The Evolution of Audience Targeting in Demand Gen
Google is fundamentally restructuring how advertisers reach new customers within its Demand Gen campaigns. In a significant move toward an AI-first ecosystem, Google has announced that Lookalike segments will transition from strict targeting constraints to optimization signals. Scheduled to take full effect by March 2026, this shift represents a departure from the traditional “walled garden” approach to audience building, favoring a more fluid, machine-learning-driven model.
Demand Gen campaigns, which replaced Discovery Ads, are designed to capture interest across Google’s most visual and immersive surfaces, including YouTube (Shorts, In-stream, and Feed), Google Discover, and Gmail. Central to these campaigns has been the “Lookalike” segment—a tool that allows advertisers to upload a seed list of existing customers and ask Google to find similar users. Under the new update, the role of that seed list is changing from a hard boundary into a directional compass.
The Technical Shift: From Constraints to Signals
To understand the weight of this update, it is essential to distinguish between a “constraint” and a “signal.” In the legacy version of Lookalike targeting, advertisers selected a similarity tier: Narrow (top 2.5% of similarity), Balanced (top 5%), or Broad (top 10%). The algorithm was strictly bound to these percentages. If a user fell outside that specific similarity pool, they would not see the ad, regardless of how likely they were to convert at that specific moment.
Starting in March 2026, these tiers will act as “optimization signals.” This means that while Google’s AI will prioritize the users within those defined similarity pools, it is no longer forbidden from venturing outside of them. If the system’s predictive modeling identifies a user who is highly likely to convert but technically falls outside the “Broad” 10% similarity tier, the system can now serve an ad to that user.
This transition effectively reframes the Lookalike segment. It is no longer a fence that keeps the campaign within a specific yard; it is a signal that tells the AI where to start looking, while granting it the autonomy to follow the scent of a conversion wherever it leads.
Comparing the Before and After Models
The practical implications for digital marketers are vast. Let’s break down the structural differences between the two models to better understand the impact on day-to-day campaign management.
The Legacy Model (Pre-March 2026)
Under the old system, advertisers had a high degree of predictability regarding who would see their ads. By choosing a “Narrow” tier, a brand could ensure that their budget was spent only on the users most mathematically similar to their existing customer base. This was ideal for niche products or brands with very specific buyer personas. However, the downside was a “scale ceiling.” Once the system exhausted the high-intent users within that narrow pool, performance would often plateau or costs-per-acquisition (CPA) would spike as the system struggled to find more conversions within a limited set of users.
The New Signal-Based Model
In the new model, the tiers still exist, but they function as a weighted priority. The AI uses the Lookalike list as a high-quality data source to understand the characteristics of a “good” customer. However, it combines this with real-time intent signals—such as recent search history, app usage, and video consumption—to find conversions that a strict similarity model might miss. This approach is designed to maximize conversion volume and lower the average CPA by allowing the algorithm to bypass the artificial boundaries of a percentage-based list.
The Synergy with Optimized Targeting
A critical component of this update is how it interacts with Google’s existing “Optimized Targeting” feature. Optimized Targeting is a setting that allows Google to look beyond your selected audience segments to find conversions you may have missed. When Lookalike segments become signals, they will stack with Optimized Targeting to create a powerful, albeit less transparent, engine for growth.
If an advertiser enables both, the Lookalike signal provides the “who,” while Optimized Targeting provides the “how and when” for expansion. This layering allows Google’s AI to pursue a broader reach while still keeping the campaign anchored in the brand’s first-party data. For performance marketers, this means the system has more freedom than ever to pursue the most efficient conversions across the entire Google network.
Why Google is Moving Toward AI Signals
The shift toward signal-based targeting is not an isolated event; it is part of a broader industry trend toward “Black Box” advertising. Several factors are driving Google to make this change, ranging from technical necessity to performance optimization.
1. Overcoming the Scale Cap
Strict Lookalike targeting often leads to diminishing returns. As campaigns mature, they frequently hit a wall where they cannot find new users within the narrow similarity pool. By converting these pools into signals, Google allows the campaign to scale more naturally. This is particularly important for Demand Gen campaigns, which are designed to sit at the top and middle of the marketing funnel, where high volume is a primary goal.
