AUDIENCE SIGNALS PMAX: FULL COVERAGE WITHOUT LIMITS

One of the most common misconceptions in managing PMax campaigns concerns audience signals in Performance Max: many advertisers avoid them or use them cautiously, fearing they will limit campaign coverage. The logic seems intuitive, if I tell the algorithm to focus on certain segments, don't I risk excluding the others?

The answer is no, and understanding why radically changes how you configure this part of the campaign. PMax audience signals are not targeting constraints in the traditional sense. They don't work like audience targeting in a standard Display campaign. They are a starting point for machine learning, not a perimeter defining who can see the ads.

In this article I analyze how to add audience signals to PMax correctly, starting from official Google documentation and flagging where direct experience on real accounts leads to more precise conclusions, particularly on which signals genuinely accelerate learning and which add no measurable value.

What Audience Signals Are in Performance Max

According to official Google documentation, audience signals are "suggestions provided by the advertiser to help the algorithm find customers with a higher probability of conversion more quickly." Google explicitly specifies that these are a starting point for optimization, not a targeting constraint.

The mechanism works as follows: without signals, the algorithm starts from scratch and must explore a very broad universe of users to understand who converts and who doesn't. With quality signals, Google uses that more precise starting area to learn faster, autonomously expanding toward adjacent segments with a similar conversion probability.

The practical consequence is counterintuitive compared to how targeting works in standard campaigns: adding very precise signals does not reduce final coverage, because the algorithm only uses them in the initial phase. After 4–6 weeks, the model built on the account's conversion data weighs far more than the original audience signals. The signals remain one of the inputs, but not the primary determinant of placement decisions.

What experience shows: the most visible difference between campaigns with and without quality audience signals is not in final coverage, it is in the stability of the first weeks. Campaigns with active Customer Match and quality lists show a less volatile ROAS in the first 2–3 weeks compared to campaigns without signals, where performance swings are systematically wider. Coverage reached after 8 weeks is substantially equivalent in both cases, the difference is in the speed of reaching stability.

The Types of Audience Signals Available in PMax

According to official documentation, four categories of audience signals can be added per asset group in PMax. I analyze each with a comparison between what Google states and what emerges from practical use:

1. Customer Match, the most powerful signal

According to Google's Customer Match documentation, this feature lets you upload customer lists (email, phone number, address) that Google matches to its own users. It is the signal that most precisely describes who the advertiser's real customer is.

Customer Match in Performance Max works best with lists of at least 1,000 matched users updated within the last 90 days. Older lists or those with fewer matched users have significantly reduced impact, the system doesn't have enough fresh data to build a reliable profile.

Inference on Customer Match: on accounts where the Customer Match list was segmented, recent purchasers separated from newsletter subscribers, the signal produced by the purchaser list is more precise and produces a more stable learning phase than a single list that mixes different intent levels. Granularity in list segmentation improves signal quality.

2. Website visitors, remarketing as a signal

Remarketing audiences in PMax based on website visitors, technically defined as audiences based on browsing behavior, are one of the most effective signals after Customer Match, because they represent users who have already expressed explicit intent toward the brand. Not all visitors have equal value as a signal: someone who visited the product page or cart is a much more precise signal than someone who only visited the homepage.

The correct practice is to segment remarketing audiences by intent level and use high-intent ones (abandoned cart, product page) as primary signals. Generic site visitors can be used as secondary signals, but their impact on accelerating learning is lower.

3. Custom audiences based on intent keywords

According to Google's custom segments documentation, it is possible to create segments based on keywords the target audience searches for, sites they visit, or apps they use. In PMax, these custom audiences work as a more granular intent signal compared to Google's predefined in-market segments.

A custom audience built on sector-specific queries (e.g., "running shoes for overpronation", "CRM software for SMBs") allows more precise user targeting than a generic in-market segment like "Sportswear" or "Business software."

4. In-market and interest segments

Google's predefined in-market segments are the signal type with the lowest impact on learning acceleration. Google already knows them, they are part of the contextual signals the algorithm evaluates autonomously for every auction. Adding them as audience signals in PMax is generally not enough to produce significant incremental results beyond what the algorithm would discover on its own in the first weeks.

