PMax: The Signals That Guide Ad Placements

The most frequent question from people managing PMax campaigns is also the hardest to answer with concrete data: how does the algorithm decide where to show my ads? Understanding the logic behind Performance Max signals is not just a technical curiosity exercise — it is the foundation for optimising campaigns on solid ground rather than acting blindly on a black box.

Performance Max PMax campaigns represent a distinct campaign type — one that uses machine learning to find customers most likely to convert across all channels. That description is correct but vague. What the official documentation does not articulate is which specific signals feed the model, how they are weighted against each other, and above all how the advertiser's choices — in terms of assets, audience signals, and conversion data — affect the quality and consistency of the signals the algorithm receives.

This article analyses the Performance Max signals mechanism starting from Google's official documentation, flagging where direct experience on real accounts — and inference from observed patterns — leads to more precise conclusions about how PMax decides placements and what can be done to shape those decisions.

The Signal Categories PMax Uses in Every Auction

This list is a starting point, not a complete description. From pattern analysis on real accounts and cross-referencing with available documentation on Smart Bidding and Smart Creative solutions, the signals the PMax algorithm uses for each placement decision fall into four distinct categories with different dynamics and weights:

Google's AI combines these four signal categories simultaneously across Search Display YouTube, Gmail, Maps, and Discover — selecting the placement and bid that maximise the estimated probability of conversion for each user in each moment.

PMax Contextual Signals: The Auction Moment

PMax contextual signals are the fastest-changing ones — they vary from auction to auction based on the characteristics of the moment the placement request occurs. Google documents them explicitly in the About Smart Bidding section as determinants for estimating conversion probability in every single auction.

The main contextual signals PMax evaluates in real time include:

• Search query — the exact text searched by the user, including semantic variants and implicit intentions in the phrasing. On Search, this is the most powerful contextual signal: a query like "men's running shoes for overpronation" contains more intent information than any demographic signal. Search themes added at asset group level act as hints that steer the algorithm toward relevant query territory without restricting it to exact matches.
• Device — mobile, desktop, or tablet. PMax automatically adapts bids and ad formats to the device. From experience, conversion rates by device differ significantly across almost all sectors — and PMax learns these patterns from the account within a few weeks.
• Time and day — the time of day and day of the week. PMax does not have manual dayparting like standard Search campaigns, but it incorporates time and day as bid-weighting signals. In many accounts, impression concentration in the time slots with historically higher conversions is observable — a direct effect of learning from these signals.
• Geographic location — the user's position at the time of the auction, with granularity down to city and neighbourhood for mobile. Geographic conversion patterns are learned by the algorithm and influence budget distribution across geographic areas.
• Page context — for Display and YouTube placements, the theme and keywords of the page or video where the ad would appear. This is the contextual signal that distinguishes "relevant" display placements from generic ones. Keyword lists used in custom audiences help reinforce this contextual layer by describing the topics and searches that define your category.

PMax User Signals: The Behavioural Profile Over Time

PMax user signals are the ones Google accumulates on the user over time across the entire Google product network — Search, YouTube, Gmail, Maps, Chrome, Android. These signals build a behavioural profile that goes far beyond a single session or a single query: they represent the user's pattern of interest, intent, and purchase behaviour over an extended time horizon.

The main components of user behavioural signals that PMax evaluates:

• Recent search history — queries the user has typed on Google in the past hours and days. A user who has searched "best trail running shoes" three times in the past week receives greater weight in auctions for outdoor footwear campaigns than someone who has never searched anything related.
• YouTube videos watched — the type of content consumed on YouTube builds a detailed interest profile. A user who regularly watches tech product review videos is a commercial intent signal for those categories, regardless of what they are searching for at that moment.
• Gmail behaviour — emails received and opened (e.g. e-commerce newsletters, competitor order confirmations) can constitute commercial interest signals that Google uses to segment users in its own systems.
• Chrome browsing history — for users signed in to their Google account on Chrome, browsing history provides very precise interest and intent signals. It is the highest-granularity signal Google has available.
• Previous ad interactions — clicks on similar ads, views of promotional videos, visits to product pages after an ad interaction. PMax uses these signals to identify users "in the consideration phase" — not yet ready to convert but in an active purchase journey.
• Custom segments — audiences built from specific keyword searches, website visits, or app usage patterns. These are the most effective user-side inputs for describing your ideal customers at the intent level, since they let you define the audience by the exact searches and behaviours that characterise a qualified prospect.

