What smart bidding is, and the three ways it goes wrong

Smart bidding is Google Ads setting your bids automatically, using machine learning to chase whatever your conversion actions count as success. When it is fed honest conversion data it is genuinely good, and I leave it switched on in almost every account I manage.

What is smart bidding actually doing?

Instead of you choosing a fixed cost per click, Google adjusts the bid for every individual auction based on signals it can see and you cannot, such as device, location, time of day and search behaviour, to hit the goal you set, for example maximise conversions or a target cost per lead. The crucial detail sits in that phrase, your conversion actions. Smart bidding optimises toward whatever the conversion action counts, with no opinion about whether that event is a real enquiry or a tag misfiring. If you are unsure what yours actually counts, start with what a conversion action is, and why yours might be lying.

What happens when it is fed fake conversions?

It optimises toward junk, diligently. In one account I took over in 2026, the conversion tag fired on page load, so 75% of clicks were counted as conversions. Smart bidding duly hunted for people likely to load a page, which is everyone, and cheap low-intent traffic looked like brilliant performance. The reports climbed while the inbox stayed empty, a pattern I break down in Google Ads says you are getting conversions but no leads arrive.

What happens when it is starved of data?

It degrades. Smart bidding needs a steady flow of conversion data to keep its predictions sharp, and when a tag breaks it falls back toward guessing. The classic case is a website relaunch that silently removes the tracking code, which I cover in your website relaunch broke your leads. Nothing in the interface shouts about it. Campaigns keep running, bids drift, and performance sags for reasons that look mysterious until someone checks whether the tag still exists.

Why does it take so long to recover after a fix?

Because smart bidding re-learns slowly. Repairing the tag restores honest data from that day forward, but the model has been trained on months of bad signal and works its way back gradually, over weeks rather than days. Judging a repaired account in its first few days punishes the fix for the sins of the old data. Give it a proper learning period before drawing conclusions.

What should I do before blaming smart bidding?

Fix measurement before judging campaigns. Most accounts I rescue have a bidding system doing exactly what it was told with a broken definition of success, and verification matters because suspicion alone misleads: of 70 tracking problems I suspected across takeovers, 30 proved false once checked inside the GTM containers. Once the data is clean, hold whoever runs the account to real evidence, which is the subject of the three receipts and is my ads agency doing a good job.