How taming AI Max netted us 12% more conversions

As early as September, Google Ads will start replacing many of its features with its new automation suite, AI Max.

Google calls it a “comprehensive suite of targeting and creative features,” which is a fancy way of saying it helps you make and serve ads.

If you try to dig a little deeper, you’ll find marketers split between those who think you’ll never have to work again, and those who think ads will never work again. So I badgered a lead marketer on our Paid Ads team to see just where the truth lies.

“We were actually one of the first advertisers to test it,” says Jonathan Tablada. “From our early testing [we saw] a 12% increase in conversions. But it’s definitely not autopilot.”

Below, he shares what we learned from 40+ campaign trials, including what worked, what didn’t work, and what you need to make the max of AI Max.

Adsmaxxing

“How people are asking questions with AI is very different from how they asked questions in the past,” Jonathan says.

Searches are getting longer and way more specific, and that’s before you even layer in how answer engines personalize responses according to user context. Which means the old ways of exact match keyword targeting may not get you in front of the right people anymore.

“AI Max is able to go after very broad terms, but it can also pick up on clues from the user to understand who’s a good fit [for your ad] or not.”

But despite all the hype lately about AI doing everything for you, it’s not a checkbox solution.

“We don’t just go through every campaign and turn on AI Max [indiscriminately]. What we’ve done is gone through rigorous testing using Google Ads Experiments.”

Set it but don’t forget it.

That rigorous testing isn’t a one-off experiment to decide whether to broadly use AI Max or not. Jonathan and his team routinely test numerous sample campaigns to see where AI Max works and where manual targeting makes more sense.

There are a million guides to running ads experiments, so rather than spending time on the nuts and bolts, here’s how they create experiments that quickly show whether AI Max is right for a given campaign.

1. Use Google Ads’ native experiment tool.

What you don’t want to do is simply turn on AI Max and see what happens. A so-called “lookback test” can torpedo your conversion rate without ever actually showing you where you went wrong.

If a campaign fails, would you know if it was due to your changes, an external traffic shift, or simply bad timing?

“The Google Ads Experiment tool forces the traffic to split. We set it to a 50/50 split, and it’s cookie-based, so it’s truly a separate set of users for each variant.”

So the first step is to create an A/B test for key campaigns where the only variable is whether AI Max is turned on.

2. Aggressively garden your negative keywords.

Like anything based on statistical analysis, there’s going to be a margin of error. It’s important to not have a “set-it-and-forget-it” mindset.

“The broader you get, the more you want to keep a close eye on your search term report. Otherwise, you could be spending on areas that are very inefficient.”

For example, Jonathan shares a story about finding ad dollars being spent to show up for the term “HubSpot customer support.”

Someone searching for our customer support already knows we exist, so that’s a waste of budget. So that term gets plunked onto our negative keywords list — the list of terms you explicitly tell Google not to show an ad for.

And again, this is not a one-off effort. You should be continually refining your negative keywords list.

“Over time, as you add negatives, AI Max should improve the campaign’s targeting.”

3. Keep an eye on landing page choice.

“AI Max will read your landing page, and from what I’ve seen, there’s no way to opt out of this. So it’s not just looking at keywords, it’s looking at your landing page and using that to target users,” Jonathan explains.

“If your landing page is out of date, if you don’t have the right landing page, or the page copy isn’t really relevant to what users are searching for, that’s really critical.”

That means it’s more important than ever to choose the right landing page and to make sure it represents a logical path from ad to conversion.

“We see better performance by having dedicated paid ad landing pages, even outside of AI Max.”

Diving In

Running well-crafted experiments will get you most of the way to success, but Jonathan shared a handful of tips for using AI Max like a pro.

1. Don’t jump into the deep end.

If you’re just getting into Google Ads, don’t jump straight to AI Max.

“It’s like going from a kiddie pool to the ocean. It’s going to be overwhelming. You’ll see a massive amount of irrelevant search terms, and it’ll be really daunting to have to add them all as negatives.”

Instead, Jonathan recommends graduating slowly from exact match, then phrase match, and finally to AI Max.

“[When you switch] from exact match to phrase match, you’ll get some irrelevant terms. Add them as negatives, get to a good spot, and by the time you get to AI Max, you’ve already blocked a lot of the irrelevant stuff.”

2. Audit smartly, not exhaustively.

Even if you’re a veteran of Google Ads, Jonathan says it’s important to do your prep work first.

“Before you turn on AI Max, I would look back at the last month or two and see what terms you can add as negatives right away.”

But this is definitely an area where it pays to work smarter, not harder.

“Instead of starting at the top and scrolling down — which can be mind-numbing — think about things like ‘refund,’ ‘support,’ and ‘customer service’ to find irrelevant terms fast.”

You can also use filters to suss out terms that provide zero conversions or have excessively high cost-per-acquisition.

3. Try one setting at a time.

The first thing Jonathan’s team does when they turn on AI Max is to turn off the “final URL expansion” and “ad text customization” settings.

“Final URL expansion will essentially crawl your whole website and serve whatever landing page the algorithm thinks is best. Ad text customization means the AI will actually write copy for you based on the user’s query.”

Jonathan is quick to say these aren’t bad features; it’s just very easy to get overwhelmed.

“If you haven’t used AI Max before, testing AI Max on its own is enough to get a read. You can always layer in those other features as tests at another time.”

4. Give it a chance to work its magic.

You’re not going to magically see whether AI Max is working in 24 hours. But just how long should you let it work its mojo?

“On average, eight to 12 weeks,” Jonathan advises. “If you let something go too long, it’s a lot of noise. But if you can’t get a read within eight to 12 weeks, it’s probably not statistically significant.”

Of course, this does depend on how much data you have to work with.

“For really high-volume campaigns, I’ve done tests in just two to three weeks.”

5. Measure down-funnel performance, too.

“It’s not just about running AI Max to get more conversions. It’s about whether AI Max actually boosts revenue in some meaningful way.”

So, perhaps the most important bit of advice of all: In addition to watching conversions and cost-per-acquisition, you also want to pay attention to down-funnel performance. That might be sales, downloads, views, bookings — whatever your ultimate business goal is.

As with any kind of paid ad, it’s possible to find keywords that bring in vast throngs of people that… simply aren’t qualified prospects.

“[With some keywords], we saw tons of conversions, but then it was only two people using the product and zero paying for it.” To prevent that, “we look at different stages: Are they actually using HubSpot? Did they become a customer within 90 days? How much was that revenue worth?”