How To Fix Google Ads Smart Bidding With A Primary vs. Secondary Conversion Framework

Many Google Ads accounts have a conversion tracking problem that is disguised as a strategy problem.

The ad account in this case has every action labeled as a “conversion.” The conversions are form fills, key button clicks, page views, cart adds, and checkouts started, which are all flowing into the same column, and all weighted equally. This less-than-ideal conversion setup is training Google’s Smart Bidding to optimize toward a vague composite of “engagement” instead of what matters for the advertiser.

When accounts are set up like this, the outcome is unfortunately predictable.

The campaigns look healthy inside Google Ads and on reports, the conversions look high, and the return on ad spend appears strong and efficient, yet none of it matches what the advertiser is actually experiencing inside the business. When the business team looks at their financial statements or the money in the bank, the story is different and doesn’t match. The advertiser isn’t growing, and doesn’t feel successful, and the internal reality doesn’t line up with what the Google Ads team claims.

The fix isn’t here isn’t to test another bid strategy.

The solution is to examine the conversion architecture. By using a primary‑versus‑secondary conversion framework, the ads manager can control what Google’s machine learning is allowed to learn from and, more importantly, what data the platform should ignore. When the conversion framework is applied correctly, primary and secondary conversions become an additional lever that actively shapes algorithm behavior and brings the account back into alignment with real business outcomes.

Here’s an example, and while the numbers are fictional, patterns like this show up in real accounts all the time.

A Performance Max campaign generated 4,000 clicks and produced 37 purchases, yet the ads platform reported a 62% conversion rate. That math only works when roughly 90% of the “conversions” are button clicks, form interactions, and abandoned checkouts being treated as equal to revenue. In practice, this looks like:

• Add‑to‑cart events counted as conversions, even when the user never returns.
• Checkout‑start events weighted the same as completed purchases, inflating return on ad spend.
• Button clicks or page scrolls logged as “micro‑wins,” overwhelming the real signals.

That is a signal‑to‑noise ratio of roughly 9:1 against the algorithm.

The Smart Bidding Signal Crisis

It’s important to reset how we think about Smart Bidding in Google Ads. Smart Bidding is not just a bidding tool; it’s a pattern‑matching engine. Google’s own documentation makes this clear when it explains that Smart Bidding evaluates audiences, the queries a user searched before and after, and a wide range of signals we can’t see. In other words, the bidding system isn’t optimizing for keywords in isolation; the bidding algorithm is optimizing for patterns in user behavior. And the patterns that are learned look at conversion architecture you feed it.

Every primary conversion you record teaches the algorithm what an “ideal customer” looks like. The model uses signals like device, time of day, audience cluster, query intent, landing page behavior, and more to find more users who match that pattern.

When the ad account is set up and mixes high-intent actions with low-intent micro-actions in the same primary pool, the model loses contrast. The bidding algorithm cannot distinguish a buyer’s pattern from a browser’s pattern, because the ad manager told it those two users represent the same outcome.

Many in the industry will say that Google Ads chases the easiest conversion. However, taking a step back, it is more than the system that does what it is designed to do. Yes, Google Ads takes the path of least resistance. This is because button clicks are vastly easier to generate than purchases. Cart adds are vastly easier than completed transactions. So the bidding algorithm aggressively hunts for users who do the easy things unless it is guided differently by the human in charge.

This is not a bug in Google Ads. It is the algorithm executing the instructions perfectly.

The Architectural Fix: Signal Engineering, Not Tag Management

The Primary vs. Secondary framework reframes conversion tracking from a reporting concern into an algorithmic training concern.

Two settings, two completely different jobs:

• Primary (Optimization): Populates the “Conversions” column. Actively used by Smart Bidding to train, predict, and bid. This is the algorithm’s curriculum.
• Secondary (Observation): Populates the “All conversions” column. Strictly ignored by the bidding strategy. This is the ad manager’s diagnostic layer.

The mistake many ad managers make is ignoring the secondary conversions. These are switches that determine which data the machine learning model is actually allowed to see during training.

Think of primary and secondary conversions as data architecture, not data management. When the account is set up, it’s important to consider what gets fed into the model and what gets stored in the warehouse for human review later. Those are two distinct surfaces with two distinct audiences.

