Performance Max is now a core campaign type for ecommerce accounts. It’s not optional for ecommerce anymore. But most stores running PMax are getting two things wrong: how they structure asset groups and how they handle creative. These are the two areas where operator decisions actually matter in PMax, given how little control Google gives you elsewhere.
This article covers what asset groups are, the architectural distinction most guides miss, how to decide between margin-tier and product-theme structuring, the 2026 platform changes that quietly changed the operator workflow, and the common mistakes that flatten campaign performance across the catalogue. It assumes you’ve already run PMax and know the basics. If you’re setting up your first PMax campaign, start with the Performance Max campaign structure guide for ecommerce first, and if you’re on Shopify, the Shopify Google Ads & Shopping setup guide covers feed and product-data setup first.
What asset groups actually are
An asset group in PMax is a collection of creative assets (images, videos, headlines, descriptions, logos), audience signals, and search themes. Google combines these inputs dynamically to serve ads across Search, Shopping, YouTube, Display, Discover, Gmail, and Maps. You can have up to 100 asset groups per campaign. Asset groups cannot be shared between campaigns.
The structural difference from Search ad groups matters here. In Search, ad groups are organised around keywords, and your ads serve only against those keywords. In PMax, asset groups are organised around creative themes and audience signals, but the algorithm decides which placements, channels, and queries to serve against. You’re providing inputs. Google decides outputs.
This shifts the operator job. In Search, structure decisions are about query control. In PMax, structure decisions are about creative relevance and signal quality. Get that distinction right and the rest of the structure question clarifies fast.
The architectural distinction most PMax guides miss
Most PMax guides conflate two different levels of segmentation: campaign segmentation and asset group segmentation. They’re not the same decision.
Campaign-level segmentation operates on budget and ROAS targets. Different campaigns can have different bid strategies, different budget caps, and different aggressiveness profiles. This is where margin-tier decisions live. A 60% gross margin product and a 10% gross margin product need different ROAS targets and different budget priorities. Putting them in the same campaign means the algorithm optimises for whichever converts easiest, which is almost never the high-margin product.
Asset group segmentation operates on creative relevance and audience signal quality. Asset groups within a single campaign share the same budget pool and ROAS target. What changes between them is the creative theme, audience signals, and search themes. This is where product-line and product-theme decisions live.
Get this distinction wrong and you’ll either end up with one giant campaign that the algorithm starves of strategic guidance, or with too many fragmented campaigns that don’t have enough conversion volume each to optimise. Get it right and structure clarifies into two separate questions that have different answers based on different inputs.
Why audience-signal-as-structure doesn’t work in PMax
The biggest PMax structuring mistake is treating audience signals like Display campaign audience targeting. They’re not the same thing.
In Display campaigns, audience targeting is a hard constraint. Set audience targeting and your ads serve only to that audience. In PMax, audience signals are hints to the algorithm. They tell Google “people like this tend to convert for us.” The algorithm uses signals as starting points and then expands aggressively based on its own conversion modelling. Your asset group with a “site visitors” audience signal will serve to plenty of people who never visited your site, if the algorithm thinks they’ll convert.
This is why structuring asset groups around audience signals doesn’t do what most operators expect it to. Creating one asset group for “retargeting site visitors,” another for “lookalike of purchasers,” another for “in-market shoppers” assumes those groups will serve to different audiences. They won’t. They’ll all serve to whoever Google’s algorithm thinks is convertible, with the audience signal acting as a soft starting bias.
There’s a related mistake. Some operators duplicate identical asset groups across different audience signals, hoping to compare audience performance. After about 30 days, performance across the duplicate groups converges to the same average. The algorithm collapses the distinction because the signals aren’t enforced as hard bounds.
Audience signals belong inside asset groups as inputs that match the group’s theme. They don’t belong as the segmentation principle itself.
Campaign-level structure: margin-tier segmentation
For ecommerce accounts with margin variance across the catalogue and enough conversion volume to support segmentation, margin-tier campaign structure is the most consequential architectural decision you’ll make. Getting that structure right is a large part of Shopping and Performance Max management.
