Last updated: Rules, fees and platform policies change. This guide is reviewed and kept current.
Performance Max is Google Ads’ offer to you: “give us the budget and the goal, and we will pick the channels.” Set up correctly, it works a far wider inventory than a small team could manage on its own. Set up incorrectly, it does this: it spends most of your budget reselling to customers who were already searching for you by brand name, and shows a wonderful ROAS in the dashboard.
This post is not a call to switch PMax off. The aim is to make clear where you need to intervene: machine learning optimises well, but you have to tell it what to optimise for.
Performance Max: seven steps to keep control
Using the automation without surrendering to it.
- Conversion dataIf measurement is wrong, every other setting works against you
- Separate brand searchStop budget flowing to traffic you would have had anyway
- Product feedRoughly seventy per cent of the result comes from feed quality
- Split campaigns by marginGroup by profitability, not by category
- Build testable assetsSo you can tell which image and copy actually worked
- Search themes and negativesSteering and exclusion belong in the same plan
- Measure against the learning periodIntervening early resets what the system has learned
1. If the conversion data is wrong, everything else is wasted
Automated bidding strategies learn from the signal you give them. A system that learns perfectly from the wrong signal takes you to the wrong place quickly.
- Leave a single primary conversion: purchase. Mark events like “add to cart”, “contact form” and “page view” as secondary so they are not used for bidding.
- Send the conversion value as the real amount. If you assign a fixed value, the system treats a 200-lira order and a 2,000-lira order as the same thing.
- Feed returns and cancellations back. Report returns as negative conversions where you can. Otherwise your most-returned products look like your “most successful” ones and budget flows there.
- Turn on enhanced conversions. In a world of tightening cookie restrictions, incomplete measurement means incomplete learning.
Spending a day here before you build the campaign determines the efficiency of the next three months.
2. Separate out branded search
PMax’s best-known problem: it reports users who searched for your brand, and were already intending to buy from you, as its own achievement. That traffic is cheap and converts well, which makes it an attractive target for the algorithm.
The fix has two steps: apply a brand exclusion list on the PMax campaign, and manage brand traffic through a separate search campaign. That way you can answer two questions separately: “what am I spending on defending my brand?” and “what am I spending on acquiring new customers?”
3. The product feed is seventy per cent of the ad
In shopping inventory, product data speaks as loudly as your ad copy; louder, in fact. Every field you fix in the Merchant Center feed shows up directly in performance:
- Title: brand + product + distinguishing feature + variant. Not a keyword pile; the form people actually search in.
- Image: clean background, product filling the frame, a main image with no campaign text on it.
- Category and attributes: size, colour, material, gender, age group. Every field you leave blank is a query you cannot match.
- Stock and price accuracy: inconsistency produces both disapprovals and wasted clicks.
- Custom labels: tag breakdowns like margin band, season, best seller, stock to clear. Building the campaign structure on those labels is the most practical way to hold control inside PMax.
4. Split the campaign by margin, not by category
Most stores build campaigns on category logic: shoes, bags, accessories. But what determines your budget allocation is not category, it is profitability. Use custom labels to build three groups:
- High-margin products: aggressive target ROAS and most of the budget.
- Mid-margin / volume products: a more cautious target; the aim is to cover fixed costs.
- Low-margin or high-return products: either delist them or test them deliberately with a capped budget.
Setting a target ROAS without that split means applying the same standard to products with different margins. Mathematically, it is wrong.
5. Build assets so they can be tested
You can add text, images and video to every asset group. If you do not add a video, the system generates one automatically — and it is usually not the image your brand wants. Put in at least one vertical, short video that works watched on mute.
Do not fill the headlines in at random either: some should focus on the benefit, some on price or promotion, and some on a trust element (easy returns, fast delivery, warranty). Replace the ones marked “low” in the asset performance report regularly; that is one of the few direct levers you have left.
6. Search themes and negatives
Search themes are the way to tell the system “consider these intents too”. They are valuable when you open a new product group or a season, because there is no data yet. Negative lists, on the other hand, protect your budget from irrelevant queries: people looking for free stuff, people looking for spare parts, competitor brand names, people applying for a job.