Conversion and Account Signals: The Learning Engine

In practice, PMax uses conversion data to do two distinct and complementary things:

• Build the profile of the "typical converter" — by analysing who has converted in the past (demographic, behavioural, and contextual characteristics at the time of conversion), the algorithm builds a statistical model of who is likely to convert in the future. This model is continuously updated as new conversions arrive.
• Estimate conversion probability in every auction — for every user and every placement context, the algorithm estimates in real time the probability that user will convert, comparing their profile against the typical converter model. This estimate directly determines the bid in every auction.

The quality of this model depends entirely on the quality and volume of available conversion data. This is where the most practically relevant consequences for campaign optimisation emerge:

• Minimum conversion volume: below 30–50 monthly conversions in the account, the model does not have enough data to be statistically reliable. Placement decisions become noisier and less precise — explaining the performance instability typical of small accounts running PMax.
• Quality of conversion actions: if the account tracks "easy" conversions (e.g. page view, scroll depth) alongside conversions with real value (e.g. purchase, qualified lead), the typical converter model is built partly on the former — skewing placement decisions towards users with a high probability of scrolling, not necessarily of buying.
• Conversion lag: for products with long decision cycles (B2B, high-ticket), the lag between first contact and conversion can be weeks. PMax learns slowly in these contexts because the feedback (the conversion) arrives late relative to the auction that generated it.

The role of audience signals as ignition signals

According to the official Google documentation on audience signals, these do not define who sees the ads — they define the starting point for learning. The algorithm uses audience signals as acceleration inputs in the initial phase, then detaches from them progressively as it accumulates its own conversion data.

The practical consequence is that audience signals matter most in the first 4–6 weeks of a PMax campaign's life, when the account's own conversion data is scarce or absent. High-quality audience signals — Customer Match with a recent customer list, visitors of high-intent pages — allow the algorithm to start from a more precise exploration area, reducing the "noise" of the initial learning phase.

After the learning period, the weight of audience signals decreases progressively relative to conversion signals built by the account. They never disappear completely, but they become one of many inputs rather than the algorithm's primary orientation.

How to Influence the Signals PMax Receives

Understanding the Performance Max signals categories is useful above all because it identifies concrete levers for improving the quality of the algorithm's decisions. Some of these levers are documented by Google; others emerge from experience on real accounts:

• Clean up conversion tracking — this is the lever with the greatest impact on conversion signal quality. Keep as primary conversions only actions with real economic value, and demote soft actions to secondary conversions not used for optimisation. Google's documentation on primary and secondary conversions describes how to configure this distinction in the account.
• Quality Customer Match audience signals — providing lists of existing customers with verified, recent emails (last 90 days) is the most effective way to influence the starting point of learning. A Customer Match list of 1,000+ active customers tells the algorithm with precision who your real customers are — accelerating construction of the typical converter model. First-party data of this kind is the strongest input an advertiser can provide, because it reflects actual purchase behaviour rather than inferred interest.
• Create asset groups coherent with the activation context — asset group content is an explicit signal the advertiser provides about when and how to activate each group. Thematic asset groups with headlines and images aligned to the product or service segment let the algorithm better match content to query or placement context. If you do not supply a video, PMax auto generates video from other assets — a process that reliably produces lower-quality output than a brand-supplied creative and reduces signal precision on YouTube placements.
• Complete product feed for e-commerce — feed data is an explicit signal PMax uses for matching between products and Shopping queries. A feed with optimised titles, correct GTINs, and complete attributes improves the precision of contextual signals for Shopping auctions.
• Offline conversion import — for businesses with offline sales or purchase cycles, importing offline conversion data (phone sales, signed contracts, in-store purchases) via Google Ads' Offline Conversion Import function provides the algorithm with conversion data more faithful to commercial reality than clicks and online forms alone.