Google Ads Performance Max Best Practices: How to Actually Control the Black Box (Before It Controls Your Budget)

Performance Max has a 78% adoption rate among Google Ads accounts spending over $10k/month. The majority of those advertisers have no idea what it’s actually doing with their money.

That’s not a knock on the people running those accounts. It’s a knock on how Google sold PMax — as an autonomous, AI-powered campaign type that “finds conversions across all Google channels.” What they glossed over: the AI optimizes aggressively toward whatever conversion signal you give it, on whatever inventory it decides is worth bidding on, with almost no transparency into why. If your conversion tracking is even slightly off, or your asset groups are lazy, or you forgot to add a brand exclusion list, PMax will find creative ways to spend your budget on things that look like wins but aren’t.

The Honest Case For Performance Max (It’s Not All Smoke)

Before we go full contrarian, let’s give credit where it’s due. For the right account, PMax genuinely outperforms what a human can build manually. The reason is reach arbitrage: Google’s AI can find conversion opportunities across Search, Shopping, Display, YouTube, Discover, Gmail, and Maps simultaneously — and it can shift budget between those channels in real time based on auction signals no human has access to.

If you’re running an ecommerce brand with a clean Google Merchant Center feed, a solid 6-month conversion history, and products in a category with real demand across multiple channels, PMax can find incremental revenue that a standard Shopping + Display stack would miss. We’ve seen well-structured PMax campaigns pull 20–35% more conversion volume at comparable ROAS compared to Shopping-only campaigns for established ecommerce clients.

The catch — and it’s a big one — is that “well-structured” is doing a lot of heavy lifting in that sentence. Most accounts don’t launch PMax well-structured. They launch it default, watch the metrics look okay for 30 days because the AI is hoovering up existing demand, and then call it a success. That’s not PMax working. That’s PMax surviving on the demand you already built.

When Performance Max Actively Hurts You (And Why Your Reports Won’t Show It)

Here’s the scenario we’ve watched play out more times than we can count: An account manager adds a PMax campaign alongside existing Search and Shopping campaigns. Conversions go up account-wide. Everyone’s happy. What’s actually happening: PMax identified your branded search terms, started bidding on them aggressively, and is now reporting those conversions — which would have happened anyway at a fraction of the cost — as PMax wins.

Google’s attribution model hands PMax credit generously. Because it runs across every channel, it’s almost always in the conversion path somewhere. That makes isolating its true incremental contribution genuinely difficult, and most standard reporting won’t flag it automatically.

PMax also tends to underperform — or outright damage efficiency — in these specific situations:

• Low conversion volume accounts. Under 30 conversions per month, the algorithm is basically flipping coins. It needs volume to learn, and without it, you’ll see erratic CPAs and budget waste that looks like “normal variance.”
• Highly regulated or niche B2B industries. If you sell something with a 90-day sales cycle and your conversion event is a form fill, PMax will optimize toward form fills — including from people who will never buy. Quality signal degrades fast.
• Accounts with weak or incomplete conversion tracking. PMax doesn’t question your tracking setup. It just optimizes toward whatever fires. If your thank-you page tag also fires on back-button navigations, PMax will find those people with terrifying efficiency.
• New campaigns with no historical data. The learning phase for PMax is brutal — typically 4–6 weeks where the algorithm burns budget experimenting. If your monthly budget is under $3,000, you may not have enough spend to survive the learning phase at a sustainable CPA.

PMax Asset Groups: The Lever Most Advertisers Ignore Completely

Asset groups are where most Performance Max best practices advice stops at “add all your assets and let Google optimize.” That advice will cost you. The structure of your PMax asset groups is the primary way you communicate intent to the algorithm — and if you throw everything into one group, you’re essentially telling Google to figure it out.

The right approach: build asset groups the same way you’d build tightly themed ad groups in a Search campaign. One asset group per distinct product category, audience intent, or funnel stage. If you sell kitchen appliances, you’re not building one asset group called “All Products.” You’re building separate groups for stand mixers, espresso machines, and cookware — each with headlines, descriptions, images, and audience signals that are specific to that category.

