Putting Google Ads AI Max text customization to the test across three account types
Google Ads AI Max includes a text customization feature that generates Responsive Search Ad assets for every ad group automatically. A series of structured tests across three companies found that the feature works well in neglected, long-tail campaigns but underperforms hand-written assets in highly optimized ones, and can actively harm performance when it undermines audience prequalification.
How the tests were set up
The tests focused on the text customization capability within AI Max, which tailors ad assets in each ad group to the keywords in that group. Three business types were selected: ecommerce, B2B lead generation, and B2C lead generation. Each account contributed two campaigns that received heavy optimization attention and two long-tail campaigns that received less.
To qualify, campaigns had to meet several criteria:
- No brand keywords.
- Minimum monthly spend of $20,000.
- At least 100 ad groups.
- Limited use of pinning, since heavily pinned campaigns would not allow fair testing of new assets.
- No URL expansion, so the test isolated asset performance from landing page routing.
Final URL expansion was turned off across all test campaigns to ensure only the asset layer was being evaluated.
Messaging restrictions and asset review
Before launch, each company built messaging restrictions to keep auto-created assets aligned with brand guidelines. The recommended workflow took about an hour or two:
- Use a Gemini prompt to draft initial assets.
- Prompt the system to write overly promotional ads with claims the brand would never approve, surfacing the worst-case output.
- Create messaging restrictions that block those unwanted patterns.
- Re-run the promotional prompts and confirm that the new assets stay within brand guidelines.
During the test, each company reviewed the auto-created assets as they appeared and removed any that drifted off-message. The default asset report filter does not include ads, so the filter had to be changed manually to surface the new assets. Across the ecommerce and B2C lead gen accounts, roughly 19% of auto-created assets were removed during monitoring.
Ecommerce results
The ecommerce account sells more than 100,000 SKUs, and many shoppers return to the site and search again if the landing page does not match their product. At first, AI Max and text customization appeared to deliver strong gains. A closer review showed that AI Max was poaching impressions, clicks, and conversions from other campaigns, and total account revenue actually declined.
The company responded by adding high-performing search terms as keywords to steer ads to the correct ad group and campaign, then layering in more negative keywords and audience exclusions to slow cannibalization before rerunning the tests. The conclusion was that text customization did not match human-managed assets for highly optimized campaigns in this account, but it did help the long-tail campaign that received less day-to-day attention.
B2B lead generation results
Prequalifying the audience is one of the central jobs of B2B RSA assets. Ads need to deter consumer searchers and appeal to business buyers. This company had previously relied on extensive pinning to enforce that qualification. For the test, the pins were removed so Google could optimize freely.
Click-through rates spiked. Conversion rates fell sharply because the auto-created assets started attracting B2C searchers. Messaging restrictions instructed the system to prequalify for B2B audiences, and some individual assets met the criteria, but the ads shown to users did not consistently appeal to B2B buyers.
The other two tests ran for over a month. The B2B test was stopped after three weeks because results had deteriorated badly. The company reverted to pinning and removed the auto-created assets. Within a week, performance returned to pretest levels.
B2C lead generation results
The B2C lead gen company localizes its ads through geographic ad copy and geographic insertion. Its top campaigns had tailored copy for the keywords in nearly every ad group. Its long-tail campaign had a few headline assets per ad group but reused most assets across ad groups, the kind of formulaic copy that is common in lower-priority campaigns and a natural fit for AI assistance.
Auto-created assets did not beat the human-written assets in the top campaigns, but they performed well in the long-tail campaign, exactly the area where the team had less time to optimize.
Where AI Max automated assets actually help
Across all three tests, a clear pattern emerged. For ads where the team spends significant time refining messaging, human-written assets still outperform AI-generated ones. When the asset must do a specific job, such as prequalifying B2B buyers, promoting a specific offer, or running a short-term promotion, the advertiser should keep control of the copy rather than hand it to AI.
Auto-created assets earn their keep in the campaigns that never get enough attention. With solid messaging restrictions and a regular review process, AI can lift performance in those neglected ad groups. AI in ad creation still requires oversight and review, but using it to handle the heavy lifting in areas the team does not have time to optimize, and then spending human time reviewing and tweaking the output, is the strongest use case the tests surfaced.
FAQ
What is text customization in Google Ads AI Max?
Text customization is a feature within AI Max that automatically generates Responsive Search Ad assets tailored to the keywords in every ad group, reducing the need for per-ad-group customization by the PPC team.
Which campaigns benefit most from AI Max auto-created assets?
Long-tail and lower-priority campaigns that receive less day-to-day optimization benefit most. Highly optimized campaigns and assets that need to prequalify a specific audience, such as B2B buyers, tend to perform better with human-written copy.
How were the AI Max text customization tests structured?
Each of three companies (ecommerce, B2B lead gen, B2C lead gen) tested two heavily optimized campaigns and two long-tail campaigns. Campaigns used no brand keywords, spent at least $20,000 per month, had at least 100 ad groups, used limited pinning, and did not use URL expansion, so only asset performance was measured.
This article summarizes reporting from searchengineland.com.