AI Max: our views and recommendations

AI Max is Google's new, one-click upgrade for Search campaigns. It matches ads to searches beyond your keyword list, writes ad copy from your site and chooses landing pages for you. Google pitches it as a way of finding demand your keywords miss – taking your Exact and Phrase keywords and expanding on them using this new match type labelled ‘AI Max’. 

We wanted to know whether the traffic from the new searches AI Max found resulted in conversions, so we ran Google's AI Max experiments against our existing campaigns.

How we tested

Each test used Google Ads' built-in AI Max experiment. Traffic was split between the original campaign (control campaign) and a copy of the campaign with AI Max switched on (trial campaign). Both campaigns ran at the same time against the same auctions.

The trial campaign allowed AI Max to use our Exact and Phrase match keywords as a start point and extrapolate out to identify similar, relevant, long tail searches.   

What happened

AI Max raised clicks significantly in two of the three accounts, but no test showed a significant change in conversion volume. 

• Client 1: no difference. Both campaigns spent the same and drove almost the identical volumes of clicks and conversions.
• Client 2: more traffic, same conversions. AI Max reached 40% more impressions and generated 30% more clicks, a significant uplift. However, conversion volumes remained consistent with the control campaign, meaning each conversion achieved in the AI Max campaign cost 28% more that those seen in the control campaign. The increase in cost for AI Max could be expected and isn’t an immediate red flag as the terms are broader, which could mean a slightly colder audience. In a subsequent test it would be worth weighing up if the slight increase in cost is worth the ROI from the conversions. 
• Client 3: lots more traffic, expensive conversions. AI Max tripled clicks and spent 2.6 times as much as the control campaign. It did produce conversions, where the control produced none.

This top line results data that Google shares on the experiment overview dashboard is only half the story. To gain more insight, we need to dive into the search terms that drove conversions to see if AI Max had found new, relevant searches or if the conversions were coming from the keywords we already had in the Exact and Phrase match keyword groups.

Investigations revealed that many of the conversions in the AI Max campaigns actually came from Exact or Phrase Match keywords already in the campaigns. 

The topline results reported by Google could mislead you into believing AI Max is working effectively when, in fact, the majority of the conversions came from Exact and Phrase match types and budget spent on AI Max terms essentially is wasted. 

For example, although the experiment for client 2 was relatively positive, generating increased traffic and minor fluctuation in CPA for control vs trial campaign, the conversions that came through the trial campaign that were from new, AI Max search terms were 150% more expensive than those that came through Exact and Phrase match terms. This analysis gives a very different picture to the experiment results.

What we learned

• AI Max does little where keywords in a campaign already cover all possible relevant searches. 
• Where it does expand the campaign, the extra traffic is cheaper per click but of poorer quality. Cheaper clicks with lower engagement point to broader, less relevant searches.
• More clicks have not meant more conversions. 
• Spend grows with reach, so budgets can be compromised. 
• Click uplifts show up fast; in-depth insights on conversions take longer to understand. 
• Don’t just take the experiment insight screen as gospel, dive deeper into the data.

We would recommend you Test AI Max, but don't just switch it on. Our three tests give three different answers, so a controlled experiment is the only way to know how it will behave in your account.

Our recommendations

1. Pick campaigns with room to expand.
2. Judge success on conversions and cost per conversion, not clicks. 
3. Check the search terms early on. Lower click-through rates suggest looser matching. Review what AI Max is matching to in the first two weeks and add negatives where needed.
4. Cap the downside. Set a cost-per-conversion threshold before launch and stop the test if it's breached.
5. Narrow the scope of the test if a broad test fails. There could be areas of your account where this tool would be useful but it needs through investigation.

If you would like help deciding how to use AI Max in your paid search campaigns, please get in touch.