Google’s AI Ads: What 2026 Means for ROAS

Let’s be real, AI has totally upended digital advertising, and Google’s constant algorithm changes don’t make it any easier. If you want to get consistent results, you have to completely rethink your strategy. So how do you keep campaigns afloat when Google keeps moving the goalposts?

Campaign Teardown: “Urban Explorer” Footwear Launch

Back in mid-2025, we launched “Urban Explorer,” a new line of sustainable outdoor footwear. The goal was to drive direct-to-consumer sales, hitting a younger, environmentally conscious audience. Seeing how much Google was leaning on its AI for everything, we bet a huge chunk of our budget on Performance Max campaigns, with some supporting YouTube and Display ads on the side.

Strategy and Objectives

Our whole strategy was to give Google’s AI exactly what it wants: lots of different inputs and really clear conversion signals. We were shooting for a Return on Ad Spend (ROAS) of 3.5x and a Cost Per Lead (CPL) of under $15 for newsletter sign-ups, which for this kind of product is a great sign of intent. We gave ourselves an eight-week window and a $120,000 budget to make it happen. We knew from the start that PMax needs broad audience signals and good creative to work, so we just focused on feeding the machine as much good stuff as we could.

We defined our target audience by their interests in outdoor life, sustainability, city fashion, and tech. Our bet was that one big, AI-orchestrated campaign would work better than a bunch of separate, siloed ones. That meant we had to consciously stop ourselves from creating super-detailed audience segments in each campaign type and just trust that Performance Max could find the right people across all of Google’s properties.

Creative Approach: Fueling the AI Engine

With AI campaigns, you need a ton of different creative. We went all in and built a full asset library:

• Video Assets: We made ten different videos, from quick 15-second spots to 60-second stories, showing the shoes in the city and out in nature. We had product demos, lifestyle shots, and even some testimonials from micro-influencers.
• Image Assets: We had over thirty high-res photos. This included clean product shots, people wearing the shoes, infographics about the sustainable materials, and some user-generated content we’d collected. We made sure to have a mix of square, field, and portrait formats so they’d look good everywhere.
• Headlines and Descriptions: We wrote a bank of 15 unique headlines and 5 long descriptions. Each had a different tone and call-to-action so the AI could test countless combinations. We also made sure to include relevant keywords (without stuffing them) to give Google’s algorithm context.
• Audience Signals: This was huge. We uploaded our own customer lists of past buyers and newsletter subscribers. Then we built custom segments based on people visiting competitor websites or searching for related terms. It gave the AI a solid head start on who to look for.

We deliberately didn’t try to guess which creative would be the winner. The whole point was to let the AI figure that out by serving different combinations and seeing what people actually responded to. Honestly, that takes a leap of faith that a lot of advertisers aren’t comfortable with because they’re used to being in complete manual control.

Targeting and Bid Strategy

We put most of our chips on Performance Max because it hits all of Google’s channels, Search, Display, Discover, Gmail, Maps, and YouTube, from one campaign. We set the bid strategy to “Maximize Conversion Value” and plugged in a target ROAS (tROAS) of 350%. For our other campaigns, like branded search and a basic display remarketing campaign, we used “Maximize Conversions” with a target CPA (tCPA) to mop up any high-intent users efficiently.

The audience signals we fed into PMax were absolutely critical. By integrating our first-party data with Customer Match, we gave the AI a perfect picture of our existing customers to find lookalikes. We also layered on custom segments targeting people who had recently searched Google for things like “sustainable running shoes” or “eco-friendly sneakers.” This mix of specific signals and broad reach gave the AI the ammo it needed to go find new customers.

Campaign Performance: What Worked and What Didn’t

Across the eight weeks, the campaign pulled in 18.5 million impressions and got 125,000 clicks, which works out to a Click-Through Rate (CTR) of 0.67%. We spent a total of $118,500.

Metrics Breakdown:

What Worked:

1. Performance Max’s Broad Reach: The campaign found niche audiences in places we never would have targeted manually. For example, YouTube in-stream ads were converting like crazy with some very specific interest groups we hadn’t even considered.
2. Diverse Creative Assets: Having so many different videos and images was key. It let the AI constantly test and learn, showing people the right ad at the right time. The system quickly figured out that our longer, story-based videos were great for building awareness on YouTube, while simple, product-heavy images worked best for closing sales on the Display network.
3. First-Party Data Signals: Handing over our customer lists made a huge difference in the quality of audience matching. A recent eMarketer report said brands using their own data get twice the performance, and our experience definitely backed that up.