The pattern most stores benefit from:
A Hero campaign for the top 20% of products by margin contribution. These get aggressive tROAS targets and a disproportionate share of budget. The logic is straightforward: every dollar spent here has higher profitability than the catalogue average, so the campaign should be allowed to run with lower efficiency expectations.
A Standard campaign for the next 60% of the catalogue. Moderate ROAS targets, balanced budget. This is the workhorse campaign carrying most of the catalogue weight.
A Clearance or low-margin campaign for the bottom 20%. Tighter ROAS targets, smaller budget. The goal here isn’t growth; it’s preventing the algorithm from spending freely on products that don’t justify the ad cost.
Implementation requires custom_label_0 mapping in your product feed. Each SKU gets tagged with its tier (Hero / Standard / Clearance), and the PMax campaign uses listing group filters to scope to the right tier.
The conversion volume gate matters. Below 30-50 conversions per month, you don’t have enough data for one campaign to optimise reliably, let alone three. The rule of thumb: 30-50 conversions per month = one or two campaigns; 100+ conversions per month = start margin-tier segmentation. Below that threshold, run a single campaign with broad coverage and revisit segmentation when volume supports it.
Margin-tier segmentation only works if your feed has clean margin data. If you don’t know your gross margin per SKU (or if your COGS data is unreliable), margin-tier structure is premature. Build the data discipline first.
Asset group-level structure: product theme segmentation
Within each campaign, asset groups should segment by product theme for creative relevance. Different headlines, different images, different descriptions, calibrated to the products the asset group sells.
A clothing store might have asset groups for menswear, womenswear, and accessories. A furniture store might have asset groups for living room, bedroom, and outdoor. A consumer electronics store might have asset groups for laptops, audio, and accessories. The theme is whatever creative grouping makes sense given the catalogue and the placements PMax serves to.
The reason for this segmentation is the combination logic. PMax dynamically combines headlines, descriptions, and images at serve time. If you dump all creative into one asset group, Google will try to combine “premium leather handbag” with an image of a kitchen knife. That doesn’t happen literally because Google is smarter than that, but you’d be surprised how often suboptimal pairings serve when creative is too broad.
Asset group themes also feed the audience signal and search theme logic. A menswear asset group has different audience signals (men’s style interests, men’s site visitor segments) and different search themes (“men’s leather jackets,” “smart casual menswear”) than a womenswear asset group. The segmentation lets you provide more targeted inputs to the algorithm.
How many asset groups should you run
The industry consensus has converged on 2-4 asset groups to start, with expansion based on performance data. This is more disciplined than the “more is better” approach some operators take.
The reason for the cap: each asset group needs sufficient conversion volume per month to optimise. Industry guidance puts this at around 20 conversions per asset group per month minimum. Below that, the group operates in semi-permanent learning mode and the algorithm makes bid decisions without enough signal.
Math this out for a typical account. A campaign generating 100 conversions per month can support 4-5 asset groups at the 20-per-group floor. A campaign generating 40 conversions per month supports 2 asset groups at most. A campaign generating 20 conversions per month should run one asset group.
The expansion pattern: launch with 2-4 focused asset groups by primary product theme. Run for 6-8 weeks. Identify which themes are pulling weight and which aren’t. Either consolidate weak groups into stronger ones or add a new asset group when conversion volume in an existing group exceeds 40-50 per month (giving you headroom to split it).
The most common over-segmentation mistake is creating 6-10 asset groups at launch because the catalogue has 6-10 categories. The categories exist in your business; that doesn’t mean PMax has enough conversion data to optimise each one as a separate creative theme.
How budget shapes PMax channel behaviour
PMax doesn’t distribute budget evenly across Search, Shopping, Display, YouTube, Discover, Gmail, and Maps. At small budgets, PMax exhibits a heavy bias toward Search and Shopping placements. The algorithm gravitates to the highest-intent, lowest-CPC channels first.
Budget is also the lever I use to control effective ROAS, which is why the bid strategy and the budget decision have to be made together rather than in isolation. The interaction between your bid strategy choice and budget pacing is covered in full in the Performance Max bidding guide for ecommerce, including why I avoid Target ROAS and lean on budget pacing instead.