Going into the search terms report once a month and adding new negatives is tedious, and one of the highest-return maintenance jobs there is.
7. Measure in line with the learning period
The most common mistake in PMax is changing settings every week. Interventions made while the bidding strategy is learning reset the process. Practical discipline:
- Wait at least two weeks after a change.
- Move the target ROAS by no more than 10–15% at a time.
- Make budget changes gradually; doubling it breaks the learning.
- Compare over windows of at least 14 days, not day by day.
Catching the campaign that looks good in the dashboard and bad at the till
Finally, close the gap between the ROAS PMax reports and your real profit. The platform talks in revenue; you have to talk in margin. Every month, put these three numbers side by side: ad spend, revenue from advertising, and the real contribution margin on that revenue. Without the third number you do not know whether the campaign is profitable — you only know it is big.
The difference between people who get good results from Performance Max and people who do not is almost never a “hidden setting”. The difference is clean conversion data, a tidy product feed, a structure built around margin, and discipline about intervening. With those four in place, the automation works for you; without them, the automation takes the easy route with your budget.
Setup order: which job, and when?
On a new account, or when tidying up one that has drifted, sequence matters. The order below speeds up learning without burning budget.
- Measurement (1–2 days): a single primary conversion, real values, enhanced conversions, return feedback.
- Feed cleanup (3–5 days): fix disapproved products, rework titles, apply custom labels.
- Brand protection (1 day): a separate brand search campaign, brand exclusion in PMax.
- PMax (first 3 weeks): one campaign, starting with high-margin products. Set the target ROAS slightly above current performance; do not write in an imaginary number.
- Expansion (after week 4): a second campaign for volume products, new intent tests with search themes.
What to do when the budget is small
On accounts with a low daily budget, the most common mistake is splitting that budget across five campaigns. Automated bidding has to see enough conversions to learn; a divided budget never gets any single campaign over that threshold. The rule is simple: few campaigns, a clear product selection, patient measurement. Advertise the profitable, well-stocked core of your range rather than the whole catalogue.
And one more thing: advertising does not rescue a bad product page. If every unit of currency you pay per click lands on a page with weak images, a missing size table or an unclear delivery time, the ad budget is really financing the page’s defects. Starting optimisation at the landing page rather than the campaign usually produces faster results.
A success story, and independent data
- Lindex (Google’s own case study, November 2022): ran Performance Max across 18 markets and, using its membership-club data, raised ROAS by 27%.
- Independent data, Optmyzr (October 2024; 9,199 accounts, 24,702 campaigns): 97% of advertisers do not use brand exclusion lists, and performance with and without them was essentially the same. 55% of campaigns run without a product feed; they show better conversion rates and CPAs but weaker ROAS.
What it means: Google’s success stories and independent data paint different pictures. Apply the controls in this article with a before/after comparison of at least 14 days in your own account, and judge by net margin rather than dashboard ROAS.
Sources: Google, Lindex case study; Optmyzr, Performance Max Study.
Related reading
- Work out your margin before you decide what number to set the target ROAS to: calculating real profit margin on marketplaces.
- So you do not lose the visitor advertising brought you on a slow page: the speed optimisation checklist.
- The new rule set on the organic side: getting cited by AI search.
Sources and official links
- Google Ads Help — About Performance Max campaigns — The inventory the campaign covers and how it works
- Google Ads Help — Getting started with Performance Max — The official order for conversion goals, assets and bidding setup
- Google Ads Help — How asset groups work — How asset groups are structured and their specifications
- Google Ads Help — Search targeting and controls for PMax — Brand exclusions, negative keywords and search themes
- Google Ads Help — Asset group reporting — Asset group reporting and top ad combinations
Rules, fee schedules and platform policies change. The figures in this post reflect the position at the time of writing; confirm the current position from the official sources above before you act.
Haftalık bülten
Pazaryeri ve e-ihracatta bu hafta ne değişti?
Komisyon oranları, mevzuat ve algoritma değişikliklerini haftada bir, tek e-postada özetliyorum. Reklam yok, tekrar yok — istediğiniz an tek tıkla çıkabilirsiniz.
