Google Ads Performance Max AI Update: What Changed for Advertisers

Performance Max has been through more structural changes in the past eighteen months than in its first two years combined. Google has been pushing AI deeper into every layer of the campaign type — from how assets get assembled and served to how the bidding model interprets conversion signals. For advertisers who set up PMax campaigns in 2023 and haven't revisited the underlying mechanics, several things now work differently in ways that aren't obvious from the UI.

This record documents the changes that matter operationally: what shifted, when it became active, and what it requires from you as a practitioner.

Summary of Changes by Area

Search Term Reporting: What You Can See Now

For most of PMax's existence, search term visibility was the biggest operational complaint. You could see conversion data by asset group but had almost no insight into which queries were actually triggering your ads. That changed in Q3 2025 when Google expanded the search terms report to include category-level query data — not individual queries at the impression level, but aggregated search theme categories that show where traffic is concentrating.

The practical implication: you can now identify if PMax is pulling in queries that belong to a different campaign type (e.g., branded queries you'd rather serve through a dedicated Search campaign, or informational queries that don't convert). Before this update, the only lever was adding negative keywords at the account level — which still applies, but the reporting now gives you enough signal to make that decision deliberately.

Generative AI Asset Creation: What It Actually Does

Google moved generative AI image and headline creation for PMax asset groups into general availability in Q4 2024. The feature uses Google's own image generation model to produce creative assets when you don't supply enough of your own — or when the system determines that generated variants might outperform existing assets.

There are two distinct behaviors to understand here:

• Automatically Created Assets (ACA) — the system generates headlines and descriptions by pulling from your landing page and existing ad copy. This is separate from the image generation feature and has been available longer. As of Q2 2025, ACA is on by default for new campaigns and the opt-out moved to campaign level.
• Generative image creation — produces new images using AI. You can review generated images before they go live, but only if you actively check the asset group. Google does not send alerts when new generated assets are added.

The quality of generated images varies considerably by category. Product-focused retail campaigns tend to produce more usable outputs than service businesses or B2B advertisers, where the AI has less visual reference material to work from. In practice, most experienced PMax managers supply their own image assets and disable ACA for headlines when copy tone is tightly controlled.

Brand Exclusions: Finally at Account Level

Before Q2 2025, controlling branded queries in PMax required a workaround: running a separate branded Search campaign with a higher priority bid and hoping PMax would defer. It worked inconsistently. Google added account-level brand exclusions for PMax in Q2 2025, letting you specify brand terms that PMax should not trigger on across all PMax campaigns in the account.

This matters because branded queries typically have very different economics than non-branded — higher conversion rates, lower CPCs, and a different attribution story. Mixing them into PMax's optimization pool can make campaign performance look better than it is on non-branded traffic, which is usually where the real acquisition challenge sits.

The exclusion applies at the account level, not campaign level. If you want branded queries in one PMax campaign but not another, that configuration isn't possible — it's all-or-nothing per account. For most advertisers this is fine, but multi-brand accounts running PMax for separate brands need to be aware of this constraint.

Bidding and Signal Changes

Enhanced Conversions and First-Party Data

Google's bidding documentation made it explicit in early 2025 that PMax's Smart Bidding model performs materially better when enhanced conversions are set up correctly. Enhanced conversions pass hashed first-party data (email, phone) alongside conversion events, allowing the model to match conversions back to Google-signed-in users even when cookie-based attribution breaks down.

This isn't new functionality — enhanced conversions have existed since 2021 — but the weighting has increased as third-party cookie deprecation has progressed. Advertisers running PMax without enhanced conversions are now operating with a meaningfully degraded signal set compared to those who have it configured.

PMax vs. Search Campaign Auction Interaction

Google's stated policy is that when a Search campaign and a PMax campaign are both eligible for the same query, the Search campaign takes priority. In practice, this has never been perfectly reliable, and Google's own documentation acknowledged in Q1 2025 that the interaction is more nuanced — PMax can win in certain auctions even when a matching Search campaign exists, particularly for queries that the Search campaign's keyword set doesn't explicitly cover.

Customer Match Weighting

Audience signals in PMax are inputs to the model, not hard targeting constraints — the campaign can and does serve ads to people outside your audience signal lists. What changed in Q4 2024 is how the model weights different signal types. Customer match lists (first-party CRM data uploaded as hashed emails) now carry significantly more weight than in-market or affinity segments in the optimization loop.

The practical recommendation: if you haven't uploaded a customer match list to your PMax asset groups, you're leaving signal quality on the table. Even a modest list of existing customers or past converters gives the model a much stronger starting point than relying on Google's own audience segments.

Placement Reporting: Display and YouTube Visibility

Until Q1 2026, PMax's placement-level reporting for Display and YouTube was buried or effectively unavailable through the standard UI. Google added placement-level data to the Insights tab in early 2026, showing which specific websites and YouTube channels your ads appeared on and their associated conversion volume.

This matters for two reasons. First, brand safety: you can now identify placements that are off-brand or inappropriate and add them as placement exclusions. Second, budget efficiency: PMax allocates budget across Search, Shopping, Display, YouTube, Gmail, and Discover automatically. If the placement report shows that Display placements are consuming a large share of spend with low conversion contribution, that's actionable information — though your only direct lever is placement exclusions, not channel-level budget allocation.

What Hasn't Changed (and Why That Matters)

Despite the reporting improvements, several fundamental constraints remain in place as of Q2 2026:

• No channel-level budget controls. You cannot tell PMax to allocate X% to Search vs. YouTube. The model decides. Placement exclusions are the only indirect lever.
• No query-level negative keywords at campaign level. Negatives can be added at the account level or through a shared negative keyword list, but not at the individual PMax campaign level.
• No individual asset performance data. Assets are rated Low / Good / Best, not by specific conversion or CPA metrics. You can't directly compare a headline's ROAS contribution against another.
• Learning periods after changes. Any significant change to an asset group, audience signal, or conversion action triggers a re-learning period. Google doesn't publish a fixed duration, but in practice it's typically 2–4 weeks before performance stabilizes.

Operational Checklist for Current PMax Campaigns

Given the cumulative effect of these changes, here's what to verify in any PMax campaign that's been running for more than a few months:

1. Confirm enhanced conversions are active and passing data. Check the Diagnostics tab in the Conversions section — look for match rate above 30%.
2. Review automatically created assets. Go to Asset Groups > Assets > Automatically created and audit any generated headlines or images for brand compliance.
3. Set account-level brand exclusions if you're running a separate branded Search campaign. This prevents PMax from competing on your own brand terms.
4. Upload a customer match list as an audience signal if you haven't already. Even a list of 500+ past purchasers will improve model signal quality.
5. Check the Insights tab for placement-level data. Add any clearly off-brand placements to your exclusion list.
6. Review the search themes report under Insights to identify query categories that should be routed to a dedicated Search campaign instead.