Why Performance Max drifts
PMax optimises towards whatever target you give it, using every inventory type Google owns. If you ask for a high return, it will find the cheapest conversions in the account, and the cheapest conversions are always people who already know you. That is not the algorithm misbehaving. It is doing exactly what it was told, in a way that flatters the report.
~30
Conversions in 30 days per campaign, the minimum for stable learning
14 days
Leave targets alone this long before judging a change
10-15%
Maximum target move per adjustment to avoid restarting learning
A structure that holds up
Simplicity beats cleverness, but not at the cost of visibility. Three or four campaigns is usually right: brand search on exact and phrase, a PMax campaign for high-margin lines, a second for the rest of the catalogue, and where volume justifies it a tightly controlled search campaign on your money terms.
Brand traffic
- What we set
- Brand exclusion list plus a separate brand search campaign
- Why
- Stops PMax claiming credit for demand you already had
Margin
- What we set
- Split campaigns by margin band using feed labels
- Why
- Lets you set different targets where profit differs
Placements
- What we set
- Account-level exclusions and content suitability settings
- Why
- Keeps spend off low-quality apps and video inventory
New customers
- What we set
- New customer acquisition goal with a value bid
- Why
- Prices growth explicitly rather than hoping for it
Creative
- What we set
- One asset group per theme, video included
- Why
- Prevents thin groups and repetitive placements
Learning
- What we set
- Stable targets for at least 14 days, then 10 to 15% moves
- Why
- Stops the campaign restarting learning every week
Exclusions and negatives, done once, properly
- Brand exclusion list applied to every non-brand PMax campaign.
- Account-level negative keyword list for competitor names, jobs, wholesale and support queries.
- Placement exclusions for low-quality apps and mobile game inventory, reviewed monthly.
- Content suitability settings tightened to match the brand, not left on default.
- Location settings set to presence, not presence or interest, unless you genuinely ship everywhere.
The first thing we find is a Performance Max campaign quietly buying brand clicks, and a monthly report celebrating the ROAS it produced.
The feed is the targeting layer
In a shopping-led account the product feed decides what PMax can show and who it shows it to, so feed work outperforms bid tinkering almost every time. Titles carrying the attributes people actually search, accurate GTINs and brand fields, real product types, and custom labels for margin band, stock depth and seasonality.
Custom labels earn their keep
Label by margin and stock, then segment campaigns on those labels. It is the only practical way to bid differently for a 60% margin bestseller and a 12% margin clearance line inside an automated campaign type.
Asset discipline
One asset group per genuinely distinct theme. Five to seven headlines, several descriptions, a full range of image ratios, and at least one vertical video, because if you do not supply video Google will generate something for you and it will look like it. Refresh the weakest assets monthly rather than rebuilding the campaign.
Reporting that tells the truth
Insist on four cuts: non-brand conversions separated from brand, new versus returning customers, cost per acquisition by margin band, and asset-group level search terms and placements. A single blended ROAS figure cannot tell you whether you grew or simply harvested demand you already had, and that distinction is the whole point of running paid media.
This is how we run every Google Ads account, on a fixed monthly fee from £850 plus VAT rather than a percentage of your media spend.
The monthly checklist
- Review search terms and placements. Add negatives and exclusions.
- Check brand leakage: non-brand conversion share should be stable or rising.
- Refresh the two weakest assets in each asset group.
- Audit feed errors and disapprovals, and fix titles on underperforming lines.
- Review new customer share against target before touching ROAS.
- Make at most one target change, then leave it alone for a fortnight.
Questions we get asked
Why is Performance Max spending on brand searches?
Because brand terms are the cheapest conversions available and the algorithm optimises for the target you set. Add a brand exclusion list at campaign level, keep a separate brand search campaign, and compare non-brand conversions before and after. Without that split your reported ROAS is mostly people who were already going to buy.
Should I use Performance Max or Standard Shopping?
Run both where volume allows: Performance Max for breadth and new customer discovery, a tightly structured Standard Shopping or search campaign for your highest-margin lines where you want manual control of query and bid. For small catalogues under a few hundred products, one well-fed PMax plus brand search is usually enough.
How much budget does Performance Max need to learn?
Give it enough for roughly 30 conversions in 30 days per campaign, and leave the target alone for at least two weeks. Under-funded campaigns thrash: they exit learning slowly, spend erratically and produce data too thin to act on.
How many asset groups should a campaign have?
One per genuinely distinct theme, usually mapped to product category or margin band, with five to seven headlines, several descriptions, and images plus at least one vertical video. Splitting into a dozen thin asset groups starves each one of data.
Do search themes replace keywords?
No. Search themes are hints, not targeting. They nudge the campaign towards intent you know converts, but they do not exclude anything. Real control comes from brand and negative lists, feed structure, audience signals and campaign-level segmentation by margin.
What reporting should I demand?
New customer versus returning, non-brand conversions separated from brand, cost per acquisition by margin band, and asset-group level placement and search-term data. If your report is a single blended ROAS number, you cannot tell growth from harvesting.
Written by
Sam, co-founder, growth and paid media
Runs growth and paid media at Mesmerise Marketing. Ten years of paid and organic campaigns for challenger brands across DTC, B2B services and hospitality.
Meet the team
