TLDR: Performance Max (PMax) gives Google’s algorithm a lot of power and without the right setup, it’ll use that power in ways that don’t serve your business. This guide walks through the steps that keep you in control.
Step 1: Get your conversion tracking right before you launch
PMax is keyword-less. It runs on user intent signals, which means it can only figure out who to target if it knows what users are doing after they land on your site. That makes conversion tracking even more critical here than it is for traditional search.
Before you launch, go into your conversion goals in Google Ads and make sure you’ve applied the right ones to the campaign. For ecommerce, that’s a purchase conversion. For lead gen, it’s form submissions, but if you’re importing offline data, you also want qualified or serviceable leads pulling through, alongside the final sale.
If you’re working with low conversion volume, consider adding micro-conversions like add-to-basket or proceed-to-checkout to give the algorithm more to learn from. The general rule: target as far down the funnel as you can but make sure you have at least 30 conversions in the last 30 days for whatever that lowest-funnel conversion is.
If you’re using target ROAS or any form of value-based bidding, check your revenue data is accurate before you go live. A quick spot check: divide your total conversion value by your number of conversions to get your average order value, then sense-check that against what you actually sell. If the number is wildly off, say £1 where you’d expect £40+, your conversion value tracking needs fixing first. Don’t skip this step.
Step 2: Build your asset groups around themes, not everything at once
Asset groups in PMax determine which creative gets served to which audience. If your asset groups are too broad, your ads lose relevance. If they’re too narrow to start with, you slow down the algorithm’s ability to find patterns.
The right structure depends on your product range and business model. For a client with a wide range of product values and categories, think a paint brand selling everything from masonry paint to fire-resistant coatings to marine paint, grouping by product type within the asset group makes sense, with campaigns structured around performance tiers (what sells well and what’s profitable).
For a footwear retailer with strong seasonality, the logic is different. Separate campaigns by product category, split asset groups by gender where relevant, and keep seasonal products in their own campaigns so the algorithm isn’t trying to serve sandal ads in November.
The principle is the same in both cases: make sure the creative inside each asset group is as relevant as possible to what you’re advertising. There’s no point having men’s walking boots creative in the same asset group as women’s summer sandals. Relevance is what keeps your ads from wasting impressions on the wrong people.
Step 3: Add audience signals that actually mean something
PMax uses signals rather than keywords to find the right users. You can and should give it as much useful signal as possible from the outset, the algorithm learns faster when it has quality data to work from.
There are two main types of signals to use. The first is audience lists: your existing customers, your high-LTV customers, in-market audiences from Google’s own native segments (such as “fashionistas” or “in-market for trainers”), and any first-party data you have available. Set these up and apply them the same way you would for any other campaign.
The second is search themes, which is PMax-specific. Because the campaign has no keywords, search themes give the algorithm a steer on what kind of search intent is relevant to your business. They work more like broad match than exact match, a search for “women’s Birkenstocks size 6” can match against a search theme of “sandals” even though the words don’t overlap, because the algorithm reads intent from the full picture of that user’s behaviour.
Add as many relevant search themes as you can, up to 50 per asset group. The goal is to maximize the useful signals you’re feeding the campaign, so use everything you have: your own first-party data, Google’s in-platform audiences, demographics, and a full set of search themes that map to what your customers actually search for.
Step 4: Upload a full set of creative assets
PMax is built to mix and match creative to suit individual users. A price-led headline might convert a deal-hunter; a features-based headline will resonate with someone focused on quality. The algorithm needs options to do this well, which means asset depth is one of the most important levers you have.
Hit the maximums where you can: 15 short headlines, 5 long headlines, 5 description lines, up to 20 images, your business name, and your logos. For video, aim for at least one, if you don’t upload one, Google will auto-generate a slideshow from your images, which is rarely what you’d choose for your brand.
Unlike traditional image extensions, PMax image assets can include text overlays and logo overlays. That makes them well suited to time-sensitive messaging: a sale announcement, a last-chance-to-buy prompt, a seasonal push. A simple animated video, even just a few slides coming together, can outperform a static image, and it doesn’t need a production budget to be effective.
