A team of PPC specialists ran side-by-side experiments with Google Ads AI Max text customization across three companies: an ecommerce retailer with more than 100,000 SKUs, a B2B lead generation business, and a B2C lead generation business. The takeaway for anyone auditing paid search accounts is that auto-created responsive search ad assets perform unevenly. They lift neglected campaigns and quietly damage the ones a team has already tuned by hand.
What the experiment actually measured
Each account had AI Max switched on with text customization enabled. To keep generated copy in bounds, the team used a short Gemini workflow: produce a baseline of assets, deliberately request off-brand or over-promotional variations, write restrictions that block those patterns, then keep prompting until every output aligns with the company voice. The full setup typically ran an hour or two per account.
From every account the team pulled four campaigns: two that the in-house team actively managed, and two long-tail campaigns that received less attention. The filters were strict. Campaigns had to exclude brand keywords, spend at least $20,000 a month, and contain at least 100 ad groups. Campaigns that depended on pinning were excluded because pinning overrides auto-created assets. URL expansion was also disabled so the test isolated copy alone.
How to make AI-generated assets visible during an audit
One operational detail showed up in all three accounts and is worth checking first. The default asset review filter in Google Ads does not display AI-generated assets. Reviewers need to change the filter to include the "Auto-created" option before anything the system produced becomes visible. Without that step, an audit can conclude an account is fine when the system is already running copy nobody has reviewed.
Across the ecommerce and B2C accounts, close to 19% of the auto-created assets were removed during the test because they drifted away from approved offers or brand voice. That ratio is a useful benchmark. If removals during your own review are running well below 19%, the system may be quietly serving copy no one has ever looked at.
Ecommerce: where the audit gets harder
The ecommerce retailer sells more than 100,000 SKUs and sees many shoppers return to search again when the landing page does not match intent. At first the AI Max results looked positive. Deeper analysis told a different story. AI Max was taking impressions, clicks, and conversions from the account’s other campaigns, and total account revenue fell during the test.
The fix the team applied is itself a checklist item. They added high-performing search terms as new keywords to push the system toward the right ad groups, layered in more negative keywords, added audience exclusions, and reran the test. After that round, auto-created assets still trailed human-managed assets on the optimized campaigns but improved performance on the long-tail campaign, where each ad group had received less individual attention.
What to check on an ecommerce account
- Compare assisted and last-click conversions for the campaigns running auto-created assets against the same period before AI Max was turned on.
- Look for impression and click overlap between AI Max campaigns and other campaigns in the same account, since internal cannibalization was the main source of revenue loss here.
- Confirm negative keyword lists and audience exclusions have been refreshed since AI Max was enabled.
B2B lead generation: the prequalification gap
The B2B account had pinned its RSA assets heavily to make sure ad copy filtered out consumer searchers and signaled business buyers. For the test the pins were removed. Click-through rates climbed sharply, but conversion rates fell because the new ads were pulling in B2C traffic. The nuance of prequalifying a B2B audience is something text customization did not handle on its own.
The messaging restrictions included instructions asking the system to prequalify for a B2B audience. A few individual assets met the criterion, but the actual ad combinations users saw did not consistently appeal to business buyers. The team stopped the test after three weeks, restored the pins, and removed the auto-created assets. Performance returned to the pretest baseline within a week.
What to check on a B2B account
- Compare post-click conversion rates by audience signal, not just at the campaign level, since blended CTR can hide a B2C influx.
- Verify any pins that previously enforced business-buyer language are still in place if auto-created assets are running.
- Review search term reports for consumer queries that AI Max may have matched to B2B ad groups.
B2C lead generation: where auto-created assets earn their keep
The B2C account localizes ads through geo-targeted copy and inserts. Its optimized campaigns carried tailored ad copy in nearly every ad group, while its long-tail campaign reused a few generic headline assets across many ad groups. That gap, hand-tuned top campaigns sitting next to a thin long-tail campaign, is a common pattern in large accounts.
Auto-created assets did not outperform the hand-tuned top campaigns. They did improve the long-tail campaign, where the comparison was against human-written copy that had been recycled across many ad groups with little customization. For teams that cannot write unique assets for every long-tail ad group, the feature raised the account’s baseline.
Pattern across all three accounts
Human-written assets still beat AI-generated assets when the team had already invested significant time in optimization. Text customization also struggles with copy that has to do a specific job, such as prequalifying B2B buyers, promoting a specific offer, or running a short-term promotion, because the system tends to generate broad, generic variations rather than audience-specific ones.
Auto-created assets earn their keep in the places a PPC team does not have time to optimize: long-tail campaigns, ad groups with generic copy, and accounts where every ad group cannot get human attention. The feature still needs active oversight. Reviewers must flip the asset filter to see what the system produced, remove roughly one in five assets before they accrue impressions, and watch for cannibalization across campaigns.
What this means for a site audit
For a technical SEO or PPC audit, the practical checklist is short. Confirm auto-created assets are visible in the asset review, since they are hidden by default. Expect to remove close to one in five of them for offer or voice drift. On optimized campaigns, treat AI Max as a risk to existing performance and look for internal cannibalization. On long-tail campaigns, treat it as a likely lift. On B2B accounts with heavy pinning, do not remove the pins without a controlled test, and check post-click conversion rates by audience before judging the results.
FAQ
What is Google Ads AI Max text customization?
Text customization is a feature inside Google Ads AI Max that automatically generates responsive search ad assets for each ad group based on the keywords in that group. It can be paired with messaging restrictions to guide the output.
How did AI Max text customization perform in the ecommerce test?
Auto-created assets underperformed human-written assets on highly optimized campaigns and initially cannibalized traffic from other campaigns, reducing total revenue. After new keywords, negatives, and audience exclusions were added, the feature improved performance on the long-tail campaign.
Why did AI Max text customization fail for the B2B account?
The auto-created assets did not consistently prequalify searchers as business buyers, so click-through rates rose while conversion rates fell. The account stopped the test after three weeks, restored its pinned assets, and returned to its pretest performance within a week.