Inference on in-market segments: on accounts where in-market segments were the only configured audience signals, first-4-week performance did not differ significantly from accounts with no signals. The difference versus campaigns with Customer Match or remarketing was, however, consistent and repeatable across multiple accounts.

How to Add Audience Signals Without Losing Coverage

The fundamental question, how to configure Performance Max audience signals while maintaining maximum coverage, has a precise operational answer. The key principle is that coverage does not depend on how many signals you add or how specific they are: it depends on how the algorithm interprets them, which is as suggestions, not as constraints.

The strategy that produces the most stable results, based on experience across accounts of various sizes:

• Use different signals for different asset groups: instead of putting the same signals on all asset groups, assign to each asset group the signals most consistent with its theme, just as you differentiate creative assets between groups. The "running shoes" asset group receives the visitors of the running shoes page as a signal, not the generic audience of the whole site. This increases differentiation between asset groups and improves the algorithm's ability to understand when to activate each one.
• Prioritize Customer Match as the first signal: for accounts with available customer data, Customer Match should always be configured before other signals. It is the signal with the highest information impact because it describes who has already bought, not who might buy.
• Combine signals at different levels: Customer Match (who has already bought) + high-intent visitors (who showed recent interest) + custom audience on keywords (who is actively searching) forms a combination that covers the funnel without being restrictive. The algorithm starts from these three levels and expands toward adjacent segments with similar characteristics.
• Don't add overly broad signals as the only input: using only generic in-market segments or broad interests as the sole signals does not significantly accelerate learning and can give the algorithm information that is less useful than what it would discover on its own.
• Update Customer Match every 30–60 days: static lists that age lose value as a signal because customer behavior changes. A regularly updated Customer Match keeps the typical converter profile aligned with the current business reality.

Inference on non-limitation of coverage: on accounts where audience signals included Customer Match with very specific and small lists (e.g., only the top-100 customers by value), total PMax campaign coverage after 8 weeks did not differ significantly from accounts with broader signals. The algorithm uses the top-customer profile as a starting point for finding similar users, not as an exclusive perimeter. Final coverage is determined by the bidding target and budget, not the specificity of the signals.

The Mistakes That Actually Reduce Coverage, and Those That Don't

Separating real mistakes from perceived ones is essential for correctly configuring PMax audience signals. The coverage of a PMax campaign is primarily determined by three factors: budget, bidding target, and conversion tracking quality. Audience signals influence the efficiency of the learning phase, not the volume of reachable coverage.

Mistakes that do NOT reduce coverage

• Adding very specific or small signals: a Customer Match list of 500 customers does not limit PMax to reaching only those 500 users. The algorithm uses them as a reference profile and expands toward millions of similar users.
• Using multiple signals simultaneously: adding 5 different signals to an asset group does not intersect audiences as traditional targeting would. Each signal is an additional input, not a restrictive filter.
• Niche sector-specific signals: even if the sector has low search volumes, the algorithm uses signals to orient itself among available opportunities, it doesn't reject them for being too specific.

Mistakes that actually reduce coverage

• Overly aggressive target ROAS: this is the real culprit behind under-coverage in most cases. An unrealistic target ROAS limits the auctions PMax can participate in, regardless of the configured signals.
• Insufficient budget: a daily budget that is too low doesn't allow the algorithm to collect enough data to explore available channels and segments. Coverage remains compressed not because of signals, but because of the budget.
• Identical audience signals across all asset groups: this doesn't reduce total coverage, but reduces differentiation between asset groups. The algorithm can't understand when to activate one over another, and tends to concentrate on the one it perceives as stronger, leaving others underutilized.
• Overly aggressive placement exclusions: excluding many sites or content categories via Brand Safety Center effectively reduces the volume of available inventory. Unlike audience signals, placement exclusions are real constraints that limit accessible placements.

Inference on diagnosing under-coverage: when a PMax campaign consistently spends less than 75% of the available budget, the cause in the vast majority of analyzed cases was the target ROAS or tCPA, not the audience signals. Removing audience signals to "unlock" coverage is one of the most frequent incorrect solutions: it leaves the algorithm without guidance during the learning phase without solving the real problem.