Every month, someone shows me a Google Ads account and asks why their "best" campaign got its budget cut. Nine times out of ten, the answer isn't the campaign. It's the attribution model quietly deciding which campaign gets credit — and most advertisers have never looked at the setting.
Attribution is the rulebook Google uses to answer one question: when a conversion happens, which click gets the credit? Change the rulebook and the same account tells a completely different story.
Why Google killed most attribution models
In 2023, Google retired first-click, linear, time-decay and position-based attribution. Today you have exactly two choices:
- Last click — 100% of the credit goes to the final ad click before conversion.
- Data-driven attribution (DDA) — machine learning distributes credit across clicks based on how much each one actually contributed, modeled from your account's own conversion patterns.
Google removed the middle options because almost nobody used them correctly, and because DDA needs to be the default for Smart Bidding to do its job. That part is true. What the documentation doesn't emphasize: DDA is only as honest as the data feeding it.
How data-driven attribution actually works
DDA compares the conversion paths of people who converted against similar paths of people who didn't. If users who saw your generic search ad before clicking your brand ad convert 30% more often than users who only clicked the brand ad, DDA shifts a slice of credit to the generic campaign.
The practical effect in most accounts: brand campaigns lose credit, generic and upper-funnel campaigns gain it. That's usually more truthful — your brand campaign was always going to catch people who already decided.
When DDA lies to you
DDA can only model what it can see, and it only sees Google clicks. Three failure modes to watch:
- Thin data. Under roughly 300 conversions a month, DDA's model is guessing with confidence. Small accounts get noisy credit shuffles that change week to week.
- Missing channels. If Meta, email or organic drive part of the journey, DDA still splits 100% of the credit among Google clicks. Google grades its own homework and never gives itself less than full marks.
- Broken tracking. If your conversion tag double-fires, fires on the wrong page, or misses 30% of conversions because of ad blockers, DDA optimizes toward corrupted data. Fix tracking before you debate attribution.
Attribution models don't create truth. They redistribute whatever truth — or garbage — your tracking collects.
Which model to use, in practice
- Use DDA if you have healthy volume (300+ conversions/month), clean tracking and Smart Bidding. This is the right answer for most accounts.
- Use last click when you need a stable measuring stick: auditing an account, reconciling with finance, or diagnosing a sudden performance change. Last click is dumb but consistent — sometimes you want the boring witness, not the clever one.
- Compare both in the Attribution report (Tools → Measurement → Attribution). The delta between models tells you where assisted conversions live. Big gaps on generic campaigns = your funnel is longer than your reporting assumes.
When to switch
Switch to DDA when your volume can support it and your tags are verified. Expect Smart Bidding to recalibrate for one to two weeks — don't panic-adjust targets during that window. Switch back to last click temporarily only for diagnosis, never as a permanent retreat because DDA "made numbers look worse." It usually made them look more honest.
And whichever model you choose, keep one rule: judge campaigns by the same model over time. Comparing last-click January against DDA March is how budgets get cut for the wrong reasons — and how your best prospecting campaign ends up dead.