Google AI Max is A TRAP?! Plus: New Data on ROAS, Demand Gen & DSA Sunset

Episode Transcript

00:00:03 Mike:
Welcome to another episode of Growing E-Commerce. We are your hosts, Mike Ryan and Chris. Chris, thank you for joining us today. What do we have today? Let’s get right into it. We have some data with us today, interesting data.

00:00:24 Chris:
I would even say that every piece of data you bring up is interesting. But this one is maybe even a bit more special because the market is really a big unknown.

00:00:36 Mike:
It is a big unknown. The secret reality is that most data is boring, and it is hard work to find the interesting part inside of it.

00:00:46 Chris:
Really. And unless you can connect it to the real challenges clients have. This is the perfect segue, because as far as I know, the market is not fully sold on AI Max for search, for instance.

00:01:02 Mike:
Exactly. That is one of two technologies I want to talk about today: AI Max for search. As a recap, this is a new DSA-like technology introduced by Google. Dynamic Search Ads (DSA) is a way of matching landing pages and search terms and generating headlines. This technology lived inside of search as dynamic search ads and it also lived inside of PMAX. Google decided that rather than bringing search to PMAX as they have done with other campaign types, they want to take PMAX into search. The confusing part for me is that it already kind of was there in the form of existing search ads. So we now have a situation where there are three very similar technologies: dynamic search ads, AI Max, and the search part of PMAX, all doing very similar activities with similar goals.

00:02:03 Chris:
Let’s be clear about DSAs. This has been a great technology that has been here for years. Before we jump into the numbers and what you found, what is your take on dynamic search ads as a standalone campaign type? Do you see a bright future for it, or do you see it being sunsetted because it is embedded in PMAX and AI Max for search?

00:02:34 Mike:
It is dead as a doornail, unfortunately.

00:02:38 Chris:
I didn’t want to be that bold, but…

00:02:40 Mike:
I will go ahead and be that bold.

00:02:45 Chris:
I want to put you on the spot now. How much time is left?

00:02:47 Mike:
By this time next year, it will be gone already. That is my prediction. I already saw a redundancy between DSA and PMAX when PMAX was launched, and they haven’t touched it since then. About 20% of search advertising costs, at least in e-commerce, still seem to be running through there. That is a lot of money, and they need to be careful with how they do these things. I think they were evaluating that strategy through focus groups, deciding whether to migrate search to PMAX. But it is now at a point where there are three highly equivalent technologies. Two of them are in Google’s so-called “power pack,” the future of search. These are their smiled-upon technologies that have received the blessing. It is clear that dynamic search is left over. I have seen Google migrations and sunsets come and go. My prediction is that in early 2026, they are going to announce that DSA is being upgraded to AI Max, and they will want that resolved by the end of September so everything is set up for the holidays. We see this pattern time and again with expanded text ads and shopping migrating to PMAX. This is the recipe.

00:04:18 Chris:
Ladies and gents, this is big news because, as you stated, roughly 20% is still running through DSAs. That is the level we are talking about. It is huge. DSAs, whether in the accounts we run for clients or the ones we have analyzed throughout the last couple of years, is a great technology. This campaign type will be sunset. The technology lives on embedded in PMAX and AI Max for search. The most important thing is that Google will call it an “upgrade.” But is it a no-brainer?

00:05:00 Mike:
It is not a migration, it is an upgrade. We saw this time and again.

00:05:04 Chris:
I love it.

00:05:06 Mike:
Me too.

00:05:07 Chris:
What is your take on it?

00:05:08 Mike:
That is not official information, but I am very sure this is what will happen. It has been clear for a while. Let’s talk about a couple of numbers first. Right now, we are seeing that about 12% of accounts have some level of AI Max activity.

00:05:29 Chris:
So it is just activation. It doesn’t correlate with spend share. Is the 12% related to the spend already running?

00:05:37 Mike:
No, not yet. The spend is at small levels now because it is not activated in all campaigns. Even when it is activated, it is basically one match type among four or more. If we look at the search spend at an account level, it is typically at less than one percent. But if you look within campaigns where it is activated, then it jumps up to about five percent of cost. Still low. I can’t assume this is satisfying for Google.

00:06:12 Chris:
I disagree. It is early. In their earnings on our last episode, a statement from their Chief Business Officer was that it is growing very fast. Of course they say that, but it just went to general availability recently after being in beta. I think that 12% of accounts testing it shows the foothold is there. Then this all won’t matter because I think there will be a migration. We remember this with PMAX; we saw a gradual rise in cost share, then it went vertical due to the migration before finding its level. I think that will happen with AI Max too.

00:07:09 Mike:
May I correct you? It is not a migration, it is an upgrade.

00:07:14 Chris:
Upgrade, I am sorry. Thank you. I was pretty bold with regards to Google in the last episode. I think they have a great set of products and they are listening to the market way more than they have been for the last 10 years. What I see from talking to clients and following influencers—including you—is that the reaction to AI Max for search is not utmost positive. New campaign types often get criticized initially by online marketers. But if you cut out the noise, Mike, what are the numbers telling you, even though it is early?

