How to Create a Performance Max Campaign: My 14-Day Setup

Most guides on how to create a performance max campaign stop at the wizard: pick a goal, drop in some assets, hit publish. That's the easy part, and it's also where the real work starts, not ends. In my experience the launch click decides almost nothing the outcome is set by the sequence you run in the two weeks after it, and by what you loaded before you ever opened the campaign builder. I've inherited enough freshly-launched PMax campaigns stuck at a trickle of spend to know that the failure is almost never "the algorithm didn't work". It's that the account was handed to automation with broken tracking, thin assets, or a target so tight the campaign couldn't breathe.

So this is not a screenshot walkthrough of the setup screens. It's the actual rhythm I follow as a performance max setup guide 2026: the minimum assets I load, the day-by-day sequence across the first 14 days, and the configuration mistakes I keep watching new accounts repeat. If you want the conceptual backdrop on what this campaign type even is before you build one, my primer on what Performance Max is is the ground I'm assuming here.

A quick word on why the sequence matters so much with this campaign type specifically. Unlike a classic search campaign, where you can hand-tune keywords and watch each ad group, PMax hands almost every lever to Google AI at once placements, audiences, creative combinations and bids all move together. That's why pmax campaigns reward preparation over tinkering: the moment you publish, the machine starts learning, and everything it learns in those first days becomes the foundation it optimises on top of. Get the inputs right up front and the automation compounds in your favour; get them wrong and you're teaching the algorithm to chase the wrong outcome with real money. The rest of this guide is organised the way I actually work a new build prep gates first, then assets, then the fourteen-day rhythm, then the errors because that ordering is the strategy.

Before you build: the two things that must exist first

Nothing else in this performance max setup guide 2026 matters if the account can't measure what it's optimising toward. Before I open the campaign builder, two things have to be true. First, conversion tracking is live, tested, and firing clean not "set up", but verified with a real test conversion so I've watched the event land in the interface. Second, at least one audience signal exists to seed the machine: a customer list, remarketing pool, or first-party segment strong enough to give the AI a starting point rather than a cold start.

I treat these as gates, not steps. Google's own guidance is blunt that a campaign created without working conversion tracking can serve, but its spend may be limited which is the polite version of "your budget won't move and you won't know why". So the very first thing I confirm is that conversion tracking is genuinely reporting, because every optimisation decision the campaign makes over the next fortnight is only as good as the signal underneath it. Get this wrong and you don't have a slow campaign, you have a blind one.

Before I build, I also make a deliberate decision about which conversion goals the campaign will optimise toward. This is where a lot of new accounts quietly go wrong: they leave every account-level conversion action switched on, so the campaign is chasing newsletter signups and add-to-carts with the same urgency as a purchase. I set campaign-specific goals so the AI is pointed at the outcome that actually pays the business, and I keep the goal list tight enough that the signal isn't diluted across a dozen soft events. If the account genuinely has multiple meaningful actions, I still rank them rather than treating them as equal.

The second gate is the audience seed, and here the strongest input I can give is first-party data. Where the account has one, I attach a customer match list the emails of people who already bought or converted because it gives the algorithm a concrete picture of who to look for rather than a vague description. On top of that I add custom segments built from search behaviour and site activity, so the audience signal describes intent as well as identity. None of this hard-targets the campaign; in PMax these are signals, not fences. But a rich seed shortens the cold-start dramatically, and a customer match list plus a couple of well-built custom segments is the difference between the machine guessing and the machine starting warm.

Finally I walk the campaign settings one field at a time before I ever touch creative location and language, the schedule, the URL options, brand exclusions because the settings screen is where the quiet, default-on mistakes live. It takes ten minutes and it's the cheapest insurance in the whole build.

The minimum assets I load before launch

Thin asset groups are the single most common reason a new PMax campaign under-serves, and it's entirely self-inflicted. Performance Max builds ads by combining what you give it across every channel, so if you hand it a skeleton, it has nothing to test and the campaign quietly throttles itself. There is a real floor here, and I don't go below it.

My non-negotiable asset floor

For text, I load the full complement Google asks for rather than the bare minimum the builder accepts: I aim for the fuller set of short and long headlines and descriptions the platform recommends, because more variations means more combinations to test, not more clutter. The long headlines matter more than people think they're what shows on the larger, more prominent placements, and skimping on them quietly limits where your ad can appear. For images, I never launch with a single format I load landscape, square and portrait so the campaign can render across placements instead of being locked out of half of them. A square logo goes in every time. And at least one video: if I don't provide one, the system will auto generate one from my other assets, and an auto-built video is almost always worse for brand consistency than something I control.