Think about what message you want to land with different audiences, then build creative that covers those bases. Price sensitivity, product features, seasonal relevance, urgency, give the algorithm the assets to match the right message to the right person.
Step 5: Set up brand exclusions and negative keyword lists
A new PMax campaign needs two to four weeks of learning before it starts to perform properly. During that period, it will cast a wide net, which means it will pull in some irrelevant, off-brand, or wasteful search terms if you don’t actively manage it.
The most efficient starting point is to apply any existing negative keyword lists from the rest of your account directly to the PMax campaign. Poor-performing terms, brand-safety exclusions, competitor terms you don’t want to bid on, if you’ve already built those lists, use them.
From day one, check your search term report every day or two (especially in the early weeks) and add negative keywords at the campaign level for anything irrelevant or off-brand. It’s not glamorous work, but it’s what keeps the algorithm learning on the right data rather than the wrong data.
On brand exclusions: these are built in campaign settings as brand lists. Add your own brand name if you’re running a non-brand PMax campaign alongside a separate brand campaign. If you’re part of a larger group, think multiple fashion or retail brands under the same parent, add sibling brands as exclusions to avoid competing against yourself.
You can also use brand exclusions for competitors where you know the audiences don’t overlap. If your customers are brand-loyal and would never have searched for a competitor in the first place, excluding that competitor’s branded terms stops the algorithm from chasing traffic that was never going to convert for you. Note that you can keep shopping ads running even with a brand exclusion in place, so it’s not all or nothing.
Finally, demographic exclusions for age and gender are available in campaign settings. If you know your core customer is a woman and male traffic rarely converts, turn on the gender exclusion. Don’t let the algorithm spend budget figuring out something you already know.
Step 6: Give the campaign time to learn before making changes
Four weeks is the absolute minimum before making any major structural changes. And four weeks only applies if the campaign is running at significant volume, hundreds of conversions, thousands of clicks, tens or hundreds of thousands of impressions. For most campaigns, six to eight weeks is more realistic.
Even after the campaign shows as “out of learning mode,” it’s still figuring out what works. Making big changes to structure, targeting, conversion goals, budget, ROAS targets, or CPA targets before the campaign has stabilised just extends the learning period and delays performance improvement.
What you should be doing in those early weeks: adding negative keywords and brand exclusions as you find issues in the search term data. That kind of reactive maintenance is fine and doesn’t reset the learning. What you want to avoid is pulling the levers on the core setup until the campaign has enough data to be confident in what it’s telling you.
Patience here is a strategy. The algorithm improves fastest when you give it consistent signals and don’t disrupt it before it has found its footing.
Step 7: Read your PMax reports without going mad
PMax reporting can be as simple or as complex as your setup demands. If your conversion tracking is accurate, your goals are set correctly, and everything is working as expected, the native platform data, revenue, ROAS, spend, clicks, impressions, impression share, gives you a solid picture of what’s happening.
Where it gets more complicated is when you have multiple tracking sources. If your source of truth is GA4 rather than Google Ads tags, be aware that GA4 data can look different inside Google Ads than it does natively in GA4. For anything beyond a surface-level read, you’ll want to visualize and manipulate that data outside of the native platform, GA4 on its own is not built for easy reporting, and combining metrics that don’t play nicely together is a common source of confusion and wasted time.
For search term data and competitive insights, use the in-platform reporting, that’s where it lives. For ROAS and revenue reporting, the platform works fine if your tracking is clean. But if you want to move beyond revenue to profit, reporting on POAS rather than ROAS, pulling in CRM data, or getting a single view across channels, you’ll need a third-party reporting tool or a proper data connection to bring those numbers together.
ASK BOSCO® connects your Google Ads data with the rest of your marketing and ecommerce stack, so you can see performance clearly across channels without the manual work of building and maintaining reports from scratch. When PMax is running well, you want to be spending your time on decisions, not wrestling with dashboards.
