Google Ads: 2026 AI Retargeting for 15% ROAS

Listen to this article · 12 min listen

In 2026, the customer journey is a chaotic web of touchpoints, so getting back in front of users who’ve already shown interest is how you actually close a sale. AI predictive retargeting is the smart way to do this, going way beyond old rule-based systems to actually guess what users want and hit them with ads they’ll care about. This tutorial is a rundown on how to set up a serious AI retargeting campaign in Google Ads using its newest predictive tools. So, how do you turn all those past clicks and views into actual revenue?

Key Takeaways

  • Your first job is to feed Google Ads good data, which means configuring Google Analytics 4 (GA4) with solid enhanced e-commerce tracking and enabling its predictive audiences.
  • Build smarter predictive segments inside Google Ads by mixing those GA4 signals with your own first-party CRM data for much better accuracy.
  • Set up dynamic creative optimization (DCO) through the Google Ads Asset Library. This lets the AI build personalized ad variations on the fly based on what it thinks a user wants to see.
  • Use Performance Max campaigns, but make sure to feed them with strong audience signals so Google’s AI can effectively find and convert high-value past visitors across its entire network.
  • Live in the “Diagnostics” and “Insights” tabs in your Google Ads account to constantly tweak your audience signals and bidding, with the goal of bumping your ROAS by 15% in the first quarter.

Step 1: Laying the Data Foundation with Google Analytics 4 (GA4)

Let’s be clear: your retargeting is only as good as your data. GA4 and its event-driven model are the only real foundation for this kind of AI work, so just having it installed isn’t enough. You have to configure it to pull out maximum insight.

1.1. Verify Enhanced E-commerce Tracking

Go into your Google Analytics 4 property. In the nav on the left, find Admin (the gear icon) and go to Data Streams. Pick your web stream. Under “Google tag,” hit Configure tag settings. From there, go to Modify events and double-check that your core e-commerce events like add_to_cart, begin_checkout, purchase, and view_item are all firing correctly with the right parameters (like value, items, and currency). Any mistakes here will poison the predictive model before it even starts. I constantly find people have forgotten the item_list_name parameter, which completely blinds you to how users are interacting with different product categories or sales.

1.2. Enable Google Signals

Google Signals is what gives you cross-device tracking and richer demographic data, which are the ingredients for good audience insights. In the GA4 Admin panel, go to “Data collection and modification,” click Data Settings > Data Collection, and flip the switch for Google signals data collection. This isn’t optional for AI campaigns. It gives the algorithms a much more complete picture of a single user’s behavior as they hop between their phone and desktop, and the AI needs that full picture to make good predictions.

1.3. Configure Predictive Audiences

The predictive stuff baked into GA4 is a huge advantage. Inside your GA4 property, go to Configure > Audiences and click New Audience. You’ll immediately see “Suggested Audiences” under a “Predictive” heading for things like “Likely 7-day purchasers” or “Likely 7-day churning users.” Just pick one, look over its default conditions, and hit Save. You can get more specific by creating a custom predictive audience, maybe something like “Users likely to buy in 7 days AND who have also viewed more than 3 product pages in the last 30 days.” That’s a powerful segment. Once you save these, they’ll show up automatically in your linked Google Ads account, ready to go.

Step 2: Crafting Advanced Audience Segments in Google Ads

Okay, your GA4 data is solid. Now it’s time to build the actual audiences inside Google Ads, and this involves more than just importing the default GA4 audiences. You have to sharpen them with other signals.

2.1. Link Google Analytics 4 to Google Ads

First, a quick sanity check: make sure the accounts are actually talking to each other. In Google Ads, go to Tools and Settings (the wrench) > Setup > Linked Accounts. Find “Google Analytics (GA4),” click Details, and follow the steps to connect your property. This connection has to be in place for your predictive audiences to flow through.

