Home·Playbooks·Conversion Signal Quality in Performance Max: An Audit Before You Touch Bids
UPLIFY Playbook · Google Ads / PMax

Conversion Signal Quality in Performance Max: An Audit Before You Touch Bids

10 min read· ~1916 words· published May 21, 2026

Performance Max optimizes toward the goals you configured, not the ones you had in mind. Here is how to read your conversion setup, find duplicates and low value events, and decide what to fix first.

Why might Performance Max optimize toward the wrong conversion, and how do you check for that before changing bids?

Performance Max optimizes toward the goals you configured, not the ones you had in mind. Here is how to read your conversion setup, find duplicates and low value events, and decide what to fix first.

Performance MaxSmart BiddingGoogle Adsконверсійні діїосновні конверсійні діїдругорядні конверсійні діїспеціальні цілі конверсійцілі за замовчуванням акаунтакоригування конверсійвиключення данихTag Assistantконверсионные действияосновные конверсионные действиявторостепенные конверсионные действияспециальные цели конверсийцели по умолчанию аккаунтакорректировки конверсийисключения данныхconversion actionsprimary conversion actionssecondary conversion actionscustom conversion goalsaccount-default conversion goalsconversion adjustmentsdata exclusions
Contents 9

Performance Max can quite easily optimize toward a mistakenly selected, duplicated or low value conversion, because its bidding follows whatever goals you configured in the account or in the campaign. It is a goal-based campaign type that optimizes toward the conversion goals you set, using Smart Bidding (Google Help), and Smart Bidding makes auction time decisions based on conversion data and contextual signals (Google Help). This is a risk, not an automatic outcome. But until the meaning of your events and the correctness of tracking have been verified, changing bids, budgets or creative can distract you from an error in the inputs themselves.

What a dirty conversion signal looks like in practice

A problematic signal does not always look like a broken tag. It can be an event that fires correctly on a technical level but means something other than what you have in mind, or means far too little.

Common varieties worth recognizing:

  • an intent event rather than an outcome: add to cart, phone number reveal, checkout start. On its own it is useful for observation; the problem starts when it takes part in bidding on equal footing with a purchase;
  • a single purchase counted twice because of parallel sources (for example, the Google Ads tag plus an import from analytics), which gives some orders more weight than others;
  • an event with no value, or with the same nominal value for every product, while real margins differ;
  • conversions that are later cancelled or refunded but stay in the data the system learns from;
  • an action a user can repeat many times in one session: a phone click, opening a messenger, resubmitting a form.

Google recommends using accurate conversion data and building campaigns with a clear goal (Google Help). Within our audit, accuracy means not only that the tag fires, but also what exactly you are counting and how many times.

How PMax reads your goal configuration

Google AI in Performance Max uses the advertiser's goals, creative assets, audience signals and, optionally, data feeds (Google Help). It is up to the business to decide whether a conversion action matches the result it actually wants. If the configuration includes an action in bidding, the system will use it for optimization.

Primary actions, secondary actions and the custom goal exception

Conversion actions are grouped into conversion goals. Primary actions appear in the "Conversions" column and are used for bidding when their standard goal is used for bidding. Secondary actions are for observation only and appear in "All conversions" (Google Help).

One important exception: a secondary action included in a custom goal is used both for reporting and for bidding (Google Help). In other words, the "secondary" label does not prevent an action from being used in bidding through a custom goal. A custom goal assembled at some point for a test campaign can bring an intent event back into bidding.

Account-default goals versus campaign-specific goals

Primary actions inside account-default conversion goals are used for bidding across all campaigns except those that have their own goals set at the campaign level. Goals that are not applied to a campaign do not appear in the "Conversions" column, but remain available in "All conversions" (Google Help).

Check two scenarios. In the first, a new primary action belongs to an account-default goal, so campaigns without their own settings factor it into bidding. In the second, PMax has campaign-level goals, so changes to the account-default goal set do not change its configuration. Before you interpret any numbers, you need to know which mode applies to the specific campaign.

Auditing from business event to tag and report

The sequence below is an audit method, not a Google requirement. Official sources support only the statements that carry a link.

  1. Write down the business events that genuinely matter to profit: a paid order, an order confirmed by phone, a qualified lead. Separately, write down the intermediate events you want to see but do not want to optimize toward.
  2. Open Goals > Conversions > Summary and review the "Status" column for each action (Google Help). Resolve actions with an unverified or problematic status before drawing conclusions about the campaign.
  3. For each action, record three attributes: primary or secondary, which goal it belongs to, and whether that goal takes part in bidding. This map lets you spot non-obvious use of actions through custom goals (Google Help).
  4. Check the level at which goals are applied: is PMax running on account-default goals or on its own campaign-level goals (Google Help)?
  5. Check the technical implementation: run a test conversion through Tag Assistant and look at tag implementation, whether it fired, and the status of the conversion action itself (Google Help).
  6. Look for parallel sources of the same event. A hypothetical example: a purchase arrives both from the global tag on the thank you page and from an analytics import, with both actions active and both primary.
  7. Reconcile the data with the store backend, but first account for differences in attribution models and reporting principles before calling a discrepancy a tracking failure (Google Help).
  8. Check value: is it passed dynamically, is it the same for every product, does it match the store's currency and tax logic?

