Performance Max asset automation: what the AI nailed, and where query level truth still beats it

Google's own Ad Strength score is negatively correlated with the thing you actually care about. In roughly 20,000 accounts that Optmyzr looked at in April, ads rated "Excellent" carried the worst cost per conversion in the whole dataset at $28.68, and ads rated merely "Average" came in at $12.43. The "Poor" ones had the best return on ad spend, 327.65 percent. So when PMax asset automation tells you your creative is in good shape, it is grading something, and it is not grading whether the account makes money.

I run enterprise Google Ads accounts, and I want to be careful here, because the generative layer Google shipped into PMax over the last year is genuinely useful and I use parts of it. The problem is not that the AI writes bad assets. It writes fine assets, quickly, and it saves a real amount of production time. The problem is what the reporting underneath it will and will not show you afterward.

PMax now writes the ads, and makes the video whether you asked for one or not

An asset group is the bundle you hand PMax: headlines, descriptions, images, logos, video, and a set of audience signals. Google assembles those into ads across Search, Shopping, YouTube, Display, Discover, Gmail and Maps, and decides which combinations run where. That part has been true for a while.

What changed is how much of the bundle Google will now produce on your behalf. Asset Studio moved from a bulk editing tool into a generative hub across 2026. At Google Marketing Live on May 20 it picked up Gemini Omni for multimodal video generation, natural language creative briefing, and a one click creative testing feature that spins a control against a treatment straight out of any asset edit. Google said global English rollout this summer, so depending on when you read this it may or may not be sitting in your account yet.

The piece that surprises people is the video. If an asset group has no video in it, PMax generates one from your existing images and text, and it does this by default. There is a real number attached to that, and it is not flattering: advertiser supplied video delivers around 12 percent more conversions on average than the slideshow style video PMax builds for you. That figure comes from Google, reported through agency write ups rather than published by Google directly, so treat the exact 12 percent as directional. The direction has been consistent across everyone who has written it up.

Google also raised the video asset cap from 5 to 15 in January and expanded asset level A/B testing to all PMax campaigns around the same time, and started moving disapprovals down to the asset level instead of nuking a whole campaign, which is a genuine improvement in how much control you have. Those last details come from agency and industry blogs rather than Google's own documentation, so I would not build a client deck on the exact figures.

Does a higher Ad Strength score mean better performance?

No, and the Optmyzr data is the cleanest evidence we have on it. Across about 20,000 accounts in their April 2026 RSA study, "Excellent" Ad Strength produced the worst CPA at $28.68 and the lowest conversion rate at 4.97 percent, while "Average" produced the best CPA at $12.43 and the highest conversion rate at 12.65 percent. "Poor" delivered the best ROAS. Ad Strength is a completeness and diversity score. It rewards you for filling every slot with something varied. Nothing in it knows which query bought.

Worth keeping the two Optmyzr numbers straight, because they get mashed together in the coverage. The Ad Strength finding is the April 6 RSA study, roughly 20,000 accounts. The separate Q1 2026 benchmark report published April 23 covers about 21,000 accounts and says something different: ad impressions are down roughly 11 percent year over year as AI Overviews reshape what inventory exists. Different datasets, different questions, and stacking them into one claim is how bad blog posts get written.

The practical read is that Ad Strength is fine as a hygiene check and useless as a performance signal. Fill the slots. Then ignore the badge and go look at what converted, which is the same discipline we end up applying on every account we take over, and it is documented in our case studies.

The reporting does not tell you which asset earned the conversion

This is the part that has not moved much, and it is the reason "let it run" is a bad instruction even when the creative is good.

You can see asset performance labels. You cannot see a clean line from one asset, through one placement, to one conversion, at the query level. So when a PMax campaign improves, you know it improved. Attributing that to the new headline, or to a shift in where Google decided to spend, or to seasonality in your query mix, is guesswork dressed up as analysis. We wrote about that failure mode at length in our piece on over reliance on Google Ads automation.

