Performance Max is a black box. Here's how to crowbar it open.

I have a Slack channel with about a dozen agency owners. Every other week someone drops a screenshot of a Performance Max campaign that's spending $30K/mo with no idea where the conversions came from. The reactions are always the same: "Yeah, mine too."

This is the deal Google made with us in 2022, and most of the marketing world signed it without reading the fine print: hand us your budget, your creative, your audiences, and your conversion data — we'll optimize for outcomes. Trust the algorithm.

The pitch is good. For some advertisers, the results are good. For most B2B SMBs, the results are opaque, which is not the same as bad — it's worse than bad, because you can't tell.

Why Performance Max actually works

Let's be fair before we get critical. Performance Max wins for a specific kind of advertiser: high transaction volume, broad customer fit, and good first-party data. If you sell consumer products with thousands of orders per week and you've fed Google rich conversion signals, PMax can outperform manual campaigns. The algorithm has enough data to actually learn something.

For the rest of us — B2B SaaS with 50 deals/quarter, law firms doing 200 leads/month, RIAs who close eight clients a year — PMax is feeding a model that doesn't have enough training data to converge. So it falls back on its default behavior, which is to spend your money in the easiest places it can find clicks. Those places are usually not the places your customers live.

The black box, in three parts

The opacity isn't theoretical. Three specific things you can no longer see in PMax that you could see in standard search and display campaigns:

1. Search query reports are gutted.

Google used to show you every search term that triggered an ad. Now you get "search themes" — buckets of queries instead of the queries themselves. You see "B2B legal services" but not the 47 queries that made up that theme. Some of those queries were perfect-fit. Some were "free legal advice reddit." You can't tell which.

2. Placement reports are aggregated.

Standard display campaigns show every site, app, and YouTube channel where your ads ran. PMax shows you the top 30 by spend. The other 2,000? Hidden. So when your "B2B SaaS" PMax campaign is quietly burning $4K/month on placements in a Bangladeshi click farm, you'd never know — unless you happened to look at the right top-line metric on the right day.

3. Asset group performance is fuzzy.

You upload 15 ad headlines, 5 descriptions, and 12 images per asset group. Google mixes them. You get an aggregate "asset group performance" score (Best, Good, Low) per asset, not per combination. So you know "Headline #4 is performing well" but not whether it's performing well with image #2 or image #11. Granular optimization that used to be table-stakes is gone.

Performance Max is great when it works. The problem is you can't tell when it doesn't.

Five things you can still control

Now the practical part. Despite Google's best efforts to abstract you out of the loop, there are still levers. Here are the five we use on every PMax account we manage.

1. Audience signals are not "audiences."

Google rebranded "targeting" to "audience signals" in PMax. The wording matters. They're signals, not constraints — Google can and will go outside them. But quality signals do meaningfully shape early-phase learning. Feed PMax a customer match list of your highest-LTV customers, and you'll see better outcomes in week 2 than feeding it nothing. This is the single highest-leverage lever in PMax.

2. Negative keywords work — but you have to ask.

By default, PMax doesn't expose a negative keyword interface. You have to email your Google rep (or chat with their AI assistant) to add account-level negatives. Most agencies don't bother. We bother. Filtering out brand-defense queries, employment searches ("law firm jobs"), and competitor research saves us 8–15% of monthly spend on a typical legal account.

3. Asset groups are the only window left.

Run multiple asset groups, one per audience or product. Don't dump 15 headlines into a single group and let Google figure it out. Set up 3-4 asset groups per campaign, themed by intent ("estate planning", "personal injury", "business law"), and you'll see asset-level performance scores you can actually act on.

4. Conversion value tiers, not just conversions.

Every conversion looks the same to PMax unless you tell it otherwise. If a "free consultation" conversion is worth $50 to you and a "signed retainer" is worth $5,000, you need to send those as different conversion values back to Google. Otherwise PMax optimizes for free consultations — because there are 50× more of them, which makes the algorithm happy and your revenue line flat.

5. Account structure: never let PMax run alone.

The single best move we make on new accounts is running a standard search campaign in parallel on the same keywords. PMax tries to grab those queries, but standard search campaigns "win" the auction when they exist. The standard campaign gives you full search query data. PMax fills in around it. You get the algorithm's reach with the visibility of standard search.

When to fire Performance Max entirely

For a non-trivial percentage of advertisers, the right answer is "don't run PMax at all." We pull it off accounts when:

• Conversion volume is below ~30/month. The algorithm doesn't have enough signal. You're paying Google to learn slowly.
• Your product is high-consideration and B2B. Long sales cycles confuse the optimization. Standard search with intent keywords beats PMax 8 times out of 10 for B2B SaaS, professional services, and considered-purchase verticals.
• Your conversion tracking is weak. If you can't pass back values for different conversion types, PMax will chase the cheapest-to-acquire conversion. That's almost never your most valuable one.
• You have a small, specific audience. "RIAs in California with $500M+ AUM" is not a PMax audience. It's a standard search + LinkedIn audience.

For most B2B mid-market accounts we manage, we run standard search as the primary, PMax as a secondary supporting campaign with a capped budget. The reverse of Google's recommendation. The performance, every time, is better.

The bottom line

Performance Max isn't a scam. For the right advertiser with the right data, it's a legitimate optimization tool. But Google has incentives to abstract you out of the decision-making loop because that increases their leverage over your spend. Every time you accept the default — every time you click "yes, use Smart Bidding" or "yes, expand to all networks" — you're giving up information you used to have access to.

The job of a competent paid media operator in 2026 is not to refuse the algorithm. It's to put the algorithm to work without surrendering visibility. Run PMax. Just run it next to a standard campaign. Set up your asset groups deliberately. Send back conversion values that mean something. Email your rep about negatives.

Or, you know, hire someone who already does. That's what Get Leads is.