The output of this pass is best consolidated into a conversion registry. For each action, record the technical name, a plain description of the business event, the data source, the owner of the setup, its role in bidding and the way to verify it. Use a separate field to note what happens after an order is cancelled or returned. This is a Uplify editorial template, not a Google Ads feature. It exists so that the developer, the analyst and the PPC specialist describe the same event in the same terms. If an action's name cannot be explained without digging through an old chat or relying on the previous contractor's memory, the configuration is not yet ready for controlled changes.

Quick diagnostics

What you see What to check first Where to look
Many conversions, few orders in the backend Whether an intent event entered bidding through a custom goal Goal settings and action type (Google Ads Help)
The campaign does not react to changes in account goals Whether goals are set at the campaign level Campaign goal settings (Google Ads Help)
Conversions show in "All conversions" but not in "Conversions" Whether the goal is applied to this campaign Report columns (Google Ads Help)
An action has an unexpected status Tag implementation and firing, test conversion Goals > Conversions > Summary (Google Ads Help)
Ads and backend numbers do not match Attribution and reporting differences, before concluding something is broken Comparison of methodologies (Google Ads Help)

Duplicates and accidental events

Duplication can be selective: it may fire on part of the traffic, on a particular payment method, or when a user returns to the thank you page from browser history. Because of that, the total conversion count will not show you which segments picked up extra weight. The impact on the campaign should not be declared without checking the sources and the distribution of duplicates.

While checking, count the actions that can physically repeat within a single visit, and review every source that records the same event. If you import offline data or analytics data in parallel with the tag, decide which source is the one used for bidding and which stays for observation. Google Ads provides the split into primary and secondary actions for exactly this (Google Help).

Value, returns and cancellations

For a retail business, value based optimization only makes sense when the value reflects something real. Smart Bidding can optimize specifically toward conversion value (Google Help), but what that value means is up to you.

Checking value does not end at the moment of the initial hit. Orders get cancelled, goods get returned, and amounts get revised after a conversation with the customer. The data should reflect that stage of the order lifecycle too. Conversion adjustments exist for these cases: they let you retract a conversion that should no longer count (a cancelled order or a return) and restate a changed value (Google Help). Order or transaction identifiers matter for matching an adjustment to the original conversion (Google Help), so an end to end identifier is worth building in at the implementation stage rather than bolting on later.

Data exclusions or conversion adjustments

These tools address different types of problems.

Data exclusions help reduce the impact of conversion tracking issues on Smart Bidding. They affect the data used by bidding, but they do not remove conversions from reporting (Google Help). This is a tool for outages and interruptions in data collection, not for routine cleanup, and it does not promise consistent performance (Google Help).

Conversion adjustments, by contrast, work with individual events that were recorded correctly but later changed status or amount (Google Help). Cancellations and returns are their territory.

A workable dividing line for an audit: when the problem concerns data collection itself during a specific period, check whether a data exclusion fits. When a conversion that was already recorded later changed status or value, check whether an adjustment fits.

What to change first and how to verify it

The order below reflects our editorial approach to prioritization, not an official recommendation.

  1. Fix the semantics first: remove from bidding any events that are not a business outcome, including through a review of custom goals (Google Help).
  2. Next, eliminate duplication, so that only one source per event takes part in bidding.
  3. Then sort out value and set up the passing of order identifiers for future adjustments (Google Help).
  4. After that, settle the goal level: account or campaign, deliberately rather than by default (Google Help).
  5. Once those checks are done, go back to bid strategies, budgets, asset groups and audience signals.

Verification after changes should be as unglamorous as the audit itself: a test conversion through Tag Assistant, a status check for the action in the conversions summary (Google Help), and reconciliation with the backend adjusted for attribution. Change one block at a time, otherwise you will not know what actually affected the campaign's behavior.

Checklist before changing bids and assets

  • Every conversion action has an owner and a clear business meaning, written out in words.
  • For each action, you know whether it takes part in bidding, custom goals included (Google Ads Help).
  • You know whether the campaign uses account goals or its own (Google Ads Help).
  • Action statuses have been checked in Goals > Conversions > Summary (Google Ads Help).
  • A test conversion goes through and is recorded correctly (Google Ads Help).
  • There are no two sources of the same event taking part in bidding at once.
  • Value is passed dynamically and matches the real economics of the order.
  • The order identifier is passed, so adjustments are possible (Google Ads Help).
  • For periods with technical failures, data exclusions have been considered instead of manual report cleanup (Google Ads Help).
  • The date and content of every change have been logged, so later trends can be interpreted.

Performance Max uses the goals you configured along with the other signals available to it (Google Help). That makes a conversion audit part of optimization. Repeat it after adding a new event, data source or integration, and whenever reporting stops lining up with actual orders.

Request audit