Google finished retiring first click, linear, time decay and position-based attribution in September 2026. Every conversion action still running on one of those four models was force-migrated to data-driven attribution, with no opt-out and no notification worth the name. If your conversion counts moved this month and your tracking checks out clean, this is almost certainly why — and the real damage is not the reported number, it is that your Smart Bidding targets were calibrated against a model that no longer exists. Two models remain: data-driven and last click. Here is what to actually do about it.
What changed, precisely
Google Ads now supports exactly two attribution models. Data-driven is the default for every new conversion action. Last click is still selectable. The four rule-based models in between are gone — first click, linear, time decay and position-based stopped being available for new conversion actions in mid-July 2026, and the remaining ones running on them were migrated over the following weeks. The Google Ads help page on attribution models now reads as if the other four never existed.
Google also removed the old eligibility threshold for data-driven attribution. It used to take roughly 300 conversions and 3,000 ad interactions in a 30-day window to qualify. That requirement is gone, which is how a 40-conversions-a-month local services account ended up on a machine-learned attribution model this quarter.
That is worth sitting with for a second. Eligibility and accuracy are not the same thing. A model with no minimum still needs volume to learn anything, and Google's own practical guidance lands somewhere around 200 to 300 conversions a month before the weighting is stable. Below that, data-driven attribution tends to converge on something that looks a lot like last click with extra month-to-month wobble. You get the new model's label and last click's behaviour, plus noise.
Your conversions did not drop. They moved
The first call we get after a migration is some version of "we lost conversions." Almost always, the account total is within a couple of percent of where it was. What changed is the split.
Under last click or position-based, a branded search campaign that closes the deal collects most of the credit. Under data-driven, some of that credit gets handed back up the funnel to the generic prospecting campaign that introduced the customer in the first place. The account total is conserved. The campaign-level numbers are not, and campaign-level numbers are what your reports, your targets and your budget decisions are built on.
The tell is simple: if account-level conversions are flat but campaign-level conversions moved in both directions, you are looking at a credit redistribution, not a tracking failure. If the account total genuinely fell, stop reading this and go check your tags — that is a different problem, and our Google Ads and GA4 mismatch guide covers the usual suspects.
One more wrinkle: fractional conversions. Data-driven attribution assigns partial credit, so campaign rows now show values like 12.4 conversions. Any client-facing report or internal script that assumed integers will either round oddly or break. We have seen a Looker Studio scorecard silently floor every campaign to the integer below and shave four percent off the total on the page.
The expensive part: your bid targets are now miscalibrated
This is the bit that costs money, and it is the bit almost nobody does.
Smart Bidding optimises toward the conversions in your Conversions column. Change how credit lands in that column and every tCPA and tROAS target you set becomes a target for a measurement that no longer exists. Google is unusually direct about this in its best practices for attribution model changes: update your bids and targets, or the change in attribution will cause over- or under-bidding.
Work it per campaign, not per account:
- Cost per conversion fell — the campaign is picking up credit it did not get before. Leaving the old, higher target in place tells Smart Bidding it has room it does not need, and you overpay for the same volume.
- Cost per conversion rose — the campaign gave credit away. The old target now looks unachievable, the strategy throttles delivery, and a campaign that was working quietly stops spending.
- Barely moved — leave it alone. Not every campaign needs adjusting, and changing all of them at once destroys your ability to attribute the outcome.
The arithmetic is the boring kind: calculate the percentage change in cost per conversion between the old model and the current one, then move the target by that same percentage. Exclude the most recent 14 days from the comparison, because conversion lag makes recent data look worse than it is and you will talk yourself into a target cut you do not need.
Then stop touching it. Every target change restarts recalibration, and an account where someone is nudging targets weekly never produces a clean read. Two weeks minimum, longer if your average days-to-conversion is longer. We wrote about the same discipline in the context of Smart Bidding's exploration behaviour — the failure mode is identical, and it is always impatience.