I want to be fair about the visibility question, because the loud version of this complaint is wrong. Advertisers did not spend years with zero insight into PMax queries. The controls arrived late, which is a different accusation and a more accurate one. PMax negative keywords launched in January 2025 capped at 100, and that cap went to 10,000 in March 2025. Full search terms reporting rolled out from May 30, 2025. The tools exist now. They showed up years into the product's life, after a lot of budget had already moved through it, and the channel level question is still the one nobody can answer from inside the interface: how much of this went to Search Partners, how much to Display, and what did each of those buy.

Can you stop PMax from spending on Search Partners and Display?

Some advertisers can, as of about a week ago, and this is the first real crack in the box.

Google is piloting a setting called Partners (Alfa) that lets selected PMax advertisers opt out of the Search Partner Network and the Google Display Network independently. That has not been possible in roughly five years. Adriaan Dekker surfaced it on LinkedIn, crediting Renaldo Spin; the agency Moroch got access in late June; household income exclusions appeared around the same time in late July. Google confirmed to Digiday that it remains a limited alpha, so this is not a thing you can go switch on today unless you are already in it.

Two details matter more than the feature itself. Both boxes ship enabled by default, so the default remains full network exposure and the opt out is something you have to go find. And the fact that Google built the control at all suggests enough advertisers asked the channel split question loudly enough to move product. If you get access, the first thing worth doing is not switching Display off. It is running with the reporting long enough to find out what that spend was actually doing, because you may find out it was doing something.

What I would actually do this month

Every asset slot gets filled, and then I stop treating the score as feedback. Ad Strength tells you the group is complete. It does not tell you the group is working. Asset Studio earns its keep on the production side of that, because building variants really is faster with it than without it. What it cannot do is tell you which variant deserves budget, and that call still comes out of conversion data you trust.

Real video goes into every asset group that matters, because an empty slot means PMax makes a slideshow and the slideshow converts worse. A rough phone shot of the actual product usually beats a generated stitch of your existing images.

Pull the search terms report on your PMax campaigns and read the report, not the summary sitting above it. Sort by conversions and keep reading until the genuinely high intent separates from the broad guessing, then build the negative list out of that. It is a slow hour and I have never found a way around the slow hour. You have 10,000 slots and most accounts I see are using a few dozen.

Partners (Alfa) is the one item here that most people reading this cannot act on yet. If you do have it, instrument before you exclude. If you do not, ask your rep, and in the meantime treat your PMax numbers as a blend of channels you cannot separate, which is the honest way to report them internally.

None of this is an argument against automation. It is an argument against grading automation with the automation's own scorecard, and the Ad Strength inversion is the cleanest proof we have that the scorecard measures something other than money.

FAQ

Does Ad Strength affect PMax performance? Optmyzr's April 2026 study of about 20,000 accounts found "Excellent" Ad Strength ads carrying the worst CPA at $28.68, and "Average" ads the best at $12.43. So there is a relationship, just not the one most advertisers assume. What the score actually measures is whether your asset group is complete and varied. That is worth doing. It is not a performance forecast, and treating it as one is how accounts end up optimized toward a badge.

Can you see PMax search terms? Yes, and later than the product deserved. Full search terms reporting for PMax rolled out from May 30, 2025, and negative keywords went from a 100 term cap in January 2025 to 10,000 in March 2025. What is still missing is channel level rather than query level: the report will not tell you how spend split across Search Partners, Display and the rest.

Does PMax create video ads automatically? Whenever an asset group has no video supplied, yes, built out of your existing images and text. Google's own reported data, relayed through agency coverage, puts advertiser supplied video at roughly 12 percent more conversions than the auto generated version. Fill the slot yourself.

What is Partners (Alfa)? Not something you can go switch on, first of all. It is a limited Google alpha, confirmed live in late July 2026, that lets selected Performance Max advertisers turn off the Search Partner Network and the Google Display Network independently, which has not been possible in about five years. Both networks are enabled by default. Google has told press it is still a restricted pilot rather than a general release.

Are 10,000 negative keywords enough to control PMax? For query level waste, yes, with far more room than most accounts use. Placement and channel spend are a different problem, and that is the one Partners (Alfa) is aimed at.