What Didn’t Work as Expected:

1. Initial Learning Phase Length: The Performance Max campaign took almost two full weeks to get its footing and stabilize, which felt like an eternity. During that time, the ROAS was all over the place, even dropping to 1.8x on some days. You really have to budget for that initial ramp-up and just be patient.
2. Attribution Complexity: With sales happening across so many Google channels, figuring out the true customer journey was tough. Google Ads showed a great ROAS, but trying to pin a specific sale on a specific headline or image inside Performance Max’s “Asset Group” reports was nearly impossible. This is just the reality of working with these black-box AI systems.
3. Budget Pacing Fluctuations: A few times, the campaign burned through its daily budget by noon, meaning we missed out on evening traffic. The system usually fixes itself over time, but it showed we needed to keep a closer eye on the daily spend, especially on high-volume days.

Optimization Steps Taken

We didn’t just sit back and let it run. Based on what we saw, we made a few key moves:

1. Asset Group Refinement: After three weeks, we dove into the asset group reports in PMax. We found the worst-performing headlines and descriptions and paused them, then wrote new versions based on our top performers. We also added two new videos that leaned heavily into the shoe’s environmental story, since the early data suggested that message was really hitting home.
2. Conversion Value Adjustments: We were getting plenty of newsletter sign-ups (hitting our CPL target), but a direct purchase was obviously worth a lot more. We went into our conversion tracking and assigned a higher value to purchases versus sign-ups. This gave the AI a much clearer instruction to go after revenue, not just leads, a small tweak that has a big impact on AI bidding.
3. Negative Keywords for Brand Safety: Even though PMax is meant to be broad, we saw our ads popping up for some weird, irrelevant search terms. We added a small account-level negative keyword list to block those queries. It’s a reminder that even in an AI-first world, some manual cleanup is still required.
4. Budget Reallocation: We could see that the Performance Max campaign was consistently outperforming our standard Display remarketing campaign on ROAS. So, in the last three weeks, we moved 15% of the Display budget over to PMax to pour gas on the fire. This kind of quick budget shifting is exactly what you need to do when an AI finds a winning formula.
5. Audience Signal Updates: Halfway through, we refreshed our custom segments with people who had been reading recent articles about sustainable fashion. We also uploaded new customer lists from some recent events we’d done. You have to keep the signals fresh.

The biggest lesson was that you have to keep feeding the AI new, high-quality data and creative. It’s not a “set it and forget it” tool. The more information and options you give it, the more paths to conversion it can find and test. The “Insights” tab in Google Ads became our best friend for figuring out what was working so we could make these kinds of adjustments.

And another thing: don’t be afraid to push back on the AI’s first instincts. It’s optimizing for the goal you give it, but you still need a human to make sure it’s not doing something dumb for the brand or misinterpreting the data. For instance, we saw our conversion rate dip on some display placements. When we looked closer, they were mostly mobile game apps with a lot of accidental clicks. The AI was chasing clicks, but the click quality was terrible, so we had to go in and manually add those placements to our exclusion list.

The Future of AI in Advertising: Working through Google’s Parameters

The “Urban Explorer” campaign really drove home how much our jobs have changed. Google is clearly designing its platform to give the AI more and more control, which means our job is now to give it clear goals, a ton of creative, and perfect conversion tracking instead of fiddling with keywords and placements. This puts all the pressure on getting your conversion tracking right and building your first-party data lists.

This push to automation is only going to get faster. As advertisers, we have to get really good at understanding the signals we’re sending these systems. That means being obsessive about setting up conversion actions correctly, being smart about how we segment customer data for audience signals, and constantly testing new creative. Can you ever fully understand the AI’s “black box”? No, but learning to read the tea leaves from performance metrics and insights reports is the new core skill for a digital marketer.

On top of all this, privacy rules are going to keep changing how AI can use data. With third-party cookies going away, your own first-party data becomes your most valuable asset for targeting. The brands that are investing in their own data right now are going to have a massive head start in training Google’s AI to find their best customers.

Our experience with this campaign suggests success isn’t about finding a secret “hack.” It’s about building a working relationship with the AI. If you give it the right fuel, data, assets, and clear goals, it will get the job done. If you don’t, you’re going to get left in the dust by competitors who do.

Working within Google’s AI-driven ad platform means you have to commit to constant learning and adapting, with a laser focus on data quality, creative, and clear business goals. Figuring out how AI can show you your true marketing ROI is the name of the game now, which we explore in our article 2026: Geo-Holdout & AI Reveal True Marketing ROI.