2. Navigating a Cookie-Less Future
The digital advertising landscape is moving away from granular tracking and third-party cookies. As traditional tracking becomes less reliable, Google is leaning into “modeled behavior.” AI signals allow the system to use aggregated, anonymized data to predict behavior rather than relying on individual tracking. This makes the platform more resilient to privacy changes and browser-level tracking preventions.
3. Reducing Model Complexity
Maintaining high-quality similarity models for every single advertiser is a massive computational task. By shifting to a more generalized AI suggestion model, Google can streamline its internal processing while potentially delivering better results for the advertiser through a more holistic view of user intent.
Strategic Implications: What Advertisers Need to Do
For brands and agencies, the move to signal-based Lookalikes requires a shift in strategy. The focus is moving away from “who we target” and toward “what data we feed the machine.”
Prioritize High-Quality First-Party Data
Because the Lookalike segment is now a signal, the quality of that signal is more important than ever. Advertisers should focus on building seed lists based on high-value actions, such as actual purchasers or long-term subscribers, rather than just website visitors. The more “pure” the seed list, the better the AI can understand what a successful conversion looks like.
Embrace a Testing Mindset
As these changes roll out, it is vital to monitor how the expansion affects your key performance indicators (KPIs). Advertisers should run A/B tests—or “experiments” in Google Ads—to compare the performance of strict targeting versus signal-based targeting. Pay close attention to lead quality; while AI expansion often increases conversion volume and lowers CPA, it can sometimes result in lower-quality leads if the AI wanders too far from the intended persona.
Rethink Creative Assets
In an AI-driven environment, creative becomes the primary lever for targeting. If Google is expanding your reach beyond your specific list, your ad creative must do the work of filtering the audience. High-quality, resonant visuals and clear messaging in Demand Gen ads will help ensure that even if the AI reaches a “new” user, the ad itself attracts the right kind of attention.
The Opt-Out Clause: Maintaining Control
Google recognizes that some advertisers—particularly those in highly regulated industries or those with extremely specific niche requirements—may not want the AI to expand beyond their defined pools. For these users, Google has provided a dedicated opt-out form.
By opting out, advertisers can request to maintain the legacy “strict” Lookalike targeting. However, it is important to note that this may eventually lead to higher CPAs or limited scale compared to competitors who embrace the AI-driven expansion. Advertisers should only consider the opt-out if they have proven through data that expansion negatively impacts their bottom line or if they have strict legal requirements regarding who can see their ads.
The Global Trend: Mirroring Meta and Beyond
This move mirrors changes seen across the social media and programmatic landscape over the last few years. Meta (formerly Facebook) introduced “Advantage+ Audience” and “Advantage+ Shopping Campaigns,” which function on a nearly identical premise: the advertiser provides suggestions, but the AI makes the final decision on delivery.
We are seeing a universal trade-off in digital marketing: advertisers are giving up granular, manual controls in exchange for the sheer processing power of machine learning. While the loss of control can be jarring for veteran marketers who are used to tweaking every knob and lever, the reality is that in most mainstream use cases, these AI-led optimizations have led to measurable performance gains.
Timeline and Transition Period
The transition to signal-based Lookalikes in Demand Gen is not happening overnight. Google has provided a significant lead time, with the final transition set for March 2026. This gives advertisers over a year to adjust their strategies, test new audience configurations, and gather enough data to decide whether to embrace the shift or utilize the opt-out form.
During this transition period, it is recommended to start treating your Lookalike segments as signals mentally. Begin experimenting with broader tiers and observing how Google’s automated expansion affects your reach and conversion metrics. Those who adapt early will likely have a competitive advantage when the strict targeting model is eventually deprecated.
The Bottom Line for Performance Marketers
Google’s shift to AI signals in Demand Gen is a clear indicator of the future of the platform. We are moving toward a world where the marketer’s job is less about managing lists and more about managing data and creative. By turning Lookalikes from a “fence” into a “compass,” Google is betting that its AI can find customers more effectively than a human-defined percentage ever could.
As we move toward 2026, the success of a Demand Gen campaign will depend on three pillars: the quality of the first-party data used as a signal, the clarity and relevance of the creative assets, and the advertiser’s willingness to trust the algorithm to find performance in unexpected places. While the loss of strict control may be uncomfortable, the potential for increased scale and improved efficiency makes this one of the most important updates to the Google Ads ecosystem in recent years.