A Representative Example Of How This Breaks Inside The Ad Platform

Here is another fictional scenario that can help illustrate how this failure shows up inside Google Ads accounts. Imagine a Performance Max campaign with healthy spend and what appears to be performing well. In this setup, “begin checkout” and “button click” are both designated as primary conversions alongside the actual purchase event. On the surface, the ad platform reports strong results. Underneath, the data tells a very different story:

• Reported conversion rate: 62%
• Composition: roughly 90% of “conversions” are button clicks and initiated checkouts.
• Actual purchases from 4,000 clicks: 37.
• True purchase rate: 0.9%
• Spend: $5,400.
• Revenue: $11,000.
• Effective ROAS: 2.04 (well below the 4.0+ target typical for the category).

In this scenario, the Smart Bidding system is not malfunctioning. It is performing exactly as instructed: finding more users who click buttons. The model has been trained on signals that do not correlate with revenue, so it optimizes toward the wrong pattern.

Correcting this issue is not instantaneous. Moving the micro‑conversions back to secondary status forces the system into a relearn phase because the model has been shaped almost entirely by false signals. Performance then becomes volatile and often depressed for several weeks while the algorithm rebuilds its understanding from cleaner data.

The broader lesson is that poor conversion architecture compounds quietly and recovers loudly. The cost of a flat, noisy setup is not paid in the first month; it is paid in the 30‑day relearn that follows the cleanup of the conversions.

The Technical Layer: Optimization Vs. Observation

Up to this point, the article has shown that the mechanics of a conversion framework matter because misconfiguration compounds over time. The next layer is understanding how Smart Bidding interprets the signals it receives.

How Primary Conversions Train The Algorithm

Every primary action recorded in the ad platform is treated as a successful outcome. Smart Bidding then works backward to identify the conditions that produced that outcome and increases bids to replicate those conditions. This is why the criteria for primary conversions must be strict. Only true macro goals belong in this category: a completed purchase, a submitted lead form, a booked consultation. These are actions that map directly to revenue rather than actions that merely correlate with revenue.

If a direct line cannot be drawn from the action to a dollar of pipeline, it does not belong in the primary pool.

How Secondary Conversions Inform Without Polluting

Secondary conversions operate in observation mode. The bidding system does not optimize toward them, but they still populate the “All conversions” column for reporting. This separation is the core value of the framework. It allows as many secondary actions as needed to map the funnel without contaminating the training data.

Examples include:

• Pricing page view.
• Add to cart.
• Begin checkout.
• Shipping page view.
• Account creation.

Each of these steps provides diagnostic insight into where users fall off. None of them instructs the algorithm to pursue low‑intent traffic. The result is a full picture of funnel behavior without sacrificing data quality.

There is one nuance worth noting. While Google’s documentation states that secondary actions are ignored for bidding, the system likely still uses them as predictive indicators of intent. This means even observation‑only events should represent meaningful steps in the buyer journey. Filling secondary slots with vanity actions risks creating false positives in the prediction layer.

Tracking legitimate funnel steps is the best advice.

The Hidden Override: Custom Goals

Custom goals, on the other hand, override the Primary vs. Secondary tagging entirely.

If you build a custom goal and add a secondary action to it, that action will be used for bidding in any campaign assigned that goal, regardless of how it is tagged at the account level.

This is a powerful feature, and a frequent landmine in Google Ads accounts. Strategists who assume “secondary is always observation” miss that custom goals re-promote those actions back into the bidding signal. A best practice is to audit every custom goal in the account before assuming the framework is intact.

How This Architecture Affects The Learning Phase

Smart Bidding’s learning phase typically runs 7 to 14 days after a strategy change (though this window extends significantly for campaigns with low conversion volume). During this window, the bidding algorithm is actively building (or rebuilding) its model of what success looks like.

A clean Primary vs. Secondary architecture compresses learning. Fewer, higher-quality signals mean faster convergence. The algorithm has clearer contrast between “buyer” and “non-buyer” patterns and can stabilize bid logic more quickly.

A polluted setup does the opposite for the account. The bidding algorithm grinds against contradictory signals, extending the learning phase and degrading early performance over the long-term. Worse, when the eventual cleanup happens or a restructure, the system enters a forced relearn and that 30-day window where revenue dips while the model unlearns the bad pattern.

There is also a default-state trap that catches even experienced ad managers. When you import conversions from Google Analytics into Google Ads, they are set to secondary by default. If your macro-goal lives in GA4 and you assume the import handled the optimization tag, you have just disconnected your bid strategy from your true revenue signal. A best practice is to verify the status manually after every import.

Edge Cases The PPC Manager Must Architect For

Phone Calls

Phone calls are the most context-dependent action in the framework.

For some businesses, calls are pure informational requests, with questions like “What time do you close?” These belong in secondary. For others, calls are the macro-goal because they result in booked consultations, demos, or sales conversations. These belong in the primary.

