Performance Max is Google's most-discussed and most-misunderstood campaign type. It is a black box that runs across Search, Shopping, Display, YouTube, Discover, Maps, and Gmail with one campaign, one set of assets, one audience signal, and one goal. Used well, PMax delivers material incremental performance; used badly, it cannibalizes your Search campaigns and burns budget. This module covers what PMax actually does under the hood, how to set it up so it complements rather than competes with the rest of your account, and the operating playbook for keeping control.
Performance Max (launched globally November 2021) is a goal-based, automated campaign type that uses Google AI to optimize across all of Google's ad inventory simultaneously. Instead of running separate Search, Shopping, Display, YouTube, and Discovery campaigns, PMax runs one campaign and decides per-impression which channel and creative to serve.
Under the hood, PMax is doing three things at once:
Audience modeling. Using your audience signals + Google's first-party data + bidding behavior to identify likely converters in real time.
Creative assembly. Combining your uploaded assets (headlines, descriptions, images, videos, logos) into ad units that fit each placement (Search RSAs, Shopping listings, Display banners, YouTube in-stream, Discover cards).
Bid optimization. Using Smart Bidding (Target ROAS or Maximize Conversion Value, sometimes Target CPA or Maximize Conversions) to bid for impressions likely to deliver conversion value.
The result is a campaign that looks simple at the surface (one campaign, one audience signal, one goal) but is doing extraordinarily complex multi-channel optimization underneath. The operator's job is to give it the right inputs — not to manage its bids or targeting.
It’s not a competition between man and machine, but a collaboration.
— Frederick Vallaeys, Optmyzr co-founder & early Google AdWords evangelist
2. Where PMax inventory is served
PMax serves on all of Google's inventory:
Google Search — PMax can serve on search queries that don't exactly match keywords in any of your Search campaigns. This is where the cannibalization risk lives.
Google Shopping — if you have a feed, PMax surfaces products on Shopping placements. By 2026, PMax has largely absorbed Standard Shopping for most accounts.
YouTube — in-stream and YouTube Shorts.
Google Display Network — banner ads across the Display partner network.
Discover — in-feed Discover ads on the Google app, Chrome new tab, and similar surfaces.
Gmail — promotional inbox tab.
Maps — promoted pins and local listings.
The hidden volume: Most operators assume PMax spend goes mostly to Shopping. In feed-based PMax campaigns it does (60-80% Shopping), but in non-feed accounts most spend goes to Display and Gmail. Always check the channel split in the Insights tab before assuming.
Used well, the leverage is real — these are public, sourced outcomes (ranges, not promises):
+18%
avg more conversions at similar CPA for PMax advertisers
Source: Google
+545%
new-customer conversions for Hairstory via the NCA goal (+31% ROAS)
Source: Google / Hairstory
+76%
advertising revenue for KEH Camera moving Shopping → PMax
Source: KEH / Google
7
Google surfaces one PMax campaign serves at once
RGM analysis
3. Asset groups: the new campaign-structure unit
Before the detail, get the shape in your head. PMax keeps the familiar account hierarchy but swaps the ad group for the asset group. Tap through each level below to see exactly what lives where — and the single rule that matters at each.
PMax doesn't have ad groups. It has asset groups, which are bundles of:
Headlines — up to 15, each up to 30 characters.
Long headlines — up to 5, each up to 90 characters.
Descriptions — up to 5, each up to 90 characters (one allowed to be 60).
Images — up to 20, multiple aspect ratios (1.91:1, 1:1, 4:5, etc.).
Logos — required, 1:1 and 4:1 ratios.
Videos — up to 5; required to serve on YouTube. Google auto-generates if not provided (often badly).
Audience signals — the "suggestions" you give the algorithm.
Final URLs / page feeds — where clicks land.
Listing groups — product feed segmentation (Shopping/feed PMax only).
How to structure asset groups
Each asset group represents a distinct theme of creative + audience. Common patterns:
One asset group per product category. Distinct copy, images, and audience signals per product line. Best for multi-category brands.
One asset group per audience tier. Different creative for prospecting vs retargeting vs high-value customer lookalikes.
One asset group per offer / promotion. When you have meaningful promo differences (e.g., Sale, New Arrivals, Best Sellers).
Asset group reporting limitation: Until recently you could not see asset-group-level conversion reporting in PMax (just account-level). Google has been rolling out asset-group conversion reporting since late 2023; verify your account has access. Without it, asset group structure is largely a creative organization device, not a performance measurement unit.
4. Audience signals: what they actually do
Audience signals in PMax are suggestions, not targeting. They tell Google's model where to start looking; the model then expands based on conversion data. This is the biggest source of operator confusion.