2.2. Create Custom Combination Audiences

While the GA4 predictive audiences are a great starting point, the real power comes from mixing them with other data inside Google Ads. Go to Tools and Settings > Shared Library > Audience Manager. Hit the blue plus button and choose Custom combination. This is where the magic happens. For example, you could combine:

  1. Your “Likely 7-day purchasers” audience from GA4.
  2. A customer list of past high-value buyers who haven’t bought anything in 90 days (you upload this via Customer list).
  3. A website visitor list of people who looked at your “Premium Services” category page but bailed without converting.

You can use “AND” or “OR” logic to build your segment. Just be careful not to make it so narrow that nobody qualifies. You’re shooting for a segment size of at least 1,000 active users for the algorithm to have anything to work with. This layering technique is what lets the AI find patterns you would never spot on your own.

2.3. Implement Dynamic Remarketing Audiences

If you’re in e-commerce, dynamic remarketing is non-negotiable. In Google Ads, under Audience Manager, head to Audience sources. Confirm your Google Ads tag is pulling in dynamic parameters like ecomm_prodid and ecomm_pagetype. Then you can build audiences based on what people actually did, like an audience for “Users who viewed a product in the ‘Electronics’ category but didn’t buy within 3 days.” This is a goldmine. The AI uses these signals to figure out exactly which product to dangle in front of them in a dynamic ad.

Step 3: Activating AI-Powered Campaigns with Performance Max

Performance Max is Google’s all-in-one AI campaign type designed to get you conversions across every Google-owned property. It works especially well for retargeting, but only if you give it good audience signals to start with.

3.1. Set Up a New Performance Max Campaign

In Google Ads, it’s just Campaigns > New Campaign. You’ll probably choose Sales or Leads as your goal. Pick Performance Max for the campaign type. As you click through the setup, pay close attention to your conversion goals and make sure your main ones, like “Purchases” or “Contact Form Submissions,” are selected as primaries.

3.2. Configure Asset Groups with Dynamic Creative Optimization (DCO)

Inside your PMax campaign, you build Asset Groups. This is where you dump all your creative ingredients, headlines, descriptions, images, logos, and videos, and the AI acts like a chef, mixing and matching them to create the perfect ad for each user.

  1. Provide Diverse Assets: Don’t be lazy here. Upload a ton of different images (field, square, portrait), videos (you absolutely need at least one 15-second video), and plenty of short and long headlines and descriptions. The more options you give the AI, the better it can optimize.
  2. Use Google Ads Asset Library: Get organized by using the Asset Library (found under Tools and Settings > Shared Library). Putting all your creative here makes it easier to manage across different campaigns and helps the AI learn which assets are winners over time.
  3. Craft Compelling Calls-to-Action (CTAs): Try out different CTAs. “Shop Now” works, but so can “Learn More” or “Get a Quote.” The AI will test them all and figure out what works for which audience segment.

This DCO functionality is what separates modern AI campaigns from old static ads, letting you deliver a message that feels personal because it’s based on what the AI predicts the user actually wants.

3.3. Add Audience Signals to Guide the AI

In each Asset Group, you’ll find the Audience signals section. This is the single most important part of setting up predictive retargeting in PMax.

  1. Add Your Custom Combination Audiences: This is where you plug in those powerful audiences you built back in Step 2.2. You’re telling the AI, “Start with people who look like this.”
  2. Include Your GA4 Predictive Audiences: Add your “Likely 7-day purchasers” and other predictive audiences directly from GA4.
  3. Use Customer Match Lists: Upload your CRM lists of inactive customers you want to win back or recent buyers you want to upsell, and add them as a signal.

It’s important to get this right: these are *signals*, not strict targeting rules. PMax uses your lists to build a profile of your ideal customer, and then its AI goes out and finds more people like them across Google’s entire network, including people who aren’t on your lists but act just like them. I always tell clients to give the AI at least three different, strong audience signals so it has enough data to learn effectively.

Step 4: Monitoring, Optimization, and Iteration

Your campaign is live. Don’t walk away. Setting it up is just the beginning. You have to keep an eye on it to make sure it keeps working.

