Google Ads Performance Max: What $80K in Spend Actually Taught Me

Performance Max is Google's bet that machine learning can outplace you across Search, Shopping, YouTube, Display, Discovery, and Gmail simultaneously. Sometimes it does. Other times it dumps your budget into Gmail placements that generate impressions nobody asked for. These notes come from running pMax across four e-commerce accounts and two lead-gen verticals over about 14 months, roughly $80K total. Not a guide from someone who read the docs. Just what worked, what broke, and what I wish someone had told me before I turned on the first campaign.

What Performance Max actually is (and isn't)

pMax is a single campaign type that serves ads across every Google property. You give it creative assets, audience signals, a conversion goal, and a budget. Google handles the rest - where to show, who to target, which asset combination to use.

That's the pitch. The reality is more complicated.

pMax replaces Smart Shopping and Local campaigns. It runs on a goal-based bidding system (tCPA or tROAS). You build asset groups instead of ad groups, and those asset groups contain a mix of text headlines, descriptions, images, videos, and audience signals. Google assembles them into ads on the fly.

What it is not: a Search campaign with extra reach. If you treat it like Search, you will be confused by the reporting, frustrated by the lack of keyword control, and annoyed when it starts eating your branded traffic. pMax is a different animal. You have to accept less visibility in exchange for broader automation. Whether that trade-off is worth it depends entirely on your account.

When pMax works and when it falls flat

I've seen pMax crush it in two scenarios consistently:

E-commerce with 200+ SKUs and a clean product feed. One DTC home goods account went from $22 ROAS on Standard Shopping to $34 ROAS after switching to pMax. The campaign found purchase-intent audiences on YouTube and Discovery that we weren't reaching before. That account had 18 months of conversion history and a well-structured Merchant Center feed. pMax had a lot to work with.

Lead gen with a clear, high-value conversion event. An insurance lead account running tCPA at $45 hit $38 actual CPA within three weeks. But that account was feeding 80+ offline conversions per week back into Google Ads via enhanced conversions. The feedback loop was tight.

Where pMax struggled:

New accounts with thin data. One client launched a brand new Google Ads account and went straight into pMax. Spent $4,200 in the first month, got 11 conversions at $382 CPA (target was $60). There was no conversion history for the system to learn from. We switched to manual CPC Search for two months, built up 200+ conversions, then relaunched pMax. Second time around, CPA settled at $52 within two weeks.

Low-budget campaigns. Below $80/day, pMax just doesn't have enough room to test across six channels. It picks one (usually Display or Gmail) and dumps most of the budget there. I've seen $50/day campaigns where 70% of spend went to Display placements with a 0.04% CTR.

Asset group structure that survived testing

Early on I made the mistake of putting everything into one asset group. One massive group with 15 headlines, 5 descriptions, 20 images. Google technically allows it. But I couldn't tell what was working for which product category or audience.

The structure that actually gave me useful data:

Key detail: listing groups in each asset group should not overlap. If the same product appears in AG1 and AG4, you have no idea which group is driving results. Use listing group filters to carve clean segments.

For lead gen without a product feed, the split is different. I typically group asset groups by offer type or landing page. A SaaS client had one asset group for free trial signups and another for demo requests. Different headlines, different images, different final URLs. That separation let us see that demo requests came almost entirely from Search and YouTube, while free trial conversions came from Display and Discovery.

Audience signals: what they mean and how much they matter

Google's documentation says audience signals are "suggestions, not restrictions." That is technically accurate and practically misleading.

In the first 2-3 weeks, pMax leans heavily on your audience signals. They are not restrictions in the long run - the campaign will expand beyond them - but they shape where the algorithm starts looking. Bad starting signals mean a longer, more expensive learning phase.

What I include as signals:

• Customer match list. Upload your buyer list. Even 500 emails helps. This is the single strongest signal you can provide. In one account, adding a 2,000-person buyer list cut CPA by 23% in the first two weeks versus running without it.
• Website visitors (Google Ads remarketing list). Past 90-day visitors, especially converters.
• In-market segments. Pick 3-5 that match your product. Don't add 30 segments "just in case" - it dilutes the signal.
• Custom segments based on search terms. This is underused. You can create a custom segment from keywords people have searched. For an outdoor furniture brand, I added terms like "teak patio set," "outdoor dining table," "weatherproof furniture." This gives pMax a Search-like starting point.

What I leave out: demographics (age, gender, income). In my experience, pMax ignores these pretty quickly and they don't improve the learning phase. Affinity audiences are too broad to be useful as signals.

The brand cannibalization problem

This is the issue nobody warns you about until it's already happened. pMax will serve on your branded search terms. It will do this quietly, and the performance will look great because branded traffic converts at high rates regardless of which campaign captures it.