00:08:08 Mike:
I agree that Google has been much more amenable to advertisers, feeling more like a partnership again. The first reaction to AI Max was pretty positive. It offers search term reporting from the get-go, which is not to be taken for granted.

00:08:32 Chris:
This is true.

00:08:33 Mike:
There are also good levels of customization. They split it out into three main features that you can mix and match. From a product design standpoint, I really like it. From a performance standpoint, however, it hasn’t been so satisfying. We talked in the past about it appearing on competitor terms, based on sample sizes of one. It is a bit opposite to PMAX, which was satisfying from a performance standpoint but lacked features. AI Max right now has good features, but the performance isn’t quite there.

00:09:15 Chris:
Before we jump into the performance numbers, which I suspect won’t be the most positive, I find it interesting that Google launched AI Max for search in a way where you instantly have a clear understanding if it is performing. With the search term report, I can see if there is an increment. Google opened themselves to market feedback which is numbers-driven. Mike, what do you have?

00:10:14 Mike:
I looked at 271 campaigns that have AI Max activated. It is a decent first sample.

00:10:24 Chris:
Yeah.

00:10:25 Mike:
When we first started writing reports about PMAX, we had about 3,000, and we will get there with AI Max eventually. But there are a lot of impressions, clicks, and conversions tied up here. I am comparing it against broad match, exact match, and phrase match, though it is fairest to compare it to broad match. Broad match is supposed to help expand coverage and provide incrementality. It should find conversions you were missing before. Google says they can deliver 14% more conversions at a similar efficiency target, but they mention that is for non-retail campaigns. We are in e-commerce and retail, and it doesn’t look that nice.

00:11:43 Chris:
To be precise, Google is saying the incremental return on ad spend (ROAS) of these incremental conversions should be on the same level as your core existing search setup. That is a hell of a statement. It would be a no-brainer to activate.

00:12:04 Mike:
That is the dream. Who wouldn’t accept 14% more conversions at the same risk and efficiency? Usually, there is a trade-off where chasing the next conversions leads to diminishing returns. Google claims that won’t happen with this technology in non-retail campaigns, which is a bold claim.

00:12:28 Chris:
It is a bold claim. Where are these 14% additional conversions coming from if everyone adopts AI Max?

00:12:43 Mike:
Yeah.

00:12:44 Chris:
Is there suddenly 14% more demand? If you were the only advertiser with this power, fine. But it is a commoditized technology.

00:12:54 Mike:
Anyone can use it.

00:12:59 Chris:
I don’t understand the narrative. These 14% might be there if you are an early adopter.

00:13:20 Mike:
They could be there if your campaign is not well-built and you are missing low-hanging fruit. For well-set-up campaigns, this should be interesting.

00:13:42 Chris:
Okay, but this is Google’s claim for non-retail.

00:13:47 Mike:
Exactly. We are retail, and they don’t make a performance claim for us, but I will show you what it looks like. Comparing AI Max and broad match: the click-through rate for broad match is 2.3%, and AI Max is 2%. Conversion rate: broad match is 5.6%, and AI Max is 4.6%—a full percentage point less. This is concerning, although it is still a strong conversion rate.

00:14:40 Mike:
CPC: broad match is 1.57 euros, AI Max is 1.58 euros. Very comparable. However, the cost per conversion (CPA) for broad match is about 28 euros, while AI Max is 34.21 euros. That is costing several euros more per conversion—about 20 percent more.

00:15:23 Chris:
That is huge.

00:15:24 Mike:
Average order value: broad match is 112, and AI Max is 89. This is where it breaks down. You are paying more per order for less valuable orders. This results in a ROAS difference where broad match is at 4.0 and AI Max is at 2.6. That is where it falls apart.

00:15:45 Chris:
As an e-commerce director, a drop from 4.0 to 2.6 is huge. The most important question for me is whether AI Max for search conversions are really incremental or if they are eating into what I already have. The worst case is this new campaign type eating into my existing search setup at a worse ROAS. Can we evaluate this?

00:16:30 Mike:
It is possible, but it would take mapping search terms to see if they are being covered by other match types. That is a big lift because search term data is vast.

00:16:49 Chris:
If anyone can do it, it is you and some data engineers.

00:16:54 Mike:
I can, but my anecdotal observation and gut feeling is that it is not purely incremental.

00:17:16 Chris:
Every online retailer is forced to grow. You can go up-funnel or activate a new campaign type like AI Max for search in the lower funnel. If it is really incremental, maybe you can live with a worse ROAS because that is what new clients cost.

00:18:04 Mike:
Your new customer ROAS will never be the same as your existing customer ROAS.

00:18:08 Chris:
Exactly. If it is incremental, a worse ROAS is acceptable. The question remains: is it incremental?

