Google Marketing Live 2023 delivered the clearest signal yet of where Google intends to take automated advertising: Performance Max is the centerpiece of the Google Ads product roadmap, and AI-generated creative assets are becoming a core feature rather than an experimental add-on. The practical question for advertisers is not whether to engage with PMax in its expanded form — the reach and placement breadth make it effectively unavoidable at scale — but what you are actually giving up when you let Google write, select, and optimize your ads autonomously.

The headline announcements from Marketing Live include: AI-generated image and text assets for Performance Max campaigns, expanded audience signals that allow advertisers to provide first-party data inputs while Google’s AI determines the actual audience delivery, product-level reporting improvements for retail advertisers, and deeper integration between Merchant Center and PMax for shopping-led campaigns. Each of these individually represents genuine product improvement. Together, they describe a product trajectory where the advertiser’s role is increasingly to set business objectives and provide raw inputs, while Google’s AI makes the creative, audience, and bidding decisions.

What AI-Generated Assets Actually Means in PMax

Performance Max already generated ad variations from advertiser-provided assets — headlines, descriptions, images, and videos — combining them algorithmically to find the highest-performing combinations across Search, Display, YouTube, Gmail, and Discover placements. What’s new in the 2023 announcements is that Google will now generate creative assets itself — specifically images and additional text variations — when the advertiser’s provided assets are insufficient or when Google’s AI determines that additional variation would improve performance.

Google’s Marketing Live blog post describes AI-generated image assets as being produced from the advertiser’s landing page and existing creative, with advertisers retaining the ability to review and approve or reject generated assets before they serve. This is the workflow in theory. In practice, large-scale PMax campaigns with many asset groups and frequent creative refresh cycles create significant review volume — the approval mechanism exists, but the throughput required to actually review every AI-generated asset is non-trivial for most marketing teams.

The creative accountability question is real: when a Google AI generates an image or headline that appears in a brand’s advertising, who is responsible for the accuracy, appropriateness, and brand-safety of that asset? The advertiser is, contractually and reputationally. If an AI-generated image is off-brand, contains inaccurate product depictions, or appears alongside messaging that creates an unintended brand association, the brand bears the consequence even though it did not produce the creative.

This is not a theoretical risk. AI-generated images from current models have documented failure modes: incorrect text rendering, implausible physical details, brand mark misrepresentation, and demographic representation issues that can generate brand controversy. Applying these tools at the scale of a live advertising campaign, across multiple placements, with review volume that makes individual approval impractical, is a meaningful brand risk management challenge.

Audience Signals: The Difference Between Input and Control

PMax’s audience signal mechanism has been the most technically misunderstood element of the product since its 2021 launch. Audience signals are inputs that tell Google’s AI what kind of users are likely to convert — they are not audience targeting instructions in the traditional sense.

When you provide a customer list as an audience signal in PMax, you are telling Google’s AI: “people who look like these customers are probably good candidates for my campaign.” Google’s AI then decides the actual delivery based on its own optimization models, which include your audience signals as one input among many. The AI can and does serve to users outside your provided signals if its model predicts conversions from those users.

The 2023 Marketing Live announcements expand the audience signal inputs — including better integration of first-party CRM data and customer purchase history signals. This is a genuine improvement for advertisers who have rich first-party data, because it provides better raw input to Google’s optimization models. It does not change the fundamental dynamic: Google’s AI controls delivery, and your audience signals inform but do not constrain it.

The implication for advertisers is important in both directions. PMax’s AI-controlled delivery can find converting audiences that explicit targeting would have missed — this is real performance upside. PMax’s AI-controlled delivery can also serve to audiences outside your intended scope, including competitor brand names, privacy-sensitive categories, or demographic segments your campaign was not designed to reach. The reach and optimization benefits come with loss of precision control.

Brand Safety Within PMax: The Category Exclusion Gap

Brand safety controls in Performance Max have been a persistent advertiser concern since the product’s launch, and the 2023 announcements address some but not all of the gaps.

Brand exclusions — preventing PMax from bidding on your own brand’s search terms — were introduced in 2022 and have been expanded. Topic exclusions for sensitive content categories are available in Display and YouTube placement contexts within PMax. However, comprehensive keyword-level exclusions that are standard in Search campaigns are not available within PMax — the campaign type’s autonomous placement optimization is structurally incompatible with the same granular keyword control that standard Search campaigns support.

For advertisers in regulated industries — financial services, healthcare, alcohol, gambling — the keyword exclusion gap is material. A pharmaceutical company running PMax may have AI-generated or AI-delivered ads appearing against health queries where specific claim restrictions apply. The coarse topic exclusions available do not substitute for the specific keyword-level control that compliance teams rely on in standard Search.

The practical answer for many advertisers is a hybrid campaign architecture: Performance Max for the broad reach and AI optimization it does well, supported by traditional Search, Display, and Video campaigns with explicit controls for the audience and content contexts where brand safety or compliance requirements demand precision. PMax is not a full replacement for managed campaigns for advertisers with specific control requirements.

When PMax’s Automation Is Net Positive Versus Negative

The straightforward framework: PMax works well for advertisers with broad audience definitions, clear conversion signals, strong product catalog data (for shopping-led campaigns), and limited brand safety or compliance constraints. E-commerce brands with large product catalogs, clear purchase conversion events, and broad target demographics are the natural fit.

PMax works less well for advertisers with narrow target audiences (B2B enterprise, high-value low-volume conversion models), strict keyword or topic exclusion requirements (regulated industries), campaigns where creative message control is critical (political, advocacy, sensitive topics), or campaigns where the advertiser needs to understand precisely who was reached and why (sophisticated audience management).

Google’s own guidance on Performance Max best practices recommends PMax as a complement to existing campaign types for most advertisers, not a full replacement. The AI-generated assets and expanded audience signals make PMax more capable in 2023 than it was at launch. They do not change the fundamental trade-off: automation at scale in exchange for reduced transparency and control.

FAQ

What are AI-generated assets in Performance Max and how does review work? AI-generated assets are images and text variations that Google’s AI creates from your landing page and existing creative when it determines additional variation would improve performance. Advertisers can review and approve or reject generated assets before they serve. In large campaigns with many asset groups, the review volume can be substantial — setting up a consistent review process is essential before enabling the feature.

Do audience signals in PMax function the same as audience targeting in other campaign types? No. Audience signals in PMax are optimization inputs — they tell Google’s AI which user types are likely to convert — but they do not constrain delivery the way audience targeting does in Display or Discovery campaigns. PMax may serve to users outside your provided signals if its optimization model predicts conversions. This is intentional design, not a bug.

What brand safety controls are available within Performance Max? PMax supports brand exclusions (preventing bidding on your own brand terms), topic exclusions for sensitive content categories in Display and YouTube contexts, and some content suitability settings. Standard keyword-level exclusions that are available in Search campaigns are not supported in PMax. Advertisers with specific keyword exclusion requirements for compliance or brand safety should supplement PMax with separate managed campaigns.

When should an advertiser use Performance Max versus traditional campaign types? Performance Max is well-suited for e-commerce brands with large product catalogs, clear purchase conversion signals, and broad target demographics. Traditional campaign types (Search, Display, Video App) are better suited for advertisers needing precise keyword control, narrow audience targeting, specific creative message control, or detailed transparency into placement and audience decisions.