The confusion starts because Google made two things look similar. Every conversion action has a “Conversion Category” dropdown (Purchase / Lead / Sign-up / Page view / etc.), and a separate “Goal” assignment with a primary/secondary flag. Both look like they steer Smart Bidding. Only one actually does.
The Conversion Category is a taxonomy field. It groups actions in reports and helps Google understand your funnel shape. It has zero direct effect on which conversions Smart Bidding will bid toward. The primary/secondary flag on the goal assignment is what tells the algorithm this is the outcome I want more of.
What the Two Fields Actually Do
| Field | What it controls | Affects Smart Bidding? |
|---|---|---|
| Conversion Category | Reporting group; some GA4/Insights features; category-based Goal grouping | Indirect only — via Goal grouping |
| Primary for a Goal | Smart Bidding optimizes to maximize count/value of this action | Yes — this is the actual lever |
| Secondary for a Goal | Recorded and visible in Conversions column but not optimized toward | No direct bidding effect; used as observation signal |
| Include in “Conversions” column | Counts in the top-line Conversions metric | Yes — if a bidding strategy targets Conversions, this decides what counts |
The critical row is the last one. Even a secondary action, if “Include in Conversions column” is on, still counts toward the top-line Conversions metric that Maximize Conversions or Target CPA bidding uses. So the practical bidding decision is a combination of: is this action primary for the campaign's goal, and is it in the Conversions column?
The Setup We Find in 80% of Audits
The pattern is nearly universal. An account has 6–12 conversion actions. Someone at some point marked most of them as primary for the same default goal, thinking “more signal = better learning.” The actions include: purchase, add-to-cart, begin-checkout, form-submit, phone-click, page-view-of-pricing, PDF-download, and a couple of legacy events nobody remembers creating.
Smart Bidding sees eight primary actions and tries to maximize their sum. It bids toward whichever one is cheapest to generate — usually the least valuable one. Add-to-cart events explode. Purchase volume flatlines. The client asks why CPA dropped but revenue didn't move.
The Correct Pattern for Different Business Types
| Business | Primary | Secondary (observation) | Excluded from Conv. column |
|---|---|---|---|
| E-commerce, one product tier | Purchase (with value) | Begin-checkout, add-to-cart | PDP view, PLP view |
| E-commerce, subscription | First-charge purchase (with LTV proxy value) | Trial start, add-to-cart | Page views, video plays |
| Lead-gen, low-friction service | Qualified form + phone call | Any form, phone click without call | Newsletter, PDF download |
| Lead-gen, high-ticket B2B | Demo requested (with lead-value assigned) | Contact form, calendar view | Content downloads, blog views |
| Local service | Phone call (verified length > 60s) | Direction click, short calls | Menu views, gallery clicks |
The through-line: one primary that actually represents money-in. Two or three secondaries that are strong intent signals. Everything else excluded from the Conversions column entirely — kept for reporting but not counted anywhere Smart Bidding can see.
The Learning Phase Trap When You Fix This
You find the mess. You know the correct setup. You start toggling actions from primary to secondary. And then the account resets its learning phase, CPA doubles for a week, the client panics.
Not every conversion action change triggers a learning reset — but three specific ones do:
- Removing an action from the Conversions column (Include in Conversions off). This directly changes the top-line Conversions metric that bidding optimizes against.
- Changing the primary goal itself on the campaign — switching from “Purchases” goal to “Leads” goal at the campaign level.
- Adding a brand-new primary action with a different value model — for instance, promoting a previously secondary action that has 10× the volume.
What does not trigger a full learning reset:
- Demoting a primary to secondary while leaving it in the Conversions column (bidding still counts it).
- Adjusting Include in Conversions on an action that's already secondary and receives <5% of total volume.
- Adding a new conversion action as secondary from day one.
The Value-Per-Conversion Fix Nobody Notices
If your primary is a form-submit or lead action, Smart Bidding will treat every lead as equal — worth the same, whether it's a $200 order or a $20,000 contract. On any lead-gen account above $1K/mo spend, this is where most performance is left on the table.
The fix is to assign a value to the conversion at fire time. In lead-gen contexts, this is a proxy — expected value = close-rate × average deal size. In e-commerce it's literal — the order total. Once values are flowing, switch bidding from Maximize Conversions to Maximize Conversion Value or Target ROAS. This is where Smart Bidding starts to differentiate between a $200 lead and a $20K lead.
Every conversion event should carry value andcurrency. For leads, use expected-value math. For e-commerce, pass the actual order total. Don't use fixed per-action defaults — those give bidding nothing to differentiate.
Move to Maximize Conversion Value only when you have at least 30 conversions per month with real values (not fixed defaults) over 30 days. Below that, the algorithm can't differentiate reliably and behaves worse than Maximize Conversions.
One $100K order will teach the algorithm to chase whatever the campaign was serving that day. Cap conversion values at, say, 3× your median. Log the true value elsewhere for internal reporting — but feed the algorithm a bounded number.
Audit Checklist
The Reframe That Fixes Most Accounts
Every conversion action in your account is an instruction to Smart Bidding. If you have eight primary actions, you're giving eight instructions of equal weight — and the algorithm will optimize for the cheapest one to satisfy. The reframe is to treat primary/secondary as a hierarchy of intent, not a menu of things you'd like more of. One primary that represents money. Two or three secondaries that predict money. Everything else out of the Conversions column.
This one change moves more Smart Bidding performance than most creative refreshes, keyword rebuilds, or PMax launches. It's also the cheapest — no new spend, no new creative, no new campaigns. Just an honest answer to “what outcome do we actually want more of.” More Smart Bidding patterns in the Sterling Lab blog — or see how we run Google Ads accounts for a walkthrough of the audit pass this checklist comes from.
