Putting Google Ads AI Max automated ad copy to the test: results from three campaigns
A series of controlled tests across ecommerce, B2B lead generation, and B2C lead generation accounts found that Google Ads AI Max text customization can lift performance on long-tail campaigns but underperforms on tightly managed ad groups, and can actively damage results when audience prequalification matters. Roughly 19% of auto-created assets were removed during the review process for brand or offer alignment.
What is AI Max text customization?
AI Max is a Google Ads feature set that generates assets automatically based on the keywords in each ad group. When text customization is turned on, the system rewrites ad copy per ad group rather than relying on the assets a PPC team has uploaded. The capability is designed to reduce the manual work of tailoring ads across hundreds or thousands of ad groups, particularly in large accounts.
Because the system generates assets in the background, it can also produce copy that promotes products, services, or offers the advertiser does not want to advertise. Google offers messaging restrictions as a guardrail, letting advertisers set rules around brand voice and prohibited claims.
How the messaging restrictions were built
The teams running the test used a four-step workflow to build their messaging restrictions, a process that took roughly one to two hours per account:
- Use a prompt in Gemini to draft initial assets.
- Run a second prompt that intentionally produces overly promotional copy and unwanted promises, so the system surfaces claims the advertiser wants to block.
- Translate those unwanted examples into messaging restrictions.
- Re-run the promotional prompt against the new restrictions until every generated asset complies with the company’s guidelines.
Which campaigns were tested
Three account types were chosen: ecommerce, B2B lead generation, and B2C lead generation. Within each account, the test selected campaigns that met a fixed set of criteria: no brand keywords, at least $20,000 in monthly spend, and a minimum of 100 ad groups.
Two campaign profiles were pulled from each account. The first profile was highly optimized campaigns where the team already spent significant time tuning assets. The second profile was somewhat neglected long-tail campaigns that performed in aggregate but received less attention. Campaigns that relied heavily on pinning or that used final URL expansion were excluded, since the goal was to isolate the asset-generation effect.
During the test, the companies monitored auto-created assets as they were generated and removed any that conflicted with brand messaging or product offers. To see AI-generated assets in the asset review screen, the default filter had to be changed; the AI filter is not selected by default. Across the ecommerce and B2C accounts, roughly 19% of auto-created assets were removed before they accumulated many impressions.
Ecommerce results: cannibalization hurt the account
The ecommerce account sells more than 100,000 SKUs, and many of its shoppers are used to searching the site directly when a landing page misses the product they want. On the surface, both AI Max and text customization appeared to deliver strong results in this account.
Deeper analysis showed the gains were illusory. AI Max was pulling impressions, clicks, and conversions away from other campaigns in the same account, and total account revenue declined. To correct the cannibalization, the team added high-value search terms as keywords so Google would prioritize the correct ad group, then layered in more negative keywords and audience exclusions. After rerunning the test with those controls, the conclusion was that AI text customization underperformed human-managed assets on highly optimized ecommerce campaigns but provided a clear lift on the long-tail campaign.
B2B lead generation: prequalification broke down
For B2B advertisers, RSA assets usually need to actively repel B2C searchers and signal to B2B buyers that the offer is relevant. The test account had previously used pinning across its assets to enforce that qualification. For the test, the team removed the pins to let Google’s optimization run freely.
The result was a sharp split. Click-through rates climbed because the assets became more broadly appealing. Conversion rates fell significantly because the ads began attracting B2C traffic the company did not want. Even with messaging restrictions instructing the system to prequalify for a B2B audience, the ads that ultimately served did not consistently speak to B2B buyers.
The other test campaigns ran for more than a month. The B2B test was stopped after three weeks because performance had degraded so far that the company reverted to pinning its assets and removing the auto-created assets. Within a week of that reversion, B2B results returned to their pretest baseline.
B2C lead generation: the long-tail campaign won
The B2C lead generation account localizes its ads through geographic ad copy and geographic insertion. Its optimized campaigns had copy tailored to the keywords in nearly every ad group. Its long-tail campaign had a small set of keyword-tailored headlines per ad group, with most assets reused across groups, a setup that leaves a lot of room for improvement.
Auto-created assets delivered in the low-priority campaign. They did not beat the human-tuned assets in the top campaigns, where copy had been iterated on for a long time, but the AI output closed most of the gap on the long-tail campaign and meaningfully improved performance there.
Where AI Max automated assets work best
Human-created assets still outperformed AI-generated assets in the campaigns where the team had already spent significant time iterating on messaging. When the asset needs to do specific work, such as prequalifying a B2B audience, advertising a specific offer, or running a short-term promotion, handing control to AI tended to backfire.
AI Max text customization is most useful where there is no time to fully optimize creatives. With solid messaging restrictions and regular asset review, the feature can lift overall account performance, especially on long-tail campaigns that have been running on reused assets. The system is not a turn-on-and-forget setting. It needs ongoing review and adjustment, and the best results come from letting AI carry the workload in neglected areas while a human reviews and refines what the system produces.
FAQ
What is AI Max text customization in Google Ads?
AI Max is a Google Ads feature that automatically generates ad assets tailored to the keywords in each ad group. Text customization is the part of AI Max that rewrites copy per ad group, reducing the manual work of building variants across large accounts.
How were the AI Max tests set up?
The tests covered ecommerce, B2B lead generation, and B2C lead generation accounts. Each account contributed campaigns that did not use brand keywords, spent at least $20,000 per month, and had at least 100 ad groups. The test used both highly optimized campaigns and somewhat neglected long-tail campaigns, and excluded campaigns that relied heavily on pinning or used final URL expansion.
What were the main results from the AI Max ad copy tests?
AI Max text customization underperformed human-managed assets on highly optimized ecommerce and B2B campaigns, and caused cannibalization in the ecommerce account until keyword and audience controls were added. In B2B, removing pins caused click-through rates to rise while conversion rates fell, and the account reverted to pinning after three weeks. On long-tail campaigns, especially in the B2C account, auto-created assets delivered a clear lift. Roughly 19% of auto-created assets were removed during review for not aligning with brand or offers.
Related coverage
This article summarizes reporting from searchengineland.com.