What you can include as audience signals:
Custom Segments — intent-based audiences built from keywords, URLs visited, or app installs.
Your Data — Customer Match lists, remarketing lists, app users, GA4-imported audiences.
Google's model starts impression-serving with users matching the audience signals as a baseline.
As conversions accumulate, the model identifies common traits of converters within and beyond the audience signal.
The model expands serving to look-alike-shaped users beyond the original signal.
Over time, audience signals matter less than conversion patterns. Mature PMax campaigns increasingly serve outside the original signal.
Strongest audience signals: Customer Match (your converters, your high-LTV customers, your repeat purchasers) outperform interest-based audiences as signals. The model learns "people like these" faster than "people interested in this category." Always include Customer Match if you have it.
RGM Expert Trick
We seed signals into PMax even though it ‘doesn’t need them’
Audience signals don’t restrict PMax — but they do give it a warm start. Hand it your first-party lists and custom segments and the model finds the right people days sooner.
We treat signals as a head start, not a constraint, and feed our best converters in on day one instead of letting it grope around blind.
WHY IT’S RARE · ‘It ignores signals anyway’ is true for targeting, wrong for speed.
5. Listing groups for shopping inventory
If your PMax campaign has a Google Merchant Center product feed, you can segment products into listing groups within an asset group. Listing groups let you exclude products, adjust prioritization, or organize feed inventory by attribute (brand, product type, custom labels).
Custom labels: the underused lever
Custom labels are 5 fields on your Merchant Center product feed (custom_label_0 through custom_label_4) where you can put anything you want. Common uses:
Stock level (full stock, low stock, last chance) — reduce spend on low-stock items to avoid disappointing customers.
Best seller status — allocate more budget to proven winners.
New arrival status — prioritize discovery campaigns toward new SKUs.
6. Brand exclusions, brand inclusions, and the cannibalization problem
PMax's biggest historical complaint from operators: it serves on brand search queries and takes credit for conversions that would have come for free through organic brand search. Google added brand controls (and brand search themes) over 2023-2024 to address this.
Brand list and brand exclusions
Build a Brand List in Google Ads Shared Library (Tools & Settings > Brand Lists). Add your brand name and variations. Then exclude this brand list from your PMax campaign. PMax will not serve on those branded queries.
The result: PMax becomes a non-brand-only campaign, comparable to your non-brand Search campaign. Conversion attribution becomes interpretable.
Brand inclusions (for brand PMax campaigns)
If you want a dedicated PMax campaign for branded shopping or branded search, use brand inclusions to only serve on brand-listed queries. Most operators do not need this — a separate Search Brand campaign is usually cleaner.
RGM Expert Trick
We exclude brand from PMax, or it takes credit for your cheapest sales
Left alone, PMax happily serves on your brand queries, scoops up conversions that would have come for pennies anyway, and posts a gorgeous ROAS that isn’t incremental.
We apply brand exclusions and judge PMax on the non-brand growth it actually creates — the only number that proves it’s earning its budget.
WHY IT’S RARE · The inflated ROAS is so flattering that most teams never question it.
7. Search themes
Search themes (launched 2023, expanded 2024-2026) let you add keyword-like suggestions to PMax that influence which Search queries the campaign serves on. Think of them as "PMax keywords" — but loose suggestions, not exact-match controls.
How search themes work
You add up to 25 search themes per asset group.
Google treats them like phrase-match keywords — the campaign can serve on queries related to the themes.
Existing exact-match keywords in your Search campaigns still take precedence over PMax for queries that are exact matches.
Search themes do not restrict PMax to only those themes; PMax can still serve on other queries the model finds.
When to use search themes: Add search themes for high-intent product categories you want PMax to cover that aren't obvious from your feed or assets (e.g., your feed has "running shoes" but you also want PMax to serve on "workout gear" queries). For most accounts, search themes are nice-to-have, not essential.
8. The PMax + Search hierarchy — how Google prioritizes
The most-common operator question: "If I have both a Search campaign and a PMax campaign, which one serves on what queries?"
Google's priority logic (as documented in their official help):
Exact-match keyword in Search campaign — wins. If the query matches an exact-match keyword in any active Search campaign, PMax does not serve on it.
Other Search match types (phrase, broad) — Ad Rank determines winner. PMax can win if its Ad Rank is higher than the Search campaign's.
No Search campaign keyword match — PMax can serve if the model decides the query is a good fit.
Brand queries in PMax brand list with brand exclusion enabled — PMax does not serve.
This means your exact-match Search campaign is protected from PMax cannibalization for matched queries. But phrase-match and broad-match Search campaigns compete with PMax in the same auction.
