Standard remarketing audiences are backward-looking. You target users who visited a product page, who added to cart, who spent more than two minutes on the site. These audiences describe what users did. Predictive audiences describe what users are likely to do next.
GA4’s machine learning models score every user in your property on three probabilities: likelihood to purchase in the next 7 days, likelihood to churn in the next 7 days, and predicted revenue over the next 28 days. GA4 uses these scores to build five pre-built audience templates that sync automatically to Google Ads.
The performance advantage is real: you are bidding to reach users GA4’s model has already identified as high-intent, not users who match a behavioral pattern you defined manually.
This post covers what each predictive audience template is, the eligibility requirements your property must meet, and exactly how to apply each audience in Search, Performance Max, Display, and as exclusions.
How GA4 Builds Predictive Audiences
GA4’s predictive models are trained on the behavioral data in your property — events, sequences, recency, frequency — and on aggregated patterns from Google’s broader data. The models output a probability score for each user across three predictions:
- Purchase probability: The probability a user who was active in the last 28 days will complete a purchase event in the next 7 days.
- Churn probability: The probability a user who was active in the last 7 days will not be active in the next 7 days.
- Predicted revenue: The revenue expected from a user over the next 28 days, based on their purchase probability and historical order values.
These scores are updated continuously and used to populate predictive audiences automatically. You do not define the audience rules — GA4 defines them based on the model outputs.
Eligibility Requirements
Predictive audiences are not available in every GA4 property. Your property must meet minimum data thresholds before the models can be trained:
- At least 1,000 users who triggered the relevant predictive condition (e.g., 1,000 users who have purchased) in the past 28 days
- At least 1,000 users who did not trigger the condition in the same period (for contrast)
- Data must be available for at least 28 consecutive days before the model can be trained
In practice, this means predictive audiences are typically available for mid-size and larger e-commerce properties. If your store does fewer than about 1,000 monthly purchasers, your property likely does not meet the threshold.
To check: in GA4, go to Configure > Audiences > New Audience. If the “Predictive” category appears in the audience templates, your property is eligible. If it does not appear, the thresholds have not been met.
The exact purchase event must be present in your property — not a custom conversion event, but the GA4 ecommerce purchase event. If you are using a different event name for purchases, the purchase probability model will not train.
The Five Predictive Audience Templates
1. Likely 7-day purchasers
Who: Users who GA4 predicts will complete a purchase event within the next 7 days.
Use in Google Ads: Bid modifier increases on Search campaigns (RLSA). These users are in a buying window right now. Increasing bids by 30-50% for users on this list means you spend more to reach them at exactly the moment they are most likely to convert.
In Performance Max: upload as a customer list signal. PMax uses the signal to identify similar users and prioritize existing users who are in-market.
Best practice: Monitor this audience’s purchase rate weekly in GA4 Explorations (segment by this audience, look at purchase conversion rate). If the model is accurate, users in this audience should convert at a materially higher rate than users not in it.
2. Likely 7-day churning users (purchasers)
Who: Users who have previously purchased and GA4 predicts are at risk of churning — i.e., they will not be active in the next 7 days.
Use in Google Ads: Retention campaigns. This audience defines who needs a win-back intervention right now, before they go dormant. Use in Display or Demand Gen with an offer or reminder. For subscription products, use in Search to capture queries from at-risk customers comparing alternatives.
What not to do: Do not bid these users up in prospecting campaigns. They are existing customers at risk — the right message is retention, not acquisition.
3. Likely 7-day churning users (non-purchasers)
Who: Users who have not purchased and GA4 predicts will not be active in the next 7 days — they are about to disengage before converting.
Use in Google Ads: Last-chance remarketing. Capture these users with a specific offer before they disappear. This audience is a subset of your unconverted site visitors who are specifically at risk of going cold.
The window is narrow. These users are still active today but predicted to be gone within a week. Time-limited promotions or urgency-based creative works well here.
4. Predicted top spenders
Who: Users who GA4 predicts will generate the highest revenue in the next 28 days, based on their purchase probability and predicted order value.
Use in Google Ads: High-value customer treatment. Exclude these users from lower-tier campaigns to avoid wasting budget reaching them at a reduced bid. Include them in dedicated high-value RLSA layers with elevated bids and premium creative. For B2B or high-consideration products, use them as a signal for account-based targeting.
Reporting use: In GA4, compare the behavior of users in this audience to your average user. What do their sessions look like? What content do they engage with? What is their acquisition source? The answers often reveal which channels and campaigns are bringing in your highest-value customers — a more useful metric than ROAS alone.
5. Likely first-time purchasers
Who: Users who GA4 predicts will complete their first purchase in the next 7 days.
Use in Google Ads: New customer acquisition campaigns. This audience tells you which of your non-purchasing visitors are on the verge of converting for the first time. Bid them up in Search. Use them as a seed audience for similar audiences or as a signal in PMax to attract users who look like high-probability first-time buyers.
Setting Up Predictive Audiences in GA4
In GA4: Configure > Audiences > New Audience > Predictive.