At lower budget levels, PMax is effectively running as a Search and Shopping wrapper. You’re not getting meaningful Display or YouTube exposure because the algorithm hasn’t been given enough budget to expand beyond the high-intent channels.
The 2026 channel-level performance reporting improvement makes this visible. You can now see budget distribution across Search, Shopping, Display, and YouTube within a PMax campaign. Audit this on every account in the first month after launch. If your $2,500/month PMax campaign is showing 85% Search and Shopping spend, that’s expected behaviour. Don’t try to force display by uploading more video assets; the algorithm needs more total budget before it expands.
The implication at lower budget levels: PMax is competing for the same query-level inventory as your Search campaigns and Shopping campaigns. If you’re running Search + Shopping alongside PMax at a small budget, you’ll see cannibalisation. The decision becomes: consolidate into PMax only, or keep Search/Shopping separate and accept some overlap.
At larger budgets ($10,000+/month per campaign), the channel distribution shifts. Display and YouTube get meaningful share. You start seeing PMax’s full multi-channel behaviour. That’s when the architectural decisions in this article have their biggest impact.
PMax for lead-gen: the offline conversion gate
A hard position before going further: don’t run PMax for lead generation unless you have offline conversion tracking implementing qualified-lead data back to Google Ads. Without that infrastructure, PMax for lead-gen produces spam form fills, optimised aggressively.
The gap is striking. A widely cited case study showed PMax delivering a 5% qualified lead rate when running against form-fill goals only, compared to 25% qualified lead rate for the same business running Search campaigns. The difference: the Search campaigns had higher-intent inbound traffic; the PMax campaign optimised for whatever form fills were cheapest, including substantial volumes of accidental clicks, job-seekers, and competitor research traffic.
The infrastructure required:
GCLID capture in your CRM. When a lead submits a form, the Google Ads click ID needs to be stored against the lead record. This is the foundation that lets you tie back qualified outcomes to the click that produced them.
Qualified-lead export back to Google Ads. When a lead becomes qualified (an MQL or SQL definition specific to your business), the qualified GCLIDs get exported back to Google Ads as offline conversions. This can be done via CSV upload, Google Ads API, or native CRM connectors (HubSpot, Salesforce, Pipedrive, and others have these).
Conversion windows of 90+ days for B2B. Lead qualification cycles for B2B are long. The default 30-day window cuts off too much downstream signal.
Patience for optimisation. PMax for B2B lead gen takes 8-10 weeks to optimise (vs 6 weeks for ecommerce) once qualified-lead data is flowing back. Most accounts kill PMax in week 3 because metrics look bad, never giving the algorithm a chance to learn from quality data.
If your business can’t or won’t implement this infrastructure, run Search campaigns instead. PMax is the wrong tool for lead-gen without it.
Asset requirements: 2026 specs
Performance Max asset requirements have stabilised over the last 18 months, but the gap between minimum requirements and recommended supply has widened. Most underperforming campaigns hit the minimums and stop there. The algorithm needs more creative surface than the minimums provide.
Headlines: minimum 3 per asset group, maximum 15. Maximum length 30 characters per headline. At least one headline must be 15 characters or shorter (Google uses this short variant in compact placements). Most operators fill all 15 slots; if you have the copy bandwidth, do.
Long headlines: minimum 1, maximum 5. Maximum length 90 characters. Used in larger format placements.
Descriptions: minimum 1, maximum 5. Maximum length 90 characters. At least one description should be shorter than 60 characters.
Images: minimum 1 landscape (1.91:1 aspect ratio, recommended 1200×628), 1 square (1:1, recommended 1200×1200), and 1 portrait (4:5, recommended 960×1200). Recommended supply: 15-20 total images across the three aspect ratios. The portrait orientation is critical for YouTube Shorts and vertical mobile placements.
Logos: minimum 1 square logo (1:1, recommended 1200×1200). Optional landscape logo (4:1, recommended 1200×300).
Business name: required, set at the campaign level (not per asset group).