00:18:22 Mike:
This can be solved through testing, and Google is getting into the marketing mix model game with Meridian because they want to own those conversations.

00:18:41 Chris:
If someone could do it, it is smec. But you don’t think AI Max for search is purely incremental.

00:18:58 Mike:
I don’t think so. I don’t find the situation damning, and I wouldn’t tell people to stay away from AI Max, but you must go in with eyes open and control it. You can handle competitor terms with brand exclusions and use negative keywords to force it to work smarter. We are in an early, unoptimized state. Another factor is redundancy. When AI Max is activated, I found that roughly one in five times, they are also using dynamic search ads. Between a quarter and a third of the time, they are also using PMAX.

00:20:12 Chris:
So we are talking about the activation of DSAs in three different campaign types.

00:20:21 Mike:
About one in two of these advertisers are using AI Max, DSA, and PMAX all at the same time. I can’t say for sure if that is bad for performance, but it isn’t good. There is a risk of competing against yourself and splitting data between campaigns needlessly.

00:20:55 Chris:
It is not hygienic. The biggest issue is not being sure if you are competing against yourself.

00:21:21 Mike:
Only seven percent of advertisers exclusively use AI Max, which is the purest form. You can use two of these technologies side-by-side if done well, but using all three is unnecessary.

00:21:52 Chris:
You can’t convince me that is a good idea. AI Max for search is here to stay, so advice for listeners would be to test it because the product design gives you leverage to control it.

00:22:21 Mike:
Exactly. You can optimize it. Google will improve it, and in 18 months, we could be having a totally different conversation.

00:22:48 Chris:
Let’s hope for a journey like PMAX. We eventually became friends with PMAX. What else do you have for us today?

00:23:07 Mike:
I have liked Demand Gen since day one. They added more placements to what was previously Discover campaigns. They have increased transparency and controls, allowing you to see and target specific channels. On the negative side, they have added “fluff inventory,” but you can opt out. Demand Gen needed to scale, and Google provided that by adding YouTube. I think Demand Gen is a YouTube buying engine and has been part of YouTube’s growth.

00:24:19 Mike:
I looked at the “economics of Demand Gen,” analyzing the channels—YouTube, Discover, Gmail, Display—and ad types. Demand Gen isn’t purely CPC or CPM; it uses multiple billing models. Interestingly, more than 50% of the cost share is YouTube video ads. YouTube is the scaling engine of Demand Gen, and Demand Gen montezies YouTube. It is a win-win.

00:26:07 Chris:
We spoke with the European product leader of Demand Gen and YouTube. Our take was to let Demand Gen generate new demand but not force it to be a performance channel. However, the product lead was convinced Demand Gen can be performance-oriented. Google is trying to build it as a mid-to-lower-upper funnel machine with a clear performance purpose.

00:27:22 Mike:
I see that, but it is confusing. Google positions it against Meta ads. They want a mid-funnel narrative that supports search and brand lift, but that is unsatisfying to someone spending on Meta. Meta generates demand but also performs. Google is trying to square this circle, but these are the same channels running with PMAX. If Demand Gen were a performance channel, it would be called PMAX.

00:28:53 Chris:
If a client isn’t maxed out on PMAX impression share, why go to Demand Gen when you can get better ROAS in PMAX? Demand Gen should be seen as something different—great for a brand and generating demand, but without a direct performance impact. The more they try to push the performance narrative, the trickier the decision framework becomes for retailers.

00:30:12 Mike:
When PMAX rolled out, the market thought it was a full-funnel campaign because of display, but it is a bottom-funnel shopping campaign. To build a true full-funnel campaign, you would almost need to build PMAX and Demand Gen together.

00:31:02 Chris:
PMAX is good at maxing out an existing client pool. But for growth beyond the fiscal year, you need new potential. Demand Gen is great for that, even without direct performance impact.

00:31:36 Mike:
Demand Gen and PMAX make a good pair if you let them each play their role. Google needs to focus on showing how Demand Gen makes PMAX better. Meridian might help with that.

00:32:08 Chris:
It comes down to expectation management. Demand Gen is not a direct one-month performance-driving campaign. Google is working to make the mid- and long-term effects more obvious.

00:32:50 Mike:
I think they picked the wrong fight by comparing themselves to social feeds like Meta. YouTube and Discover are similar but different. I like the technology, but it is in an awkward phase of product strategy.

00:33:28 Chris:
It is a great technology. Online retailers focused on performance must think about where the next growth pocket is. Demand Gen is good at that. What a wrap.

00:34:05 Mike:
That was a good note to end on.

00:34:08 Chris:
I’m proud of myself. The outro will be the highlight again.

00:34:14 Mike:
Let’s go for it. Thank you for tuning in to another episode of Growing E-Commerce, brought to you by Smarter Ecommerce, also known as smec. You can learn more at smarter-ecommerce.com. It’s been our pleasure. We’ll see you next time.