to server-side tagging. Somewhere in that build, enhanced conversions got switched off. Maybe for two days, maybe for a week. It's back on now and everything looks fine in the tag diagnostics.
Except impressions are less than half what they were. Search impression share fell off a cliff. Nothing about your budget, bids, or campaign settings changed.
Now you're stuck with a question nobody around you can answer with confidence. Did the tracking outage do this? Or is that just the most recent thing you touched, so it's getting the blame?
Both answers get argued hard in PPC circles. One camp says impressions respond to bids, budgets, and auction competition, not to tracking, so look elsewhere. The other camp says a conversion tracking gap is exactly the kind of thing that quietly strangles delivery. They're both partly right, and the difference between them is testable in about ten minutes.
Why a tracking outage can crush impressions
Smart Bidding doesn't treat conversion data as a report you read after the fact. It treats it as live input. Every conversion that comes back tells the system that a given auction, at a given time, for a given user, was worth entering.
When conversions stop being reported, the system doesn't conclude that tracking broke. It concludes that the traffic stopped working. So it does the rational thing and bids less aggressively into auctions it now believes are unprofitable. Fewer auctions entered means fewer impressions.
The lag is what makes this hard to spot. Conversion lag means the system doesn't panic on day one. It waits, expecting late conversions to fill in. When they don't, delivery pulls back. That's why the impression crash usually lands several days after the tracking damage, not the same day, and why people connect it to the wrong cause.
Worse, if conversion volume drops far enough, the campaign can re-enter its learning period. Now you're not just recovering from bad data. You're recovering from bad data that the bidding model already absorbed.
Step 1: Check whether the timeline supports it
Before you change anything, prove or kill the theory. Do not skip this. Applying fixes to a problem you haven't diagnosed is how a bad week turns into a bad quarter.
Pull a daily segmented report for the affected campaign covering the two weeks before the migration through today. You want three columns side by side.
Conversions. Impressions. Search impression share.
Now look at the order of events. If conversions fell first and impressions followed roughly 2 to 5 days later, the tracking outage is your prime suspect. That sequence is the fingerprint.
If impressions dropped on the same day as, or before, the conversion dip, tracking is not your problem. Go check what else moved. Budget caps, a target CPA or target ROAS edit, a bid strategy change, a new competitor entering your auctions, or a seasonal drop in query volume. Any of those will move impressions on their own, and none of them care about your tag setup.
Step 2: Apply a data exclusion over the outage window
Assuming the timeline confirmed it, this is the actual repair. Google gives you a tool built for exactly this situation, and most advertisers have never opened it.
Go to Tools, then Bid Strategies, then the Advanced Controls tab. Data exclusions live there. Set your date range around whatever stretch your conversions were unreliable.
One thing people get wrong here. Set the range wider than the outage itself. Include the day before your developer started, and include the tail end where tracking was partially firing but not fully restored. Half-working tracking pollutes the model the same way no tracking does. If you're unsure of the exact restoration moment, err on the side of a longer window.
What this does is tell the bidding model to ignore that stretch entirely when learning. It does not delete the data from your reports. It removes it from the decision-making.
Keep the exclusion narrow enough to stay honest. Excluding three weeks when the outage was four days will throw away good data your campaign needs.
Step 3: Verify tracking parity before you trust anything
Your tag diagnostics saying "active" is not the same as tracking working correctly. Plenty of sGTM setups fire cleanly and still lose conversions to deduplication problems, missing consent signals, or a server container that silently drops events under load.
Check the count. Pick a 7 day window since the fix. Compare the number of leads your CRM or form backend actually recorded against the number of conversions Google Ads logged for the same period. Account for attribution differences, but if Google is showing you significantly fewer conversions than real leads received, your migration is still leaking and no data exclusion will save you.
Fix the leak before you evaluate the recovery. Otherwise you'll spend three weeks waiting for a campaign to heal on data that's still wrong.
Most of this comes down to whether your conversion data is actually complete. If you're piping leads back into Google Ads through offline conversion imports or enhanced conversions for leads, the OCT Setup Checklist walks the whole setup end to end so a migration doesn't quietly break the thing your bidding depends on. It's free.
Step 4: Wait, and don't touch the targets
This is the step people fail. Two weeks is a normal recovery window. Three is not unusual, especially if the campaign re-entered learning.
The instinct during that stretch is to intervene. Raise the target CPA to force more delivery. Bump the budget. Switch to maximize clicks to "get traffic moving again." Every one of those restarts the learning process and adds another two weeks to your timeline.
Let it run. Watch impression share trend weekly, not daily. Daily numbers during a recovery are noise, and reacting to noise is what stretches a three week problem into a three month one.
FAQs
Will my impressions come all the way back?
Usually yes, assuming the underlying tracking is genuinely fixed and you applied the exclusion. Recovery is typically gradual over 2 to 3 weeks rather than a sudden return. If you're at week four with no movement, the tracking is probably still broken somewhere, or something other than the outage caused the drop.
Should I use a data exclusion or a seasonality adjustment?
Data exclusion. Seasonality adjustments are for anticipated short-term conversion rate changes, like a weekend sale. Data exclusions are for periods where the data itself is untrustworthy, which is exactly what a tracking outage produces.
Does this apply to Performance Max too?
Yes. PMax runs on the same conversion signal, and it's arguably more sensitive to gaps because it has more surfaces to allocate across. The data exclusion applies at the bid strategy level and covers it.
Can I avoid this on the next migration?
Set the data exclusion in advance, covering the planned migration window, before your developer starts. Run old and new tracking in parallel for a few days where possible so you can compare counts. Check conversion volume daily for two weeks after the switch instead of waiting for impressions to tell you something went wrong.