Most PMax campaigns underperform for five reasons: (1) one asset group covers everything, (2) audience signals are missing or generic, (3) brand searches are leaking in, (4) creative inputs are thin, (5) conversion values aren't differentiated by product. Fix the first three in the first week. The other two take a month.
Why Performance Max breaks differently than other campaigns
PMax is the most opaque campaign type Google has ever shipped. You hand it conversion data, asset inputs, and a budget — and it decides which six channels to serve, which audiences to target, and what creative to assemble. When it works, it's the most efficient cross-channel campaign in the platform. When it doesn't, you have almost no visibility into why.
That opacity is why traditional optimization instincts fail on PMax. You can't see which keyword converted (Google shows search themes, not terms). You can't see which audience signal mattered (you see aggregate asset group performance). You can't pause individual placements. The only meaningful tuning levers are structural: how many asset groups, what signals, what exclusions, what creative variety. Get those right and PMax compounds; get them wrong and the algorithm finds the cheapest conversions it can — which usually means stealing brand traffic.
Mistake #1: A single asset group serving everything
The default PMax setup ships with one asset group. Most operators leave it that way. That's the structural mistake that contains all the others.
One asset group means: one set of creative, one audience signal, one product feed feeding one optimization target. The algorithm picks whichever combination of product/audience/creative converts cheapest, then preferentially serves that combination — which means everything else in the asset group is starved of impressions.
The fix: Split asset groups by either product category (separate groups for Furniture, Lighting, Decor) or by funnel stage (one for high-AOV upsells, one for prospecting). Three to five asset groups in a single PMax campaign is the working range. Below three you're under-segmenting; above five you're fragmenting data and stretching the learning phase.
Diagnostic: Open Insights → Asset performance. If 80%+ of conversions come from one asset group while the others sit near zero, you're over-consolidated.
Mistake #2: No audience signals — or all-broad signals
Audience signals tell the algorithm where to start looking. They're hints, not constraints. Google has reframed them in 2026 as "signals" specifically because PMax will go beyond them if it finds better conversions elsewhere.
What goes wrong: most operators either leave audience signals blank (PMax has nothing to bias toward, takes longer to learn) or stack every possible signal (Customer Match + In-Market + Custom Audiences + Demographics + Interests), which gives the algorithm no clear seed direction.
The fix: Use exactly two signal types per asset group:
- One first-party signal — a Customer Match list of past purchasers, high-LTV customers, or recent abandoners. This is the highest-fidelity seed PMax can have.
- One intent signal — a Custom Audience built from URLs (your top-converting product pages) or search terms (queries that already convert in Search).
Skip In-Market and Affinity for prospecting PMax — those are too broad and confuse the model. Save them for upper-funnel campaigns.
Mistake #3: Brand searches bleeding into PMax
This is the most common silent leak. PMax serves on Search inventory (text ads at the bottom of SERPs) including brand queries unless you explicitly exclude them. Because branded queries convert at extraordinarily low CPA, the algorithm preferentially feeds budget there — which inflates PMax's reported ROAS while quietly stealing conversions that would have been free-with-brand-Search anyway.
The fix: Add account-level negative keywords for your brand and all variants. As of 2024, you can request brand exclusions for PMax campaigns directly from Google support, which is more thorough than negative keywords. Both are worth implementing.
Diagnostic: Open Insights → Search themes. If your brand name or close variants appear in the top themes, PMax is feeding on brand. The CPA there will look great, but compare to brand Search CPA — if PMax CPA is suspiciously close, you've got cannibalization.
Mistake #4: Thin creative inputs
PMax assembles ads dynamically from the assets you upload — headlines, descriptions, images, videos, logos. Most accounts feed it the minimum (5 headlines, 5 descriptions, 4 images) and wonder why their ads look generic.
The algorithm cannot synthesize variety from a small input set. Fewer inputs = less testable surface area = slower learning = worse final performance.
The fix: Maximum-fill every asset slot:
- 15 headlines (the max). Mix benefit-led, feature-led, social-proof, urgency, and offer.
- 5 long headlines.
- 5 descriptions (the max).