Display path URLs: up to 2, optional but recommended for cleaner display.
Final URL: required, must be the conversion path relevant to the asset group’s theme.
The recommendation is to populate maximum asset coverage across all types and dimensions. Campaigns running at minimums give Google too little creative surface to optimise effectively. Even one or two additional images can move asset performance ratings from Low to Good.
Video specs and the audio question
Video is no longer optional for competitive PMax campaigns. Internal Google data shows campaigns with custom video assets deliver around 12% more conversions on average compared to those relying on auto-generated video. Other industry sources put the gap at 25-40% depending on category.
Specifications: minimum 10 seconds, recommended 15-60 seconds. Three aspect ratios required for full coverage: 16:9 (horizontal, traditional YouTube), 1:1 (square, social feeds), and 9:16 (vertical, YouTube Shorts and mobile-first placements). Up to 5 videos per aspect ratio.
If you don’t supply video, Google auto-generates one from your images, logos, and text. The auto-generated version stitches static images with generic transitions and synthesised text overlays. It serves on YouTube placements but consistently underperforms custom video. Most ecommerce stores can shoot a basic 15-second product overview on a smartphone that outperforms Google’s auto-generation.
The audio question. As of 2026, Google now auto-generates AI voice-overs for PMax videos that lack an audio track. This changed the calculus around silent video creative. The next section covers this in detail.
For now: if you’re producing custom video for PMax, decide upfront whether to include voice or leave it silent. That decision now interacts with Google’s video enhancement controls, which is a new operator workflow consideration.
The 2026 AI voice-over auto-enable
This is the most operationally significant 2026 PMax change. Google launched automatic AI voice-over generation for PMax video ads on March 20, 2026. The feature is opt-out only, not opt-in. It’s enabled by default on every PMax campaign with video enhancement control turned on.
What it does: for any PMax video asset that doesn’t already have an audio track, Google’s AI generates synthesised voice narration. The voice script comes from advertiser-provided headlines and descriptions. The narration is layered onto the existing video and saved as a new video asset, which then becomes eligible to serve.
What it doesn’t do: provide a preview function. Advertisers cannot hear the AI voice-over before it serves to users. There’s no review step.
The brand safety implications are direct. The synthesised voice is generated from Google’s text-to-speech models. It may not match the brand’s intended tone, accent preferences, or communication style. For brands with strict voice guidelines (think premium retailers, regulated industries, or brands with distinct founder-voice marketing), an AI narrator reading the ad copy with a generic Google voice is a brand consistency problem.
The operator workflow this creates:
Audit every client account’s video enhancement control settings. The setting is found under Campaign Settings → Video enhancements. Confirm whether the campaign is opted in (default) or out.
Default to opt-out unless the client has explicitly approved AI voice-over for their brand. This is part of the broader auto-generated assets permission workflow (covered next). Until the client has heard a sample AI voice on their copy and approved its use, opt out.
For clients who approve AI voice-over: document the approval, the date, and the specific brand guidelines that govern when it can and can’t be used. Some clients might approve AI voice for product-feature explainers but not for promotional offers. Capture that nuance in account documentation.
Re-audit settings quarterly. Google’s defaults can change, and accounts can get migrated between settings during feature rollouts. The video enhancement control state isn’t static.
Auto-generated assets: the client-permission workflow
The 2026 platform has three layers of automated asset generation, each with different brand risk profiles:
AI text expansion. Google can generate additional headlines and descriptions from your provided copy. Lowest brand risk because the AI is working from advertiser-written source material. Some clients are comfortable with this layer because the brand voice is preserved through the source text.
AI-generated images. Asset Studio (powered by Imagen 4) can generate product imagery from text prompts. Higher brand risk because AI-generated imagery may not match brand visual standards, may show inaccurate product details, or may include generic stock-feeling visuals.
AI voice-over (covered above). The highest brand risk because voice is a strong brand identity signal, and there’s no preview function before serving.
The Adgenix protocol for handling this:
Default state on every new account: all three layers off. Don’t enable until explicitly approved by the client.