Every individual creative asset earns its place by adding a real variation, not by padding a count a headline that just restates another headline gives the algorithm nothing new to test. Google's optimisation guidance points at a similar floor roughly the full headline-and-description set, several images across all three orientations, and at least one video and reaching it is the difference between a campaign that can serve everywhere and one that's effectively benched on most inventory. The specific pixel dimensions and file limits for each of these slots live in my breakdown of the Performance Max asset specs, which I keep open while I'm loading a new asset group. The point isn't to hit a checkbox it's that asset diversity is the raw material the learning phase actually chews on.

For any retail build there's a second asset set that matters as much as the creative: the product feed. If the campaign is going to serve Shopping inventory, the Merchant Center product feed is half your ad, and a feed with weak titles, missing images or disapprovals will bottleneck the whole campaign no matter how good your text and video are. I get the feed clean and approved before launch, not after, because a campaign waiting on a broken feed is a campaign burning learning days for nothing.

The first 14 days, day by day

This is the part almost every guide skips, and it's the part that decides whether the campaign settles or stalls. The learning phase is real, and my job during it is mostly to stay out of the way while making sure I didn't seed anything wrong on day one.

Days 1–2: launch and leave it

On launch day I set the campaign live with a budget that can realistically absorb a few conversions' worth of spend starved budgets never exit learning cleanly and a bidding strategy that isn't fighting the campaign from the start. PMax runs on smart bidding by design, so the choice isn't whether to automate bids but which flavour to open on. For a brand-new campaign with no target history, I choose Maximize Conversion volume (or Maximize Conversion Value) without a target attached, so the algorithm can find its footing before I bolt a constraint onto it. I deliberately do not set a target on day one no Target CPA, no Target ROAS because a target with no data behind it is just a guess the campaign will strangle itself trying to hit. Then I do the hardest thing: nothing. The first 24–48 hours are just data populating, and any change here resets the clock.

Days 3–7: watch, don't touch

Through the first week I'm reading, not editing. I'm checking that impressions have started, that conversions are being recorded and attributed correctly, and that spend is pacing toward the budget rather than flatlining. If the campaign got zero impressions after three business days, that's a genuine signal something is misconfigured usually tracking or a disapproved asset group and I go hunting. But if it's serving and converting, I sit on my hands. The temptation to "optimise" in week one is exactly what keeps a campaign trapped in learning.

One input I will add during this window, if I didn't at build time, is search themes. They're an optional way to tell Google AI what your customers are actually searching for, and they help the campaign ramp on Search inventory faster without forcing it. I treat a search theme as a hint, not a keyword it nudges which kinds of search query the campaign leans into while it's still finding its feet. What I don't do yet is start reading the search terms report and reacting to individual queries; that's a day-8-onward job, once there's enough volume for a pattern to mean something rather than a single fluke.

Days 8–14: the first allowed changes

Only once I have a week or two of stable conversion data do I start making deliberate, gradual moves: introducing a Target ROAS or Target CPA if the account is ready for it, tightening or loosening budget, and reviewing which asset groups and audience signals are pulling weight. When I do add a Target ROAS, I set it off the campaign's actual observed return, not the number the client wishes for then move it in small steps, because a sudden jump is read by the system as a shock and pulls serving back. Even then I change one thing at a time and give it a conversion cycle to settle. This measured cadence is the same discipline I apply to Performance Max bidding strategies across the account the right choice among the available bidding strategies is whichever one the data can currently support, and targets get earned by evidence, not assumed at launch. Google's help documentation is explicit that the learning phase typically runs one to two weeks and that large, frequent changes during it are what limit serving, which is exactly why I front-load the setup and then hold still. The full official walkthrough lives on Google's Create a Performance Max campaign help page, and it's worth reading alongside this before your first build.

The configuration errors I see on almost every new account

When I audit a struggling new PMax build, the same handful of setup mistakes come up again and again. None of them are exotic they're all the kind of thing that's easy to leave in a default state and never revisit.

The first is broken or partial conversion tracking. The campaign appears to run, but it's optimising toward an event that isn't firing correctly, so the AI is learning from noise. The second is Final URL expansion left on by accident when the advertiser didn't want Google rewriting their landing pages a setting that's on by default and quietly sends traffic to pages you never chose. The third is targets set too aggressively at launch: a Target ROAS that's wildly optimistic or a Target CPA that's far too low will strangle serving before the campaign has any data to hit it with, which is why I launch without one. The fourth is the thin asset group already covered above the silent spend-cap. And the fifth, subtler one is over-restrictive audience signals or a feed that hasn't been properly segmented, so the campaign is technically live but pointed at the wrong slice of the catalog.

That last point is where setup bleeds into structure. A clean launch can still underperform if the feed underneath it isn't organised so budget follows what converts which is a separate discipline I break down in my PMax product group strategy. Fix the five configuration errors above and you've eliminated the reasons most new campaigns stall; get the product groups right on top of that, and the setup genuinely starts working for you instead of against you.