Here’s why this matters mechanically: Google uses your asset group structure to determine which ads to show to which users. When you mix product categories, the AI averages your signals and you get mediocre relevance across the board. When your asset groups are tight and specific, the AI has a clear brief. Ad relevance improves. Conversion rates improve. The algorithm has something real to learn from.

A few non-negotiable rules for asset group construction:

• Audience signals are not optional. Add your customer match lists, website visitor segments, and in-market audiences as signals. These don’t restrict who sees your ads — they tell the AI where to start looking. Without them, it starts completely cold.
• Use final URL expansion carefully. By default, PMax can send users to any page on your site it thinks is relevant. Turn this off or use URL exclusions unless you’ve audited every landing page for conversion readiness. Nothing kills a PMax campaign faster than traffic landing on a blog post or an out-of-stock product page.
• Video assets are not optional if you care about where your ads show. If you don’t provide video, Google will auto-generate one from your static assets. These auto-generated videos are almost universally terrible. Shoot something — even a simple 15-second product or service video — and upload it. It’s worth it.

The Controls Google Doesn’t Advertise (But You Should Be Using)

PMax gets a reputation as a “set it and forget it” black box, and Google has done little to dispel that myth because the defaults benefit Google’s revenue, not yours. But there are real controls available, and most accounts don’t use them.

Brand exclusions. This is the most important one. Go into your PMax campaign settings and add your brand terms as brand exclusions. This prevents PMax from bidding on your own brand name and inflating its reported performance with conversions that your branded Search campaign should be capturing at a fraction of the CPA. If you’re not doing this, your PMax ROAS number is almost certainly misleading.

Campaign-level negative keywords. Google quietly rolled out campaign-level negative keywords for PMax. Use them. Block irrelevant query categories, competitor names (if you’re not running a conquesting strategy), and any terms you know from your Search campaign data convert at unacceptable CPAs. This is one of the highest-leverage PMax optimization moves available right now.

Placement exclusions. PMax runs on Display and YouTube inventory by default. A significant portion of that inventory is garbage — low-quality sites, mobile app placements where accidental clicks run rampant, parked domain networks. Submit a placement exclusion list at the account level. Start with standard mobile app category exclusions (mobileappcategory::69 for games is a classic one) and build from there using your placement reports.

URL exclusions. Block pages you never want to drive paid traffic to: careers pages, login pages, thin blog content, out-of-stock category pages. This takes 15 minutes and can meaningfully improve your traffic quality.

Seasonality adjustments. If you have a major sale, product launch, or external event that will spike conversion rates temporarily, use Google’s seasonality adjustment tool. PMax’s smart bidding will overcorrect wildly if you don’t signal that a conversion rate spike is temporary. We’ve seen accounts with 3x normal conversion rates during a 48-hour sale followed by two weeks of under-bidding as the algorithm recalibrated.

How to Structure PMax Alongside Your Existing Campaigns

The single biggest structural mistake we see: launching PMax as a replacement for Search and Shopping campaigns instead of a complement. PMax should expand your reach, not cannibalize your existing efficient campaigns.

The structure that works for most accounts:

Keep your branded Search campaign running separately at a high enough budget and bid to maintain 90%+ impression share. Apply brand exclusions to PMax so it can’t compete. Branded conversions belong to branded Search — they’re cheap, high-intent, and your best CPA source. Don’t let PMax claim credit for them.

Keep your best-performing non-branded Search campaigns running if they’re hitting your target CPA. PMax and Search can coexist. When there’s a query that matches both a Search keyword and PMax’s targeting, Google will generally prioritize the Search keyword — but only if that keyword is in a Standard or Exact match campaign. Broad match keywords in Search campaigns will compete with PMax in ways you can’t easily see or control.

Use PMax for incremental reach — the inventory, the audiences, the channels that your Search and Shopping campaigns can’t touch. Think of it as your expansion engine, not your foundation. Once it’s learning and performing, you can gradually increase its budget share based on actual incrementality data, not just reported conversions.