The practical structure
The 2026 dominant structure:
Brand Search campaign (exact match): explicit brand-term protection. PMax with brand exclusion does not interfere.
Non-Brand Search campaign (broad + phrase + exact): catches obvious high-intent queries. Use match-type + ad-strength to compete with PMax in the auction.
Performance Max (with brand exclusion): catches everything else — Shopping, YouTube, Display, Discover, the long-tail of Search queries not in your Search campaigns.
9. Conversion goals and value-based bidding
PMax is designed for value-based bidding. Without conversion values, PMax has limited ability to optimize toward profitable customers.
Required conversion setup
Conversion actions defined with values. For e-commerce: revenue per purchase. For lead gen: modeled value per lead (use Value Rules if you don't have transaction-level values).
Enhanced Conversions for Web (or Conversions API). Non-negotiable. PMax sub-optimizes when iOS conversions are missing from its signal.
Conversion goal selection. Choose Account Default or Campaign-Specific Goals. Most accounts run PMax on the same conversion goals as Search.
Bid strategy choice
Maximize Conversion Value (no target ROAS) — for accounts in early learning or with budget you want to fully spend.
Maximize Conversion Value with Target ROAS — once you have conversion data and a known healthy ROAS, set a target ROAS at or slightly below your blended target.
Maximize Conversions — for accounts without conversion values (lead gen without value modeling).
Maximize Conversions with Target CPA — with a target CPA constraint, same caveat as without values.
10. New customer acquisition (NCA) goal
PMax includes a New Customer Acquisition goal you can layer on top of any bid strategy. Two modes:
New customer value mode — you tell Google how much more a new customer is worth than a repeat customer. PMax weights bidding toward new customers in the same campaign.
New customer only mode — PMax only optimizes toward new customer conversions, treating repeat customer conversions as $0.
NCA requires Customer Match list uploads (your existing customer list, refreshed weekly) so Google can distinguish new from repeat. Without an up-to-date list, NCA misclassifies customers.
When NCA is worth using: When repeat-customer attribution is muddying your ROAS — you have lots of brand-loyal customers who would have bought anyway, and you want the campaign to focus budget on net-new acquisition. Most growth-stage brands benefit from NCA in PMax.
10b. The PMax launch checklist — creative, and the stack
PMax rewards strong inputs and punishes weak ones. Run this order before you turn it on, and use the readiness planner to pressure-test it first.
Fix conversion tracking + values
Value-based bidding is the whole point; counts-only starves it. See Module 6.
Apply brand exclusions
Stop PMax harvesting cheap brand searches and flattering its own numbers; keep brand in a dedicated Search campaign.
Add account-level negatives
Keep spend off junk queries and competitor/irrelevant terms.
Seed strong audience signals
Lead with first-party data — your converters and Customer Match. Signals guide, they don’t restrict.
Build a deep, varied asset group
Multiple headlines and long headlines, all image ratios (landscape, square, portrait), logos, and video — if you skip video, Google auto-generates a weak one.
Set the goal — and NCA if you want growth
tROAS / Max Conversion Value on real values; add the New Customer Acquisition goal to bid up for new customers.
Hold for the learning period, then read it
Give it ~2–6 weeks and enough conversions before judging. Use the Insights tab and asset reporting to prune and refresh.
Creative & asset best practices
Feed the full ratio set. Landscape (1.91:1), square (1:1) and portrait (4:5) images, plus logos in square and landscape — missing ratios shrink where you can serve.
Always supply video. Even a simple 10–15s cut beats Google’s auto-generated one and unlocks YouTube inventory properly.
Give range, not repetition. Cover distinct angles — benefit, offer, proof, urgency — so the model has real combinations to test (the same discipline as RSAs).
Watch asset performance ratings (Low / Good / Best) and swap Low assets — but give each enough impressions before judging.
Recommended tools & tech
Feed (e-commerce)
Google Merchant Center
Clean product feed = better Shopping/PMax coverage and titles.
Feed optimization
Feed tool (e.g. Feedonomics)
Fix titles, GTINs and attributes that drive matching.
Verdict + guardrails + asset-group structure before launch.
11. Asset reporting and creative optimization
PMax shows asset-level performance ratings (Best, Good, Low) within each asset group. The ratings are based on relative serving and engagement performance:
Best — Google's model serves this asset frequently and it performs well.
Good — Average performance; the asset gets serving.
Low — Below-average; replace or improve.
Creative refresh cadence
Replace "Low" rated assets monthly.
Add 3-5 new assets per asset group monthly to give the model fresh test material.
Replace videos quarterly (videos have the slowest refresh cycle but the largest impact when refreshed).