Here is what happened in one account. We launched pMax alongside an existing branded Search campaign. Within two weeks, the branded campaign's conversion volume dropped 40%. pMax's Insights tab showed a top search category of "[brand name] + [product]." The pMax campaign looked like a hero - $18 CPA vs the account average of $44. But it was just eating the cheap conversions that the branded campaign used to capture.

The fix: use brand exclusions. Google added account-level brand lists in late 2024. Go to Campaigns > Settings > Brand exclusions and add your brand (and common misspellings). This forces pMax to find genuinely new customers instead of claiming credit for people who already know you.

After excluding branded terms, that same account's pMax CPA went from $18 to $47. Worse on paper. But the branded Search campaign recovered its volume, and total account conversions actually increased 12% because pMax was now forced into prospecting.

Budget and bidding: what I start with

My starting framework for e-commerce pMax:

The biggest mistake I see: setting a tROAS target on day one that matches your best-performing Standard Shopping campaign. pMax needs room to explore. If your Shopping ROAS is 400%, start pMax at 320% tROAS. Let it find its legs. You can always tighten later.

Starting with "Maximize conversions" (no target) for the first two weeks sounds scary. You will spend more per conversion than you want. But it gives the algorithm maximum flexibility to explore channels and audiences. The data it collects during this phase pays off when you add a target later. I tried starting with tight tCPA on three different accounts. All three had longer learning phases and worse month-two performance compared to the "open then constrain" approach.

Reading the black box: where to find actual data

pMax's reputation as a "black box" is partly deserved and partly laziness. There's more data available than most people check. Here's where I look:

Insights tab (campaign level). Shows top search categories, audience segments driving conversions, and asset performance ratings. Check this weekly. If your top search category is your own brand name and you haven't excluded branded terms, that's your first problem to fix.

Asset details report. Each asset gets rated: "Best," "Good," "Low," or "Learning." Replace "Low" rated assets every 2-3 weeks. Don't replace everything at once - swap 2-3 assets at a time so the algorithm doesn't restart from scratch.

Placements report. Go to Reports > Predefined > Performance Max campaigns > Placements. This shows you where impressions went. It won't show conversions per placement (thanks, Google), but it tells you if 60% of your impressions went to "anonymous.google" Display placements versus youtube.com. If Display is eating your budget, your audience signals probably need work.

Search terms insight. Inside the Insights tab, look for "Search term insights." This shows clustered search themes, not individual keywords. It's less granular than Search campaign search terms, but good enough to spot if pMax is going after irrelevant queries. I caught one account where pMax was spending on "[competitor brand] reviews" - useful to know, even if you can't add negative keywords directly to pMax.

The negative keywords workaround: you can add account-level negative keyword lists that apply to pMax. Not campaign-level, account-level. It is buried under Account Settings > Negative keyword lists. Use it to block obviously wasteful queries. I add competitor brand names, "free," "jobs," and any terms that pull in the wrong intent.

Creative assets: what I learned the hard way

Google asks for up to 15 headlines, 5 descriptions, 20 images, and 5 videos per asset group. Filling every slot is not the goal. Filling them with assets that make sense together is.

My minimums for a new asset group:

• 5-8 headlines (mix of benefit-driven, price-driven, and brand)
• 3-4 descriptions (specifics beat generics every time)
• 8-10 images in both landscape (1200x628) and square (1200x1200)
• 1-2 videos, 15-30 seconds each

The video point matters more than you'd expect. If you don't upload a video, Google auto-generates one from your images. These auto-generated videos look terrible. Slideshow transitions, generic music, text overlapping at weird angles. I've seen them hurt overall campaign performance because YouTube placements got bad engagement metrics, and that dragged down the campaign's quality signals.

A quick screen recording of your product, even shot on a phone, outperforms Google's auto-generated videos. In one account I replaced the auto-video with a 20-second phone walkthrough of the product. YouTube placement CTR went from 0.3% to 1.8%. Not because the video was cinematic. Because it looked like something a real person made.

Image specs that trip people up

Google requires both landscape and square versions. If you only upload landscape, your Display and Discovery placements will either be cropped badly or not served at all. Square images are particularly important for mobile Discovery feed and Gmail promotions.

One account uploaded 12 landscape images and zero square ones. Display impressions dropped 80% compared to a similar account with both formats. The algorithm couldn't serve in placements that needed square creative, so it just... didn't.

pMax vs Standard Shopping: when I run both

There's a common setup I keep coming back to: Standard Shopping for your top 20-30 products (the ones you know convert and want max control over), and pMax for everything else.

The trick is making sure they don't fight each other. Use campaign priority settings in Standard Shopping (set to High) and listing group exclusions in pMax. That way your top sellers stay in Standard Shopping where you control bids, and pMax handles the long tail of your catalog where automation's flexibility is actually useful.

Numbers from one e-commerce account running this split for 6 months:

Standard Shopping had better ROAS but lower new customer acquisition. pMax brought in more first-time buyers at a lower AOV. Combined, the account grew total revenue 28% year-over-year. Neither campaign type alone would have done that.

