PMax Structure Audit
The 27-point checklist for diagnosing why your Performance Max is or isn't compounding.
Who this is for
- $30K+/mo PMax operators with flat ROAS
- Founders evaluating an agency rebuild
- In-house buyers running a quarterly account review
Why this exists
Most PMax accounts plateau because the structural decisions made in week one were wrong, and nobody re-audits them. This is the audit that surfaces those decisions and tells you which ones to roll back before you keep stacking optimisations on a broken base.
Read this first
PMax compounds when its structure matches how the algorithm reads signal. It plateaus when an early decision (too many themes, no audience signal, miss-routed feed) starves a single asset group of learning data and the campaign reallocates spend toward the cheapest available conversions instead of the most profitable. This audit finds the starvation point and tells you whether to feed it or remove it.
How to use this audit
Pull last 60 days of PMax data
Open Google Ads, set the date range to the trailing 60 days, and export campaign-level cost / conversion value / asset-group-level performance to a sheet. Pull insights tab data for search themes (asset group level) and audience signals separately. The audit assumes a steady spend baseline; if you launched in the last 30 days, run it at day 45 instead.
Score every asset group on the structure matrix
Use the scoring table below. Score each asset group on five dimensions, sum to a /25 total. Anything under 15 is a candidate for restructure. Anything over 20 is compounding and should be left alone unless an external constraint (margin shift, inventory change) forces a rethink.
Run the four diagnostic surveys
Search themes drift, audience signal density, feed routing, and brand cannibalisation each get a dedicated section below. Each surveys a separate failure mode that the structure score alone won't surface. A campaign can score 22/25 on structure and still be leaking 30% of its spend to branded search if the cannibalisation check is skipped.
Decide rollback vs forward fix
Every finding falls into one of two buckets: roll back a recent change (restoring prior performance), or ship a forward fix (new structure to compound from). The rollback decision tree in section 2 is the deterministic version. When in doubt, roll back first, then fix; PMax learning is faster from a known-good baseline than from a hypothetical improved one.
Asset group structure scoring matrix
Score each asset group 1-5 on every dimension. Sum to a /25 total. Below 15 is a restructure candidate; above 20 is compounding and should be left to learn.
| Dimension | 1 (poor) | 3 (workable) | 5 (compounding) |
|---|---|---|---|
| Intent purity | Mixed brand + non-brand + competitor in one AG | Brand isolated, non-brand split by category | Single intent per AG, matched to one search theme cluster |
| Signal density | Generic listing feed only, no audience signal | Audience signal layered, feed labels partially populated | Customer match + custom segments + populated custom_label_0..4 |
| Creative volume | 5-7 assets per type, no rotation | 10-15 assets per type, monthly refresh | 20+ assets per type, weekly creative ops cadence |
| Conversion concentration | Below 30 conversions/14 days at AG level | 30-100 conversions/14 days | 100+ conversions/14 days, smart bidding stable |
| Audience coherence | Audience signal contradicts feed | Audience signal partially overlaps feed segment | Audience signal explicitly maps to the feed slice |
Search themes rollback decision tree
Paste this into the audit doc and walk it row by row. The rule of three at the bottom is the most-skipped check.
IF themes added in last 14 days
AND PMax CPA increased > 20%
AND theme density > 5 themes per AG
THEN rollback the additions, revert to prior state, monitor 14 days
IF themes added > 30 days ago
AND PMax CPA stable or improved
THEN keep, document the additions in the change log
IF themes count > 7 per AG (any age)
THEN consolidate to 5 highest-density themes, archive the rest
IF same theme appears in > 2 AGs
THEN cross-AG conflict, isolate to the AG with the highest signal match
# Rule of three: theme density of 3 gives clean signal without dilution.
# Below 3 starves the AG of signal. Above 7 dilutes the signal across
# unrelated intent. The window 3-5 is the sweet spot for most ecom verticals.Audience signals audit
- Customer match list uploaded within the last 90 days (not stale)
- Customer match contains at least 1,000 hashed records (below this, Google won't activate)
- Custom segment URLs include the brand's own site, top 3 competitors, top 3 review sites in the category
- Custom segment search terms reflect the actual non-brand intent (verify against Search Console top queries)
- In-market segment selected matches the product category, not a parent category
- Life events excluded if irrelevant (e.g. avoid 'recently moved' for non-home-goods brands)
- Similar audiences flag set if Google's 2026 changes deprecated the source list
- Audience signal explicitly tagged in the AG name so future operators can see it without opening the AG
- No audience signal contradicts the feed slice (e.g. luxury feed with mass-market in-market segment)
Feed-to-asset-group routing diagnostic
The most-overlooked PMax leak is feed routing: products bleed into asset groups they were never intended for because the custom labels were left blank and the feed selector defaults to all eligible products. The fix is to populate custom_label_0 through custom_label_4 with deterministic values that the AG selectors then filter on.