The column most people never turn on
Here is the practical unlock, and it takes about two minutes.
Google Ads ships Conversions (current model) and Cost / conv. (current model) columns. They restate historical performance as if your current attribution model had been in place the whole time. Standard columns show what was reported at the time; current-model columns show what it would have been under today's model.
That distinction is the difference between a defensible client report and an argument. If your trend chart mixes pre-migration months reported under position-based with post-migration months reported under data-driven, you have a chart with a fake step change in it, and you will spend the first ten minutes of the call explaining an artifact. Rebuild the trend from the current-model columns and every period speaks the same language.
Practically: add both columns to the campaign view, pull 90 days ending 14 days ago, and you have your rebaselining dataset and your corrected client chart from the same export. This is also the moment to sanity-check anything downstream — dashboards, alerting thresholds, the sheet where someone hand-keys last month's CPA. Getting a dashboard clients actually read means it survives a platform change like this one without quietly lying.
Google Ads and GA4 will still disagree, and that is fine
Both platforms now default to data-driven attribution, which sounds like it should make the numbers converge. It does not, because they are answering different questions.
Google Ads only distributes credit across Google ad interactions and books the conversion on the date of the click. GA4 distributes credit across every channel it can see — organic, direct, email, referral — and books it on the date of the conversion. Feed both the same user doing the same thing and you get two legitimately different numbers. The gap usually widens for longer sales cycles, because that is where the date convention and the non-Google touchpoints both have more room to matter.
The answer is not reconciliation. It is designation: pick the platform that is the number of record for a given decision, say so out loud in the report, and treat the other as a cross-check. For clients where the spread is genuinely material — high ticket, long consideration, several paid channels running at once — this is the point where a dedicated attribution layer stops being a nice-to-have. That is the whole argument in our Hyros attribution guide for agencies: when the ad platforms each grade their own homework, you need one dataset that does not.
When last click is still the right answer
Data-driven is the default, not the law. Switch a conversion action to last click when:
- The funnel is genuinely single-touch. Emergency plumbing, locksmiths, same-day service. There is no journey to model and fractional credit adds nothing but decimals.
- Volume is too low to learn from. A few dozen conversions a month gives the model nothing to work with, and a simple, explainable model beats an unstable one.
- A third-party system needs to reconcile. If your CRM or an attribution platform is the number of record, matching its logic beats being more theoretically correct and permanently unexplainable.
Switch it deliberately, note the date, and rebaseline targets afterwards — the exact same way. A model change is a model change, whichever direction you go, and the second one hurts more if you skipped the rebaseline after the first.
The 30-minute version
If you manage more than a couple of accounts, do this today rather than discovering it in a quarterly review:
- Add the Attribution model column in Goals, then Conversions, then Summary. Flag every action that migrated and note the date from change history.
- Add Conversions (current model) and Cost / conv. (current model) to the campaign view.
- Pull 90 days ending 14 days ago. Compare cost per conversion, standard versus current model, per campaign.
- Move tCPA and tROAS targets by the per-campaign percentage change. Skip campaigns that barely moved.
- Freeze targets for two weeks, or one average conversion cycle, whichever is longer.
- Rebuild the client trend chart from current-model columns and annotate the migration date.
Half an hour per account, and the alternative is a quarter of Smart Bidding optimising toward a target you set for a model Google deleted.
Get a second set of eyes on it
Attribution changes are cheap to fix and expensive to ignore, and they almost never announce themselves — the account just gets slightly worse in a way that looks like seasonality.
If you would rather have someone check the work, get a free automation audit. We will look at your conversion actions, your bid targets and the reporting layer sitting on top of them, and tell you exactly which campaigns are now mispriced. If the fix is bigger than a target adjustment, our paid ads management team handles the rebaseline end to end — but most accounts just need the thirty minutes above, and we will tell you if yours is one of them.