RGM Expert Trick
We pull PMax’s hidden placement data with a script
The interface won’t show you where PMax spends — how much went to Display, YouTube, or Search. So we run a script that surfaces the channel and placement report it hides.
Once you can see a third of the budget drifting into junk Display placements, you can finally do something about it.
WHY IT’S RARE · You can’t optimize a black box you’re not allowed to look inside.
12. Insights tab
The Insights tab in PMax shows:
Search categories — what query types PMax is serving on (high-level categories, not exact queries).
Audience insights — which audience segments are converting.
Asset performance — asset ratings (see above).
Channel split — how spend is distributed across Search, Shopping, Display, YouTube, etc.
Demand forecasts — predicted seasonality.
Search categories are particularly important — you can use them to identify what queries PMax is serving on and decide whether to add Brand exclusions, search-theme refinements, or move queries to a Search campaign with exact-match keywords.
PMax performance only means something against your own category. Efficiency that is excellent for travel would be alarming for apparel. Three of the most common PMax verticals, with public benchmark ranges — then go deep on yours.
13. When to use PMax vs Standard Shopping vs Search
Use case
Recommendation
E-commerce with strong feed
PMax as primary; Standard Shopping for experiments / niche bid control
E-commerce with no feed
Search + Display + Demand Gen separately; PMax only after feed exists
Lead generation (B2B)
Search primary; PMax as secondary for upper-funnel discovery if budget supports
Local services (LSAs eligible)
LSAs + Search; PMax with caution and brand exclusion
App install
App campaigns, not PMax
Awareness / video
Demand Gen + Video; PMax not optimized for top-funnel
Small budget (under $5K/month)
Search only; PMax needs volume to learn
Before you launch, run your setup through the planner — it checks readiness, lists the guardrails to apply, and recommends an asset-group structure for your business.
14. The 12 most common PMax mistakes
No brand exclusion. PMax serves on branded queries and takes credit for organic-attributable conversions. Fix: build Brand List in Shared Library; apply as exclusion.
Insufficient creative assets. 3 headlines, 1 image, no video. PMax can't test variants. Fix: max out at 15 headlines / 5 long headlines / 5 descriptions / 20 images / multiple aspect ratios / at least 1 video.
No video assets uploaded. Google auto-generates videos — often using your text on stock footage that looks bad. Fix: upload at least one real video per asset group.
Wrong conversion architecture. No conversion values, no Enhanced Conversions, no NCA. Fix: set up value-based conversion tracking before launching PMax.
Asset group structure mirrors old campaign structure. Trying to recreate granular Shopping groups as asset groups. Fix: think in terms of audience + creative themes, not SKU clusters.
No audience signals. Letting the model find audiences from zero. Slows learning. Fix: provide Customer Match + interest-based suggestions as starting points.
Daily budget too small. Below $50/day in most categories, PMax can't test inventory effectively. Fix: minimum $50-100/day; scale based on conversion volume targets.
Constant edits. Editing PMax every few days kicks it back into Learning Phase. Fix: make changes in batches; let each batch run 7-14 days.
Running multiple PMax campaigns competing. Two PMax campaigns with same goal in same account compete in auction. Fix: consolidate to 1-2 PMax campaigns.
Stale Customer Match lists. NCA misclassifies because the list is months old. Fix: weekly customer list upload (or automated sync).
Ignoring Insights tab. Treating PMax as set-and-forget. Fix: weekly review of search categories, channel split, asset performance.
Wrong bid strategy. Maximize Conversions for an e-commerce account with rich conversion values (instead of Maximize Conversion Value). Fix: value-based bid strategies for value-based businesses.
15. Anti-patterns: what NOT to do
Do not use PMax as your only campaign. Always pair with at least Brand Search to protect branded query attribution.
Do not run PMax without conversion tracking validated. Garbage conversions = PMax optimizes for garbage.
Do not split products into many small PMax campaigns. The algorithm needs volume per campaign to learn.
Do not bid manually inside PMax. Not possible — PMax is automation-only. If you need manual bid control, use Search + Standard Shopping.
Do not panic on early performance. PMax needs 2-4 weeks to escape Learning Phase. Early CPA / ROAS is unstable.
Do not exclude all of Display / YouTube via no assets. Some operators upload only Shopping inputs hoping to suppress display serving. The model still serves on display; you just give it nothing to work with.
Do not skip account-level negative keywords. PMax respects account-level negatives. Use this to block irrelevant queries from PMax just as you would from Search.
16. The operating cadence
Daily (5-10 minutes)
Check budget pacing (PMax can spend aggressively on high-CPA days).
Check for new conversions and any large spend anomalies.