For ecommerce accounts running Standard Shopping, the PMax transition question comes up constantly. Our current guidance: don’t fully replace Standard Shopping with PMax unless PMax has proven incremental performance over at least 60 days of parallel running. The control you give up — negative keywords, query-level visibility, product-level bid adjustments — is significant.

What “Good” PMax Performance Actually Looks Like (Real Benchmarks)

One reason PMax is so easy to misread is that there are almost no widely-shared benchmarks for what healthy performance looks like. Here’s what we track, and what the numbers typically look like for accounts doing it right:

Learning phase duration: 4–6 weeks. During this period, expect CPA to run 20–40% above target. If you’re not seeing improvement by week 6, the campaign has a structural problem — usually thin conversion volume, poor asset quality, or missing audience signals.

Asset performance ratings: In your asset group reporting, Google assigns “Low / Good / Best” ratings to individual assets. Target having at least 40% of your headlines rated “Good” or “Best.” If most of your assets are “Low,” your creative is misaligned with what’s resonating. Pull the low performers and test new copy.

Search term themes: Google recently improved PMax’s search term reporting to show “search categories” driving performance. Review this monthly. If you’re seeing irrelevant category themes driving volume, that’s your signal to add negative keywords or tighten your audience signals.

ROAS/CPA vs. target: A well-optimized PMax campaign should hit target ROAS within 10–15% after the learning phase. If you’re consistently 25%+ above target CPA and it’s been more than 8 weeks, PMax either doesn’t have enough signal to learn from or the campaign structure needs a rebuild.

Frequently Asked Questions

Should I run Performance Max instead of Standard Shopping?

Not automatically, and not without testing both in parallel first. Standard Shopping gives you more control — negative keywords, product group bid adjustments, query-level data. PMax can deliver more reach and higher conversion volume for accounts with strong data history, but it comes at the cost of visibility and control. Run both for 60 days, segment your conversion data carefully, and let actual incrementality drive the decision — not PMax’s self-reported numbers.

How many asset groups should a Performance Max campaign have?

As many as it takes to keep each group tightly themed around a specific product category, audience intent, or funnel stage. A good rule of thumb for ecommerce: one asset group per top-level product category. For B2B or service businesses: one asset group per core service line or buyer persona. Avoid the temptation to consolidate everything into one or two groups for “simplicity” — the algorithm performs better with clear, specific signals.

Can Performance Max target specific keywords?

Not directly — PMax doesn’t use traditional keyword targeting. It uses audience signals, landing page content, asset text, and historical account data to infer which queries to bid on. The closest thing to keyword control is your audience signals (point the AI at high-intent audiences), your asset copy (include relevant terms in headlines and descriptions), and campaign-level negative keywords to exclude queries you don’t want.

Why is my Performance Max ROAS so high but overall account revenue flat?

This is the cannibalization problem in action. PMax is almost certainly taking credit for conversions — especially branded conversions — that your other campaigns generated or would have generated at a lower cost. Run an experiment: add brand exclusions to PMax and compare total account conversion volume and revenue before and after. If revenue stays flat but PMax ROAS drops, you’ve found your answer.

What’s the minimum budget to run Performance Max effectively?

Practically speaking, you need enough budget to generate at least 30–50 conversions per month within the PMax campaign specifically. For a service business with a $50 CPA target, that’s a minimum of $1,500–$2,500/month dedicated to PMax just to clear the signal threshold. Below that, the algorithm stays in a perpetual semi-learning state and your CPAs will be inconsistent. If you’re under that threshold, stick with tightly managed Search campaigns until your volume grows.

Does Performance Max work for lead generation (not just ecommerce)?

It can, but it requires more care. The fundamental problem with PMax for lead gen is conversion quality — the algorithm optimizes toward form fills or calls, not toward qualified leads or closed revenue. If you’re not feeding it offline conversion data (closed deals, qualified lead signals), it will optimize toward volume over quality. The fix: import offline conversions from your CRM and set those as your primary conversion action. This is non-negotiable for B2B lead gen accounts running PMax.