When introducing the question to the client, frame each layer separately. Some clients are fine with text expansion but not voice. Others want full automation. Don’t conflate the three into a single “AI generation: yes/no” question.
Document client decisions in account notes with date and scope. “Client approved AI text expansion for product feature copy. Voice-over off. Image generation off until specific product imagery brief is signed off.” That level of detail makes future audits and team handovers cleaner.
Review the documented decision annually or whenever campaign objectives change. A client comfortable with AI text expansion for ecommerce campaigns may not want it on a brand campaign. Position changes warrant re-confirming.
This protocol slows campaign launch by maybe 10-15 minutes per account for the conversation and documentation. The slowdown is worth it. Brand inconsistencies from auto-generated assets are the kind of operational mistake that ends client relationships.
Asset Studio: when to use it, when to commission custom
Google’s Asset Studio, powered by Imagen 4 (images) and Veo (video), gives advertisers in-platform AI creative generation. It’s distinct from auto-generated assets in two ways. First, you control the input prompts. Second, you can preview the output before adding it to a campaign.
This makes Asset Studio a different proposition from the auto-generated assets discussed above. It’s still AI-generated creative, but with operator oversight and editability.
When Asset Studio fits:
Client has approved AI-generated creative in general (per the workflow above), and a campaign needs more creative volume than the in-house team can produce in the available timeline. Asset Studio can fill the gap with on-brand prompts.
Testing creative concepts before commissioning custom production. A 15-minute Asset Studio session can produce 10 image variants exploring different visual angles. Use them to test which concepts get Best ratings, then commission custom production for the winning concept.
Adding variation to existing custom creative. If you have 5 strong custom images, Asset Studio variants can extend creative diversity without diluting the brand.
When Asset Studio doesn’t fit:
Client requires brand-specific imagery (specific products, real founders, real customers, brand-defined art direction). Asset Studio can’t replicate brand-specific elements reliably.
Regulated industries where claims, comparisons, or visual representations need legal review. The AI may generate content that creates compliance risk.
Brands with strong design systems where every image goes through a creative director. Asset Studio bypasses that workflow.
The decision tree: would your client’s creative director approve this asset if it came from a freelance designer? If yes, Asset Studio is probably fine for a test. If no, commission custom or stick to existing approved creative.
Audience signals: how to actually use them
Now that the structural decisions are out of the way, audience signals are the input layer that sits inside asset groups. They’re hints to the algorithm, not targeting bounds (covered earlier).
The signal types worth using, in rough order of impact:
Customer Match lists. Upload buyer email addresses (hashed by Google for privacy). The algorithm uses these as the highest-quality signal of who actually converts for your business. As of 2026, Customer Match no longer requires a minimum list size, which is significant for small accounts that previously couldn’t use this signal.
Website visitor remarketing. Cart abandoners, product page viewers, past purchasers. Reasonable secondary signal. Best when segmented by intent stage rather than dumped together as “all visitors.”
In-market segments. Google’s pre-built audiences of people actively shopping in defined categories. Useful but less differentiating than first-party data. Best as a supporting signal, not the primary.
Custom segments. Built from search terms or competitor websites. Useful in concept but requires testing to find the right combinations. Worth the operator time for higher-budget accounts; not the first place to invest signal-building effort.
Affinity audiences and demographics. Lowest signal quality. Mostly useful as filters in combination with stronger signals rather than as primary signals.
Quality matters more than quantity. One Customer Match list of verified buyers beats a dozen broad in-market segments. The algorithm is using your signals as starting points; better starting points mean faster optimisation.
Per-asset-group signal customisation matters. Each asset group should have signals appropriate to that group’s theme. Don’t reuse the same signal set across every asset group; that defeats the purpose of asset group segmentation by collapsing all groups back to the same signal input.
Search themes within asset groups: pragmatic position
Search themes are keyword-like inputs at the asset group level. They tell the algorithm which search queries to consider serving against. As of 2025-2026, the limit per asset group is 50 (up from 25 originally).
Google confirmed in 2025 that search themes take priority over the algorithm’s own search signals when the two are in tension. A well-chosen search theme can actively redirect budget away from default paths the algorithm would otherwise take. Search themes behave like phrase or broad match in the keyword prioritisation hierarchy, not like exact match. They’re a directional signal, not a constraint.
