How Performance Max Really Works: What Google Doesn't Tell You

In 2021 Google launched Performance Max, marketing it as the campaign of the future: a single campaign type capable of appearing on every Google channel, Search, Display, YouTube, Gmail, Maps, Discover, automatically, optimising for conversions.

The promise was simple: give Google your assets, set a goal, and let the algorithm do the rest.

The reality, for anyone managing this campaign type on real budgets, is far more nuanced. There are mechanisms that official documentation describes vaguely, algorithm behaviours that only emerge from direct experience, and optimisation decisions that remain opaque even to the most advanced advertisers.

This article analyses how Performance Max really works: starting from official Google sources, but clearly flagging where inference and hands-on experience tell a different or more complete story.

How Performance Max Works: The Official Version

According to the official Google documentation, Performance Max Google Ads is a goal-based campaign that uses automation to find new customers and increase conversions across all Google channels. The declared operation unfolds in four steps:

1. The advertiser provides the assets, headlines, descriptions, images, videos, logos, organised into asset groups.
2. The system dynamically combines assets to build ads suited to every placement and format across all channels.
3. Machine learning optimises bids in real time to maximise conversions (or conversion value) based on the set goal.
4. Audience signals provided by the advertiser guide the algorithm in the initial learning phase.

On paper it is an elegant system. But the documentation stays on the surface: it does not explain the internal decision-making process, does not describe how budget is allocated across channels, and does not clarify the hierarchy between PMax and other campaigns active in the same account.

Performance Max Asset Groups: Structure, Logic, and What Google Doesn't Say

According to the official documentation on how asset groups work, Performance Max asset groups are collections of related assets grouped by theme or audience, used to create personalised ads. Google allows up to 100 asset groups per campaign and recommends creating them separately by theme.

What Google does not clarify is how the algorithm chooses which asset group to activate for each individual query or placement. Here are the most critical points:

• Asset selection: Google states it combines assets to create the most relevant ad for each user. In practice the combinations are not visible; they can only be tested via A/B on separate asset groups.
• Thematic segmentation: Google recommends separate asset groups by theme, confirmed by experience, but the algorithm can decide to ignore the segmentation and use a stronger asset group for queries unrelated to that theme.
• Video impact: Google declares video optional and auto-generates videos if none are provided. In practice, auto-generated videos are often poor quality and can damage the brand. Always providing your own video is an established best practice.
• Final URL expansion: Google can expand the final URL to other pages on the site if it considers them more relevant. Expansion is on by default and must be explicitly disabled for full control over ad destination.

Performance Max Audience Signals: Suggestions, Not Constraints

The official Google documentation on audience signals defines Performance Max audience signals as a starting point for optimisation and specifies that the algorithm can exit the indicated segments to find customers with higher conversions. This is one of the areas where the gap between official communication and operational reality is most significant.

Many advertisers insert audience signals convinced they are targeting a specific audience as they would with a standard Display campaign. That is not how it works: PMax uses those signals to learn faster, not to limit itself to those segments.

The most effective audience signals, from practical experience, are:

• Recent conversions (Customer Match): providing lists of existing customers or buyers from the last 90 days significantly accelerates the learning phase.
• High-intent site visitors: users who viewed the product page or cart without converting are a far more precise signal than generic interest categories.
• Custom audiences based on intent keywords: more specific than Google's generic in-market audiences, which rarely add value beyond what the algorithm would discover independently.

What Google Doesn't Tell You About Performance Max

Here are the points that official documentation addresses vaguely or does not tackle directly, essential for anyone managing PMax campaigns on real accounts.

1. Does PMax cannibalise Search campaigns?

Yes, in some cases. The Google documentation on campaign interaction states that PMax takes priority over Search campaigns for the same queries, except when the Search campaign uses exact match with identical queries. The solution is adding brand keywords as exclusions at the PMax campaign level, one of the most important exclusions to configure.

2. Does the algorithm optimise for the right conversions?

Performance Max optimisation targets the conversions set in the goal. But if the account has multiple conversion actions (purchase, add to cart, newsletter signup), the algorithm may drift towards the easiest ones to achieve. Google recommends using a single primary conversion action for PMax, or assigning different values to conversions.

3. The learning phase: how long does it really take?

Google indicates approximately 6 weeks as the standard learning period. But on accounts with low conversion volume (fewer than 30–50 conversions/month), the algorithm can remain in a de facto learning phase indefinitely. From experience, Performance Max Google Ads tends to perform reliably on accounts generating at least 50 monthly conversions in total; below that threshold, traditional Search campaigns with Smart Bidding often remain more predictable.

4. Placement transparency: the black box

PMax does not show granular ad placement data. There is no per-channel breakdown with separate CPA or ROAS. Google now provides detailed reports on search queries and the ability to add negative keywords at campaign level, a concrete step forward, but still far from the granularity of standard Search.

Performance Max Optimisation: The Real Operational Levers

Despite the black box, concrete levers exist that influence the performance of PMax campaigns:

• Target CPA / Target ROAS: set with realistic targets based on the account's historical data. Starting slightly less restrictive and lowering gradually avoids blocking volume in the first weeks. Official Smart Bidding documentation.
• URL exclusions: exclude non-commercial pages (cart, thank-you, login) to prevent PMax from sending traffic to non-commercial URLs.
• Negative keywords: Google allows negatives at PMax campaign level directly from the interface. PMax negatives guide.
• Separate PMax by product or service type: thematic asset groups work better when supported by separate campaigns, which allow independent budget control and more readable performance data.
• Update assets cyclically: dated assets lose quality score. Rotating images and headlines every 4–6 weeks keeps ratings high.
• Quality Customer Match: customer lists with verified, recent emails (last 90 days) have a markedly greater impact than generic or outdated lists.
• Do not modify the target bid in the first 6 weeks: every significant change to the target CPA/ROAS partially restarts the learning phase.

FAQ: Performance Max Google Ads