In a traditional responsive search ad (RSA) campaign, asset testing is relatively straightforward: you see which combinations the AI shows most often and which have the highest conversion rates. In AI Max, asset testing becomes more complex because the AI is mixing assets based on context, user signals, and campaign goals simultaneously — not just evaluating creative performance in isolation.
What AI Max Asset Metrics Actually Show
In a standard RSA campaign, Google Ads shows "Best," "Good," and "Low" performance ratings for each headline and description. These ratings reflect how often the AI selected that asset for serving and the conversion rates associated with the combinations it appeared in.
In AI Max, you have access to similar asset performance data, but the interpretation is different:
"Best" asset: The AI served this asset frequently AND the auctions it appeared in had higher conversion rates. But this doesn't necessarily mean the asset caused the high conversion rate — it may correlate because the AI learned that certain assets work better for high-intent queries (and the AI was already predicting high conversion probability when it chose the asset).
"Low" asset: The AI stopped serving this asset, either because the combinations it appeared in underperformed or because other assets provided better alternatives. A "Low" rating doesn't mean the asset is bad copy; it may mean it's redundant with other assets that convey the same message.
Key point: Individual asset ratings in AI Max reflect relative AI preference within your current asset set, not absolute creative quality.
Why A/B Testing Assets Is Harder in AI Max
In standard campaigns, you can run Campaign Experiments with different RSAs to isolate which ad performs better. The experiment splits traffic between the original and variant campaign.
In AI Max, campaign experiments are available but more complex:
The AI may select different assets for different audiences: A headline that performs poorly on average may perform well for a specific audience segment the AI has learned to target. Aggregate performance data hides this audience-level signal.
Asset changes affect the entire mixing pool: When you add or remove an asset from AI Max, you're changing all possible combinations — not just the specific combination you're thinking about. Adding one new headline interacts with all existing descriptions.
The learning phase resets partially: Adding several new assets at once triggers partial relearning, which means a performance change after adding assets may be relearning effect, not a reflection of the new asset quality.
Using Google Ads Experiments for AI Max Asset Testing
The cleanest way to test creative changes in AI Max is through Campaign Experiments (Google Ads > Campaigns > Experiments > Create experiment):
- Create a draft of your AI Max campaign with the proposed creative changes (new headlines, different descriptions, revised URL strategy)
- Run the experiment with a 50/50 traffic split for 4-6 weeks (minimum 4 weeks to allow both versions to exit the learning phase)
- Compare conversion rate and cost efficiency between the original and the experiment
This approach properly controls for the learning phase effect — both the original and experiment run simultaneously for the same time period, so seasonal effects and market changes affect both equally.
Important limitation: For AI Max for Shopping (https://yositeup.com/blog/google-ai-max-shopping-replacing-performance-max-2026), experiments must include a product feed in both versions. Changes to feed attributes (titles, descriptions) within the experiment are not supported directly — feed experiments require changing the live feed and observing effects without a controlled experiment.
The June 2026 Deletion and Historical Asset Performance Data
The June 2026 data deletion (https://yositeup.com/blog/google-ads-reporting-data-deleted-june-2026) deleted historical asset performance data for the affected period. If your AI Max campaign was building asset performance signals during the deletion period, those signals are gone — the AI's asset preferences after the deletion may be based on a shorter data window.
Practical implication: If you notice asset ratings changing or performance fluctuating after the deletion, this is partially explainable by the AI rebuilding asset preferences without the deleted historical data. Give the asset performance 8-10 weeks to stabilize before making creative changes based on the new ratings.
RSA Asset Best Practices Still Apply in AI Max
The principles of RSA copywriting carry over to AI Max asset creation:
- Write for independence: Each headline and description should make sense even if paired with any other asset. Avoid sequences ("Step 1: Sign up" as headline, "Step 2: Launch" as another headline) — the AI may serve them out of order or not together.
- Cover the funnel: Include assets for different funnel stages — awareness (brand credibility), consideration (product benefits), decision (call to action, price points, offers)
- Use numbers and specifics: "Save 20%" outperforms "Save Money" in click-through. Specifics are more credible and the AI can mix them with contextually relevant descriptions.
- Include your target keyword themes: AI Max's RSA component uses the ad copy as a signal along with the query. Including your primary product terms in headlines helps the AI select the most relevant assets for specific queries.
Asset Performance After DSA Migration
When migrating from DSA to AI Max for Search (https://yositeup.com/blog/google-ads-dsa-ai-max-migration-february-2027), DSA auto-generated headlines from page content. These auto-generated headlines don't exist in AI Max — you need to write RSA assets manually for the AI Max campaign.
Write a minimum of 10-12 unique headlines and 4+ descriptions for your AI Max for Search campaign at launch. This gives the AI a large enough asset pool to test combinations effectively during the learning phase. Launching with only 3-4 headlines produces limited variation and can extend the learning phase as the AI exhausts combinations quickly.
The July 2026 ToS and AI-Generated Assets
The July 2026 ToS (https://yositeup.com/blog/google-ads-tos-july-2026-ai-automation-what-changed) covers AI Max's ability to auto-generate additional assets beyond what you've uploaded. This feature (called "Automatically created assets") allows the AI to generate new headlines based on your landing page content, business profile, and existing assets.
If you enable this feature, the AI-generated assets appear in your asset list and are subject to the same quality assessment and policy review as your manually created assets. You can see which assets are auto-generated (labeled "Auto") in the Assets view.
For creative testing purposes: treat AI-generated assets as the AI's proposed copy. Review them, and if they're performing well, adopt them as your own assets (copy the text into a manual asset). Pausing the "Automatically created assets" feature after adoption ensures the AI uses your deliberate version, not the potentially inconsistent auto-generated version going forward.
Top comments (0)