The decision is not based on the action label. It is based on the post-call data. If you cannot evaluate call quality, you cannot configure this correctly. Pull a sample of calls and talk to the humans answering the phones. Then categorize the calls and make an informed decision.

Imported Google Analytics Events

GA4 events imported into Google Ads default to secondary. This is intentional because Google does not want imported actions inadvertently changing your bid strategy.

But it means every macro-goal sourced from GA4 must be manually promoted to primary. This step is missed constantly, and the symptom when it is missed is subtle. A campaign that “should be” optimizing toward purchases is actually optimizing toward whatever else was already tagged primary in the account.

Low-Volume Accounts And The Cold Start Problem

For accounts that have not yet reached Smart Bidding’s data threshold (typically 30 to 50 conversions in a 30-day window, although this varies by strategy type), the framework in this post still applies, but the secondary layer becomes more strategically valuable for the account to perform well.

While Google does not officially support using secondary actions as bidding signals, many practitioners infer that the algorithm uses them as predictive indicators of intent, though Google has not confirmed this. For a low-volume account, that prediction layer can offer the algorithm enough texture to begin pattern-matching even before the macro-goal hits volume.

This is why secondary action quality matters more than secondary action quantity. Every meaningful step in the funnel, from a pricing page view to a demo video watch, or configurator interaction, is data that gives the algorithm a directional signal during the cold start.

Of course, garbage actions in this slot, however, create false positives the model may build a flawed early model around. So, it is important to apply discernment when adding secondary conversions.

When To Promote A Secondary Action To Primary

Almost never.

The framework in this article has limited tolerance for upgrading secondary actions to primary. Cart adds, checkout starts, and page views and chats should not be promoted, regardless of how much volume they generate. Volume does not equal intent. A high-volume “begin checkout” or “chat” action still represents users who did not buy.

The legitimate scenarios for status changes are narrow:

• Phone calls reclassified. When the business validates that calls are higher-intent than initially assumed (or vice versa).
• Imported GA4 events corrected. When an imported macro-goal lands in secondary by default and needs to be promoted to primary.
• Lead quality redefined. When the business shifts from “all leads” to “qualified leads only,” and the qualification action becomes the new macro-goal.

That is the entire list. If a strategist is regularly promoting micro-actions to primary, the framework is not being used, but it is being eroded.

Tactical Checklist: Auditing Your Primary Vs. Secondary Architecture

Before any Smart Bidding strategy test, walk the ad account through this audit:

• One macro-goal per campaign objective. Confirm a single primary conversion that maps directly to revenue.
• All micro-actions tagged secondary. Cart adds, button clicks, checkout starts, page views, and chats, none of these should be primary.
• GA4 imported events verified. Confirm any macro-goal imported from Analytics has been manually promoted to primary.
• Phone calls evaluated by quality. Sample recent calls and tag the action based on actual business value, not assumed business value.
• Custom goals audited. List every custom goal and confirm which secondary actions are being force-promoted into bidding.
• Funnel coverage in secondary. Every meaningful step between the click and the conversion has its own secondary action, which is not for bidding but for diagnostics.
• Reporting columns verified. “Conversions” column equals bidding signals. “All conversions” column equals full diagnostic layer.
• No vanity events in secondary. If an action is not a legitimate buyer-journey step, remove it. Quality over quantity is critical here.
• Learning phase respected. After any framework change, allow seven to 14 days for the algorithm to recalibrate before evaluating performance. Make sure to document the change in order to explain dips in conversion volume or efficiency.
• Relearn budget accounted for. If the account is being cleaned up from a flat setup, plan for a 30-day relearn period of depressed performance.

The Strategist’s Role In 2026

Smart Bidding will continue to absorb tasks that used to be human-controlled. This will happen from bid management to creative and audience targeting. The visible surface of paid search keeps getting smaller but it is important to remember to keep humans in the loop.

What does not absorb is signal architecture and how humans think through problems and outcomes. The algorithm cannot decide what data it should learn from, rather that is a business decision that needs a rational decision and not an optimization decision.

Doing this work requires understanding pipeline math, sales cycles, lead quality, and revenue attribution.

The Primary vs. Secondary framework is where that judgment lives at this point in paid search. If it is configured well, and the algorithm scales in the direction of the right outcomes. If it is configured poorly, and the algorithm scales the wrong outcomes faster and finds more pockets of “wrong.”

The framework is the strategy. The bid is just the output of the setup that has been outlined.

More Resources:

Featured Image: Master1305/Shutterstock