Select the template you want. GA4 shows a description and, importantly, an estimated size. If the estimated size is very small (under a few hundred users), the audience may not be useful in Google Ads campaigns due to minimum audience size requirements (typically 1,000 users for Search RLSA, 100 for Display).
Name the audience clearly: “GA4 - Likely 7d Purchasers,” “GA4 - Predicted Top Spenders.” The name is what appears in Google Ads.
Set the membership duration. For purchaser audiences, 7 days is appropriate (aligns with the prediction window). For top spenders, 30 days gives the audience time to accumulate enough users.
Save. GA4 will begin populating the audience and sync it to linked Google Ads accounts within 24-48 hours.
To link: Admin > Google Ads Links (if not already linked). The audience appears in Google Ads under Shared Library > Audience Manager > GA4 audiences.
Applying Predictive Audiences in Google Ads Campaigns
Search (RLSA)
In Search campaigns, add predictive audiences in the Audience segment settings at campaign or ad group level.
Use “Bid only” mode to apply a bid modifier to users in the audience while still showing ads to all users matching your keywords. This is the safest approach — you do not restrict reach but you increase bids for high-probability converters.
Recommended bid modifiers:
- Likely 7-day purchasers: +30% to +50%
- Likely first-time purchasers: +20% to +30%
- Predicted top spenders: +40% to +60%
- Likely churning non-purchasers: +10% (urgency play, but cap spend)
Use “Target” mode (RLSA targeting) only when you want to show ads exclusively to users on the list. Useful for retention campaigns targeting churning users.
Performance Max
PMax does not support RLSA in the traditional sense, but audience signals guide the model. Add predictive audiences as signals in the Asset Group > Audience signals section.
The signal tells PMax: “find users who look like this.” PMax does not guarantee it will show ads only to these users, but the signal shifts the model toward similar users.
For high-value customer campaigns, create a dedicated PMax asset group with top spender audience signals and creative focused on retention and upsell. Separate this from your acquisition-focused asset groups so the model and creative are aligned.
Display and Demand Gen
Add predictive audiences as targeting audiences in Display campaigns. For churning purchasers, a Display retargeting campaign with a “we miss you” or win-back offer is the standard use case.
For Demand Gen, use the likely purchaser audiences to reach users who are actively in a buying window across YouTube, Discover, and Gmail placements.
Exclusions (Often Overlooked)
Predicted top spenders should be excluded from your prospecting and low-bid campaigns. You do not want your most valuable customers clicking a generic acquisition ad at a generic acquisition bid — you want them in a dedicated experience.
Exclude churning users from new customer acquisition campaigns. They are not new customers; they are at-risk existing customers. Reaching them with acquisition messaging is both wasteful and confusing.
Measuring Whether Predictive Audiences Are Working
The validation question is: do users in the “likely purchaser” audience actually purchase at a higher rate than comparable users not in the audience?
In GA4 Explorations:
- Create a Free-form Exploration.
- Add a user segment for “GA4 - Likely 7d Purchasers.”
- Add a contrasting segment for “All users not in the purchaser audience.”
- Metrics: Sessions, Purchases, Purchase rate.
- Compare the purchase rates.
If the model is working, the audience purchase rate should be meaningfully higher — often 3-5x or more compared to unscored users.
In Google Ads: monitor conversion rate and ROAS for campaigns and ad groups where you have applied bid modifiers for predictive audiences vs. those without. If the bid modifier is justified, the ROAS for audience-targeted clicks should be higher than baseline.
When Predictive Audiences Do Not Work
Property is below eligibility threshold: The Predictive category does not appear in the audience builder. Continue growing your purchase event volume and check again after 28 days.
Audience size is too small for Google Ads: Even eligible properties may have predictive audiences too small to meet Google Ads minimums (1,000 users for Search RLSA). This is common for small-medium e-commerce stores. In this case, use GA4 predictive metrics in BigQuery for analysis but rely on behavioral audiences in Google Ads until volume grows.
Purchase event is not the standard GA4 event: If your property tracks purchases under a custom event name, the purchase probability model will not train. Ensure your conversion events use the standard purchase event name.
Key Takeaway
Predictive audiences are GA4’s most differentiated feature for Google Ads advertisers. Behavioral audiences are available in every analytics platform. Purchase probability and churn scoring are not.
The strategic shift is from “who visited my product page” to “who is likely to buy in the next 7 days.” The audience definition becomes a machine learning output rather than a human-defined rule — and for high-volume properties, the model’s definition consistently outperforms manual segmentation.
The setup is simple once eligibility is met. The value is in the activation: using these audiences as bid signals, exclusion layers, and PMax signals across your full Google Ads account structure.
Up next in the GA4 Advanced Series: GA4 Reporting Identity — Why Your User Counts Are Wrong and How to Fix It
Related Posts
GA4 Channel Groupings: Why the Default Is Wrong and How to Fix It
12 min read
GA4 Cohort Analysis: How to Measure Retention and Campaign Quality Over Time
11 min read
GA4 Reporting Identity: Why Your User Counts Are Wrong and How to Fix It
11 min read
Need Help With Your Google Ads?
I help e-commerce brands scale profitably with data-driven PPC strategies.
Get In Touch