4.1. Analyze Performance Max “Insights”

Inside your PMax campaign, go to the Insights tab. This is where Google’s AI tells you what it’s learning about which audience segments are performing best, what your top ad assets are, and which search trends are driving conversions. You need to look for patterns. For example, if you see “New customers” are converting like crazy from a specific asset group, it’s time to make more creative aimed at first-time buyers. If “Returning customers” are all over a certain product video, make more videos like it.

4.2. Review “Diagnostics” for Potential Issues

Also in your PMax campaign, click into the Diagnostics tab often. Think of it as the check-engine light for your campaign. It will warn you about problems like weak asset variety, budget caps, or broken conversion tracking. Fixing these issues fast is the key to keeping an AI campaign running at peak performance. A really common alert I see is “Limited by budget,” which is Google’s way of telling you that you’re leaving money on the table because you’ve found a high-value audience but aren’t spending enough to reach all of them.

4.3. Refine Audience Signals and Bid Strategies

This isn’t a set-it-and-forget-it system. You have to use the performance data to keep refining your audience signals. If one of your custom audiences is a dud, try tightening its rules or just swap it out for a new theory. For bidding, PMax will probably be on “Maximize Conversions” or “Maximize Conversion Value,” maybe with a target ROAS or CPA. You need to adjust those targets based on real-world performance. If you’re crushing your target ROAS by 20%, maybe it’s time to raise the target a bit to push the AI toward even more efficient conversions.

When you get it right, AI predictive retargeting completely changes the game. By being careful with your GA4 setup, building smart audiences in Google Ads, and then giving Performance Max clear directions with strong signals and good creative, you can get way better conversion rates and a much healthier return on ad spend. The trick is to trust the AI’s ability to optimize while you focus on feeding it high-quality data and clear goals.

What is the primary difference between traditional retargeting and AI predictive retargeting?

Traditional retargeting hits people based on a simple rule: they visited a page, so they see an ad. It’s dumb. AI predictive retargeting analyzes a user’s entire pattern of behavior to guess what they’ll do next, like who’s about to buy or who’s about to churn. This lets you serve much smarter, more personalized ads to the people who are actually likely to convert.

How often should I update my audience signals in Google Ads?

The AI is always learning, but you should probably go in and review your audience signals every quarter. A quarterly check is a good rhythm. You should also do it any time you have a major product launch, a big marketing push, or see a fundamental shift in how customers are behaving. You just want to make sure the AI’s starting point is always relevant.

Can AI predictive retargeting work for B2B businesses?

Yes, absolutely. The goal just changes. For B2B, you’re not predicting a product purchase. You’re predicting who is most likely to ask for a demo, download your big industry report, or fill out a contact form. The process is the same: track the right events in GA4 (like time on the pricing page or case study views), build audiences based on those signals, and feed them to a Performance Max campaign.

What if my website traffic is low? Is AI predictive retargeting still effective?

It’s tougher, but not useless. AI models love big data sets, so with low traffic, it will take longer for the AI to learn and its predictions won’t be as confident. That said, it can still help you focus your limited budget on the handful of visitors who are most engaged. Your priority should be capturing as much detail as you can with your event tracking, even if the volume isn’t huge.

What are common pitfalls to avoid when setting up AI predictive retargeting?

The biggest mistakes I see are bad data in, bad results out. People mess up their GA4 event tracking or just don’t have enough of it. Another one is being lazy with creative and not giving the DCO enough assets to work with. Others get too specific and create tiny audience segments the AI can’t use. And finally, people launch the campaign and then ignore the “Insights” and “Diagnostics” tabs, essentially flying blind and letting easily fixable problems kill their performance.

Editorial Team

The Editorial Team at Creative Ads Lab is a group of writers and editors who cover marketing, brand, and growth. They research each story and publish under a shared desk byline — experts in their fields, working as one team.