The honest position: search themes have no proven long-term performance lift in PMax. They help short-term learning by directing the algorithm toward relevant query territory faster. Whether that short-term lift compounds into long-term performance gain is unclear from the available evidence.
The pragmatic operator approach:
Use search themes that match clear commercial intent for the asset group’s product theme. A footwear asset group should have search themes like “men’s leather boots Melbourne,” “comfortable work shoes,” “running shoes brand.” Not “footwear” alone (too broad to be useful as a signal).
Refresh quarterly based on the Search Categories Report. Themes that aren’t translating into the right search territory should be replaced.
Don’t fill all 50 slots reflexively. 20-30 high-quality themes outperform 50 mixed-quality themes. The signal quality matters more than slot coverage.
If you’re running PMax in a brand-protected setup (Search campaigns with exact match keywords protecting high-value queries), search themes can complement that without conflicting. The Search exact match keywords take priority for those specific queries; PMax search themes direct PMax to nearby queries.
Creative quality and asset performance scoring
Google rates each asset with one of three labels: Best, Good, or Low. Ratings are based on click-through rates, conversion rates, and combination performance with other assets in the group.
Low-rated assets drag down combinations they appear in. Google’s serving logic still uses them, but combinations including Low assets get downweighted relative to combinations of Best and Good assets. A campaign with 5 Best assets, 5 Good assets, and 5 Low assets is performing worse than the same campaign with the Low assets removed.
The operator workflow for asset quality:
Review asset performance ratings weekly in the first month after launch. Some sources recommend monthly; weekly is more aggressive and surfaces issues sooner.
Replace Low-rated assets within 2-4 weeks of the rating appearing. Don’t wait for “more data.” Google has already concluded those assets aren’t performing.
Patterns associated with Best ratings: specific outcomes rather than generic value claims (“Save 40% on annual energy costs” beats “Save money”); numbers and data where relevant; clear calls to action; problem-acknowledgment framing.
Patterns associated with Low ratings: stock images that look generic; feature-focused copy that doesn’t connect to outcomes; long descriptions that bury the value proposition; images that don’t render well across multiple aspect ratios.
The single biggest driver of creative quality is asset variety. PMax dynamically combines headlines with images with descriptions. The more variety you provide, the more combinations Google can test. A campaign with 15 headlines, 5 descriptions, and 15 images produces vastly more combinations than one with 5 headlines, 1 description, and 3 images. Creative testing on Facebook and Instagram is covered on our Meta ads specialist in Melbourne page.
Common asset group mistakes
A consolidated list of the failure patterns that show up across audit work for Melbourne ecommerce accounts:
One asset group with everything. The most common mistake. All headlines, all images, all products dumped together. Google can’t match the right creative to the right context because there’s no creative differentiation by theme.
Identical asset groups duplicated for audience signals. Operators creating “asset group A: retargeting” and “asset group B: lookalike” with the same creative in each, hoping to compare. Performance converges in 30 days because audience signals aren’t enforced as bounds.
Minimum-only asset supply. Asset groups with 3 headlines, 1 description, 1 landscape image, 1 square image. Google has nothing to combine. Asset performance ratings end up uniformly Low because there isn’t enough variety to find the strong combinations.
Stock images. Pulling stock photography off image libraries because custom creative is constrained. Stock images consistently accumulate Low ratings. Even basic smartphone-shot product imagery outperforms stock.
All headlines at maximum 30 characters. Missing the short-headline requirement. PMax has compact placements that need ≤15 character headlines. Without a short variant, your ads don’t serve in those placements.
No custom video. Letting Google auto-generate video. Auto-generated video underperforms custom by 12-40%. Even a basic 15-second product overview beats it.
Audience signals reused identically across all asset groups. Collapses asset group segmentation back to a single effective signal input.
Not auditing Video Enhancement Control after March 2026. AI voice-over auto-enabled on most accounts. Operators who didn’t audit are serving AI-narrated videos without client approval.
