Your secondary conversion actions now change how Google bids

Every conversion action you demoted to secondary is now training data. Journey-aware bidding, currently in beta, lets Target CPA Search campaigns learn from non-biddable goals like calls, form fills and newsletter signups, not just the primary action you bid toward. The list you built to keep reporting tidy has quietly become an input to delivery.

This is the least discussed change in Google Ads this year and it has the highest chance of causing an unexplained performance shift in a well-run account.

What changed, mechanically

Before, the split was clean. Primary conversion actions trained the bidder. Secondary actions were recorded for reporting and did nothing else. That is why the standard advice was to mark anything soft as secondary and stop worrying about it.

Journey-aware bidding treats those soft actions as evidence of progress along a path. A user who called, or downloaded a spec sheet, or subscribed, is now understood as further along than a user who did neither, even though only the final sale carries a bid value. The model uses that signal to value the click.

Which means the quality of your secondary list now shows up in your CPA.

Why this bites good accounts hardest

Sloppy accounts have three conversion actions. Mature accounts have twenty-two, accumulated over years, half of them created by someone who has left, several of them measuring the same event twice through different tags.

The classic set of problems in a mature account:

• A thank-you page view and a form submit event both firing for the same enquiry.
• A GA4 imported goal and a native Google Ads tag counting the same action in parallel.
• Click-to-call counted as a conversion at every touch, so one person calling three times reads as three journeys.
• A newsletter popup signup, which mostly measures people trying to close a popup.
• A PDF download attached to a resource nobody buying anything ever reads.

Under the old model those were harmless clutter. Under journey-aware bidding they are instructions. Duplicate counting teaches the model that certain traffic produces double the progression it really does. Popup signups teach it that low-intent visitors are valuable.

The audit

Do this before you enable anything, because it is worth doing on its own merits.

1. 1.Export every conversion action with its category, counting method, click-through window and view-through window.
2. 2.Mark each one against a single question: does this event mean the business moved forward? Not is this interesting. Moved forward.
3. 3.Find the duplicates. Anything measuring the same user event through two tags gets deleted, not demoted. Demoting a duplicate no longer removes it from the system.
4. 4.Fix counting method. Lead-type actions should count one per click, not every. Calls and forms are the usual offenders.
5. 5.Check the windows. A 90-day click window on a soft action stretches the journey model over a period where the user has forgotten you exist.

Then, and only then, decide which secondaries you want the model to learn from.

Feed it a real outcome

The strongest version of this setup pairs a clean secondary list with offline conversion import. Push qualified lead and closed sale back into Google Ads from your CRM with the GCLID, and the model has an anchor at the end of the journey rather than a guess.

Without that anchor, journey-aware bidding optimises toward journey progression, which is a proxy. With it, progression is being weighed against something real. For any business with a sales team between the enquiry and the revenue, this is the difference between a bidder that finds you enquiries and one that finds you customers.

Two adjacent things that will break your inputs

First, attribution model choice is no longer a debate. First-click, linear, time-decay and position-based were removed from Google Ads and GA4 back in 2023. Last-click and data-driven are all that remain. Data-driven attribution in GA4 has volume requirements that are rarely quoted: at least 400 conversions for the specific action and 20,000 total conversions across all actions within the lookback window. Below that, you are on modelled output with thin support, and no amount of bidding sophistication fixes a starved measurement layer.

Second, check your consent configuration. From 15 June 2026, ad_storage became the sole control over whether GA4 shares cookies and identifiers with linked Google Ads accounts, and the Google Signals toggle was demoted to controlling only signed-in behavioural reporting. Two failure modes followed. Accounts with Signals off but ad_storage granted now collect more than they intended. Accounts that leaned on Signals being on while ad_storage defaults to denied lost Ads data silently, with no error and no warning.

The check takes ten minutes. Open DevTools, filter the network tab for google-analytics.com, and compare the gcs and gcd parameters on a consent reject versus a consent accept. Do not use GA4 Tag Diagnostics for this, it lags by two to three days and will tell you everything is fine while conversions quietly stop arriving.

The order to do this in

Clean the conversion actions. Verify consent signalling. Wire offline conversion import. Then enable journey-aware bidding on one campaign and compare against a control for a full sales cycle.

Doing it in the other order means you find out what your secondary list was worth by watching your CPA move for reasons nobody can explain.

Your conversion actions list is now a strategy document, so treat the next hour you spend in it accordingly.