Seven mistakes I've made (so you can skip them)

1. Setting tROAS too tight on launch. Started an e-commerce pMax at 500% tROAS because that's what Standard Shopping was doing. Campaign barely spent $20/day of its $150 budget. Dropped to 300% tROAS. Spend normalized within 48 hours and ROAS settled at 360% by week three.

2. Not excluding brand terms. Already covered above, but it deserves repetition. pMax will silently eat your branded conversions and make itself look better than it is.

3. Reacting to day-one data. Day one of a pMax campaign will look awful. CPA 4x target, spend going to random placements, terrible CTR. This is normal. I killed a campaign after 3 days that probably would have been fine by week two. Never again. 14-day minimum before any judgment.

4. Ignoring the auto-generated video. Covered above. Upload your own video, even a bad one.

5. One giant asset group. Also covered. Split by product category or offer type. You need the segmentation for reporting if nothing else.

6. Forgetting URL expansion. By default, pMax can send traffic to any page on your site, not just the final URL you specify. This is called URL expansion, and it's on by default. I had a B2B client where pMax was sending people to their blog posts and careers page instead of the demo signup. Turn off URL expansion if you need users to land on a specific page, or add URL exclusion rules for pages that shouldn't receive paid traffic.

7. Comparing pMax CPA to Search CPA directly. pMax serves across six channels. Search CPA is naturally lower for high-intent queries. Comparing them straight is like comparing your retargeting CPA to your cold prospecting CPA and concluding retargeting is "better." They do different things. Compare pMax to your total account blended CPA for a fairer picture.

The reporting problem (and my workaround)

The hardest part of pMax is telling clients what's going on. "The machine learning is learning" doesn't satisfy anyone who's spending $5,000 a month.

My reporting template for pMax campaigns pulls from three sources:

1. Google Ads performance data: conversions, CPA/ROAS, spend, conversion value
2. Insights tab exports: top search categories, top audience segments, asset ratings
3. GA4 acquisition report filtered by google/cpc: new vs returning users, engagement rate, pages per session

I combine these into a weekly snapshot that shows: "Here's what pMax is doing, here's who it's reaching, here's how engaged those people are after clicking." It's not as clean as a Search campaign report with keyword-level data. But it gives enough visibility to make decisions.

One specific metric I track that most people skip: new customer acquisition cost. In GA4, segment by first-time visitors from google/cpc and look at their conversion rate separately. If pMax is mostly re-converting existing customers, your new customer CPA might be 3-4x what the blended CPA shows. That changes whether the campaign is actually growing the business or just taking credit for repeat purchases.

FAQ

Should I use Performance Max or Standard Shopping campaigns?
If you have fewer than 200 SKUs and need tight control over bidding per product category, Standard Shopping gives you more levers. pMax works better when you have a large catalog, decent creative assets, and enough conversion data (50+ conversions per month) for the automation to learn from. Many advertisers run both - Standard Shopping for top sellers and pMax for catalog coverage.

How long does Performance Max take to learn?
Google says 2 weeks. In practice, expect 3-4 weeks of inconsistent performance before the campaign stabilizes. Accounts with existing conversion data and audience signals ramp faster. Brand new accounts with no history can take 6+ weeks and significant test budget.

Can I see where my Performance Max ads are showing?
Partially. The Insights tab shows top-performing audience segments and search themes. The Placements report (under Reports > Predefined) shows which networks got impressions. You can also run placement exclusion lists. But you will never get the keyword-level and placement-level transparency of Standard Search or Display campaigns.

How much budget does Performance Max need?
Set daily budget at minimum 3x your target CPA. If your target CPA is $40, start at $120/day minimum. Below that threshold the campaign cannot collect enough data per day to optimize across all six channels.

Does Performance Max cannibalize my Search campaigns?
Yes, frequently. pMax gets priority on Shopping placements and will serve on branded search terms unless you add brand exclusions. Check the search terms report in Insights and compare it against your branded Search campaign performance. If branded CPA suddenly improves in pMax while your Search brand campaign volume drops, that's cannibalization.

Bottom line

Performance Max works when you feed it the right inputs: clean data, strong audience signals, decent creative assets, and enough budget for the system to learn. It fails when you give it nothing to work with and expect magic.

I've moved most of my larger e-commerce accounts to a pMax + Standard Shopping split. Lead gen accounts depend on conversion volume - if you're getting 50+ conversions per month, pMax can usually beat manual campaigns within a month or two. Below that threshold, you're better off with manual Search until you build up enough data.

The biggest shift in running pMax versus traditional campaigns is accepting that you're managing inputs (assets, signals, budget, goals) instead of managing execution (keywords, bids, placements). If you're the type who needs to see every search term and set every bid manually, pMax will drive you crazy. If you're willing to judge it by business outcomes rather than campaign mechanics, it's genuinely capable of finding customers you wouldn't have reached otherwise.