- 20 images. Lifestyle, product, social-proof screenshots, infographic-style. Don't just upload your hero shot in 20 sizes.
- 5 logos.
- At least one short-form video. If you don't have one, Google will auto-generate a slideshow from your images — it'll be worse than what you'd make in 30 minutes with a templated tool.
If you're spending $5K+/month on a PMax campaign and only feeding the minimum, you're knowingly slowing your own algorithm. The creative ROI is enormous.
Mistake #5: Undifferentiated conversion values
PMax can optimize for Maximum Conversion Value or Target ROAS — but only if your conversion data tells it which conversions are worth more. Most accounts track all purchases at value=1, or only at a single dollar value, which means a $50 order and a $500 order look identical to the algorithm.
Result: the algorithm bids equally hard for both. Your high-AOV products get the same effort as your low-AOV ones, which is structurally inefficient.
The fix: Pass dynamic conversion values from your checkout. In GA4 + GTM, fire purchase events with the actual order value as the value parameter. Link GA4 to Google Ads. Switch your bid strategy to Maximum Conversion Value. The algorithm will start prioritizing high-AOV products without you doing anything else.
Bonus: If you have meaningful margin variance across products, pass a custom profit conversion value instead of revenue. Now you're optimizing for margin, not vanity revenue.
We rebuild Performance Max from the ground up.
Asset group structure, signal layering, brand exclusions, creative variety, value-based bidding. Run as a 30-day sprint inside our Google Ads management. From $1,100/month.
See Google Ads service →The PMax build checklist (in order)
- Verify conversion tracking + dynamic values are firing correctly.
- Decide asset group structure (3-5 groups). Each group has a clear product/audience theme.
- Build Customer Match signal lists for each group (purchasers, abandoners, etc.).
- Build one Custom Audience per group (URLs from top-converting pages or query lists).
- Add account-level brand negative keywords + request Google Support brand exclusion.
- Upload maximum creative variety per asset group. Include at least one video.
- Set bid strategy: Max Conversion Value with optional tROAS only if you have ≥30 conv/mo.
- Give it 4-6 weeks of learning. Don't make >20% budget changes during this window.
Frequently asked questions
Is Performance Max worth running in 2026?+
Yes for most e-commerce and most multi-channel-objective accounts. PMax serves Search, Shopping, YouTube, Display, Gmail, and Discover from a single campaign, which is unbeatable for reach efficiency. The catch is it requires real conversion data (≥30 conversions/month at the campaign level) before the algorithm can find a productive equilibrium. Under that threshold, PMax often underperforms a well-structured Search campaign.
Should Performance Max replace Search campaigns?+
No. PMax should complement Search, not replace it. Search campaigns let you control exactly which keywords you bid on and what creative serves; PMax cedes that control to the algorithm in exchange for cross-channel reach. The right structure is: Search for your high-intent keywords (including brand defense), PMax for everything else.
How long does Performance Max take to optimize?+
Plan on 4-6 weeks of learning before judging performance. PMax's bidding model needs that long to gather signal across all six channels it spans. Frequent changes to the campaign during learning extend that window — every meaningful edit (budget changes >20%, target CPA changes, asset group restructure) resets the clock partially.
How do I see what Performance Max is actually doing?+
Open Insights → Asset performance and Insights → Search themes. Those are the two reports that show meaningful data on what the algorithm has decided to serve. PMax intentionally hides per-placement performance, so you cannot see exactly which YouTube videos or display sites you're appearing on. You see categories, not individual placements.
Can Performance Max cannibalize brand search?+
Yes, this is one of the most common mistakes. PMax will preferentially serve on cheap branded queries because they convert at the lowest CPA — making PMax's reported ROAS look great while it's just stealing conversions that brand Search would have captured anyway. Always exclude your brand name and obvious variants from PMax via account-level negative keywords.
Bottom line
PMax rewards structural discipline more than tactical fiddling. Get the asset groups, signals, exclusions, and creative variety right at the start, and the algorithm does the rest. Get them wrong and you can spend months optimizing inside a broken structure.
If you'd rather have a 12-year operator rebuild PMax from the ground up, the Google Ads service includes a PMax restructure as a 30-day sprint.