Asset groups for low-volume campaigns. Creating 4 asset groups for a campaign generating 20 conversions per month. Each group sits in semi-permanent learning mode. One asset group would have outperformed.
GMC disapprovals affecting campaign delivery. If your Merchant Center has product disapprovals, those products don’t serve in PMax Shopping placements regardless of asset group structure. The Google Merchant Center disapproval troubleshooting guide covers the diagnostic process.
Asset group reporting limitations and workarounds
PMax reporting has historically been opaque. The 2026 improvements include channel-level performance breakdowns (Search vs Shopping vs Display vs YouTube within an asset group) and expanded asset insights showing individual asset combination performance.
What you still can’t see directly: query-level performance within asset groups (only Search Categories, which aggregates queries). Individual ad combination performance for specific user segments. Exact attribution paths through PMax to conversion.
Workarounds operators use:
Search Categories Report. Available at the asset group level. Shows the search territory PMax is serving against, grouped by category. Use this to verify that search themes are translating to the intended query territory and to spot waste (categories you don’t want to serve against).
Asset combination report. New in 2026, more detail than before. Shows which combinations are serving most often. Use this to identify which headlines and images are doing the heavy lifting.
Channel-level performance reporting. The 2026 improvement. Confirms whether PMax is distributing budget across channels in the way you expected (or whether it’s stuck heavily in Search and Shopping at small budgets).
GMC Diagnostics for feed health. Separate from PMax reports but essential for ecommerce. If your feed has disapprovals, those products don’t serve regardless of PMax structure.
Custom Looker Studio dashboards. For larger accounts running multiple PMax campaigns, custom dashboards can join PMax data with offline conversion data, channel breakdowns, and asset performance into a single view. The setup time is meaningful (4-8 hours), but the visibility gain is substantial for ongoing optimisation work.
The bigger reporting principle: don’t rely on PMax reports alone. Cross-reference asset performance with GMC Diagnostics, with channel-level revenue data from analytics, and with offline conversion data from your CRM. The full picture comes from multiple sources, not from a single PMax interface.
FAQ
Should I run one asset group or multiple?
Start with 2-4 asset groups segmented by product theme within each campaign. Expand to more asset groups when individual groups consistently exceed 40-50 conversions per month (giving headroom to split). Don’t create more asset groups than your conversion volume can support; each group needs around 20 conversions per month minimum to optimise.
What’s the minimum creative I need to launch PMax?
Technically: 3 headlines, 1 long headline, 1 description, 1 landscape image, 1 square image, 1 portrait image, 1 square logo, business name, and a final URL. Practically, launching with minimums means accepting Low asset ratings and poor performance. Aim for 15 headlines, 5 descriptions, 15-20 images across aspect ratios, and at least one custom 15-second video per aspect ratio if your creative bandwidth allows.
Should I let Google auto-generate video for my PMax campaign?
No, if you have any capacity to produce custom video. Auto-generated video underperforms custom video by 12-40% depending on the source data. A basic 15-second product overview shot on a smartphone outperforms Google’s auto-generation in most cases. If you genuinely cannot produce custom video, consider Asset Studio (Imagen 4 + Veo) as an alternative that gives you operator oversight on the output before it serves.
How do I disable the AI voice-over on PMax videos?
Campaign settings → Video enhancements → turn off “video enhancement control.” This disables AI voice-over generation for that campaign. The default state for new and existing campaigns is enabled (opt-out), so accounts that haven’t been audited since March 2026 may be serving AI-narrated versions of videos without operator awareness. Audit this setting across all PMax campaigns on every account.
When should I add a second asset group to a campaign?
When the existing asset group exceeds 40-50 conversions per month consistently for 4-6 weeks, and when the catalogue contains distinct product themes that would benefit from different creative treatment. Adding asset groups too early splits learning data and slows optimisation. Adding them too late leaves creative relevance gains on the table. For most ecommerce accounts, the second asset group becomes worth adding around month 2-3 of PMax operation once the algorithm has stabilised.
Whatever’s actually happening in your Performance Max setup, the Free Revenue Leak Report tells you in 48 hours. No commitment.