A working convention: custom_label_0 is brand isolation (brand vs non-brand), custom_label_1 is margin tier (low / mid / high), custom_label_2 is seasonality (always-on / seasonal / clearance), custom_label_3 is performance tier (winner / mid / new), custom_label_4 is reserved for ad-hoc tagging (test groups, geo splits, supplier promotions). Each AG's listing selector then filters on the relevant labels and the feed routes products correctly without any guesswork.
Verify the routing by sampling: pick five products at random, look up the custom labels in the feed export, predict which AG they should land in, and check Google Ads to confirm. If the prediction misses for any of the five, the routing is broken and the diagnostic surfaces which label is doing the breaking.
Brand vs non-brand cannibalisation, the seven exclusion patterns
Add these to the campaign-level negative keyword list and the brand exclusions list inside PMax. Most accounts have three or four; few have all seven.
| Pattern | Example (replace with your brand) | Where to exclude |
|---|---|---|
| Exact brand term | yourbrand | PMax brand exclusions + branded campaign in Search |
| Brand variants | yourbrand.com, your brand, yourbrandshop | PMax brand exclusions list |
| Brand misspellings | yourbradn, yourbrnad | Pull from Search Console top queries, add to brand exclusions |
| Brand plus product | yourbrand serum, yourbrand + flagship SKU | Brand exclusions, plus carve out as branded campaign keyword if separate |
| Brand plus location | yourbrand uae, yourbrand london | Brand exclusions |
| Branded competitor terms | competitor + alternative, competitor vs yourbrand | Branded campaign in Search, exclude from PMax via campaign neg list |
| Brand plus review intent | yourbrand reviews, yourbrand worth it | Brand exclusions if low non-brand value, separate branded campaign if high intent |
Smart Bidding posture for the post-audit window
Once the structure work lands, the temptation is to push tROAS targets up immediately. Don't. Smart Bidding optimises against whatever conversion volume it has accumulated; if you change the structure on Monday and the target on Wednesday, the algorithm reads two simultaneous shocks and resets its learning. Hold the target flat for 14 days after a structural change. Watch the daily POAS, not the headline ROAS, and only adjust the target after the bidding has stabilised against the new asset group shape.
The exception is when the structure change explicitly removes a poorly-converting asset group. In that case, the campaign-level conversion volume drops by the pruned share, but the per-conversion margin tightens. tROAS at that point can move 10-15% upward without choking volume because the remaining asset groups are higher-quality from the start.
If the campaign was on Maximize Conversions before the audit, switch to Maximize Conversion Value with a tROAS floor only after Conversions API and Enhanced Conversions are wired and EMQ scores read 8+ across the platforms you spend on. Smart Bidding optimises against the value signal it sees; pumped-up tROAS against a half-broken value signal is the second most common reason post-audit ROAS drops in week three.
Pre-audit data prep checklist
- Date range: trailing 60 days for steady accounts, trailing 90 if your spend is below $30K/month (need more conversion volume to read signal)
- Export segments: campaign, asset group, day, device, audience signal performance
- Pull the search themes insights tab for every asset group separately, not the campaign-level rollup
- GMC product status export with disapproval reasons, even if you think the feed is clean
- Search Console top 1,000 queries by clicks for the last 28 days (you'll need this in the brand-cannibalisation section)
- GA4 ecommerce data with new vs returning split for the same window, so you can compare against ad-platform attribution
- Account change history for the last 90 days, to flag what changed and when (rules out 'we shipped a structural change two weeks ago and forgot' surprises)
- List of every campaign currently spending plus its objective, so you don't audit a paused campaign and recommend changes that won't compound
What good looks like at the end of this audit
Every asset group scores 18+/25 on the structure matrix. Search themes are 3-5 per AG, no duplicates across AGs. Audience signals are populated on every AG with at least one of customer match / custom segment / in-market. Feed routing is verified against five sample SKUs. Brand cannibalisation patterns are excluded at both the campaign and brand-exclusions level. Smart Bidding posture is documented and aligned with the new structure. From here, ship one optimisation per week and let the campaign learn between changes.
External resources
Authoritative references we link to alongside the template. Read them before running the audit.
- Google Ads, Performance Max overviewOfficial feature reference. Bookmark for arguing with Google reps.
- Google Ads, asset groups in Performance MaxAuthoritative spec on how asset groups consume signal.
- Google Ads, search themes in Performance MaxHow themes feed PMax. Read once before running the rollback tree.
- Google Ads, audience signals in Performance MaxHow audience signals seed the targeting model in section 3.
- Google Ads, Smart Bidding overviewHow tROAS and Maximize Conversion Value consume the value signal you feed them.
- Google Merchant Center, custom labelsReference for the feed-routing diagnostic in section 4.
- Google Skillshop, Performance Max certificationFree training. Worth running through once before deep-auditing your first PMax account.
- Search Engine Land, Performance Max coverageTopic index. Useful for tracking platform changes between Google's quarterly updates.
Want this run for you?