Google Performance Max has been one of the most debated campaign types in paid media since its broader rollout — and in 2026, after running it across a range of client accounts, Choco Media has a clearer picture of where Google Performance Max actually delivers and where it quietly drains budget. This post is a field report, not a press release. We are writing it for marketing managers and founders who are running PMax or being pitched it and want a direct read on what to expect.
Performance Max consolidates Search, Shopping, Display, YouTube, Discover, Gmail, and Maps into a single campaign. Google’s machine learning decides where to show ads, which assets to combine, and how to allocate budget across surfaces. In theory, it reduces complexity and finds conversions you would not have found manually. In practice, the results depend heavily on how you set it up, how much data you feed the algorithm, and how willing you are to give up control in exchange for scale.
If you are running e-commerce, lead generation, or a local service business and you have at least 30–50 conversions a month tracked reliably, this post will tell you what to expect from PMax and what to watch.
What Google Performance Max actually optimises for
Before anything else, it helps to understand what PMax is doing when it runs. The campaign optimises for your stated conversion goal using a tROAS (target return on ad spend) or tCPA (target cost per acquisition) bid strategy. Google’s systems decide which combination of assets, audiences, and placements is most likely to hit that target.
The catch is that “conversion” means whatever you have told Google Analytics 4 or your conversion tracking to count. If your tracking is off — double-counting, misattributing, or including soft events like page views — PMax will optimise hard for the wrong thing. We have seen campaigns with healthy-looking CPA numbers that were quietly inflating results by counting form views instead of form submissions. PMax is only as smart as the signal you give it.
- Set up server-side conversion tracking before launching PMax — client-side is more vulnerable to ad blockers and browser restrictions
- Use a primary conversion action per campaign and exclude low-intent events from bidding
- Give the algorithm at least 6–8 weeks and 50+ conversions before drawing conclusions
- Avoid switching bid strategies mid-learning; each change resets the learning phase
The asset group structure that actually works
One of the most common mistakes we see with PMax is treating it like a single ad group where you dump all your creative. Google can mix and match any combination of your headlines, descriptions, images, and videos. If your assets span wildly different products, audiences, or offers, the algorithm will produce incoherent ad combinations.
Our approach is to build asset groups around tightly themed product or service clusters. For an e-commerce client selling three product lines, we run three asset groups — each with creative, copy, and audience signals specific to that category. This gives the algorithm a coherent creative palette to work with and makes performance analysis legible.
What goes in a tight asset group
- Headlines: 6–10 headlines all speaking to the same offer or category
- Descriptions: 4–5 descriptions, each reinforcing one specific benefit
- Images: square, landscape, and portrait variants all featuring the same product family
- Videos: at minimum a 15-second and a 30-second version; no video means Google auto-generates one from your images, which is rarely great
- Audience signals: customer list, past purchasers, or custom intent audiences relevant to that cluster
Audience signals are worth spending time on. They do not restrict who sees your ads — PMax can and will show ads to people outside your signals — but they give the algorithm a starting point for who to prioritise. A warm customer list or a well-built custom intent audience shortens the learning phase considerably.
Where PMax wastes budget (and what to do about it)
PMax has a reputation for being a black box, and some of that reputation is earned. The most common budget waste patterns we find in account audits:
Brand keyword cannibalisation
PMax will happily bid on your brand terms and take credit for conversions that would have happened anyway through organic or direct. We add brand keywords as negative keywords at the campaign level (or use a separate brand Search campaign with a higher priority) to prevent this. Without it, your PMax CPA can look deceptively efficient.
Search term opacity
PMax does not give you the full search term report that standard Search campaigns do. You can see some search categories in Insights, but the granularity is limited. This makes it harder to add negatives for irrelevant queries. Google does allow account-level negative keywords and campaign-level negatives for PMax — use them. Build a negative keyword list before you launch, not after you have spent the first €2,000 learning what the algorithm misunderstood.
Display and YouTube bleeding
A portion of PMax budget always goes to Display and YouTube placements, especially early in the campaign. If your goal is bottom-of-funnel conversions, these placements rarely perform at the same efficiency as Search. You cannot fully exclude them in PMax, but you can set a tROAS that is tight enough to discourage the algorithm from pursuing low-quality placements. Setting a realistic but firm target forces the algorithm to be selective.
In client work, we have found that the accounts where PMax performs best are the ones where tracking is watertight, creative is organised by theme, and the team is willing to wait the full 6–8 week learning period before adjusting. Impatience is the most common reason PMax underdelivers.
PMax vs standard Shopping and Search: when to use which
PMax is not a replacement for every other campaign type. There are situations where standard Shopping or Search campaigns still outperform it.
- High brand search volume: if a significant share of your conversions come from branded search, keep a dedicated brand Search campaign and exclude brand from PMax
- Specific product categories needing manual control: if you have a product line with much higher margins where you want to control bids precisely, standard Shopping with a manual CPC or target ROAS gives more levers
- Small budgets under €50/day: PMax needs data volume to learn; at very low budgets, a tightly managed Search campaign usually outperforms because PMax cannot exit the learning phase
- New accounts with no conversion history: the algorithm starts cold; seed it with a Search campaign first, generate 30+ conversions, then introduce PMax
For accounts that already have conversion history, a meaningful product catalogue, and a budget above €80–100/day, PMax can genuinely find incremental volume that Search alone would not reach — particularly through YouTube and Discovery placements that warm audiences before they search.
Reporting on PMax: what the numbers are actually telling you
PMax reports conversion value and cost at the campaign level, but it attributes conversions across all touchpoints Google can observe. This means a customer who saw a YouTube ad, clicked a Discovery ad, and then converted via a Shopping result gets counted — but you will only see the aggregate in the PMax campaign report.
This makes PMax look very efficient in last-touch or Google-reported attribution. The honest way to evaluate it is to run an incrementality test: pause PMax for two weeks on a cohort where you have strong organic and direct traffic, and see whether total conversions drop. If they do not drop, PMax was largely taking credit for conversions that would have happened anyway.
- Use GA4 data-driven attribution as your primary measurement model rather than last-click
- Compare PMax periods against matched prior periods, controlling for seasonality
- Track view-through conversions separately — they inflate reported performance
- Check Google Ads Insights for the top-performing asset combinations and use that data to improve future creative
Our paid media service includes a full attribution review before we launch any PMax campaign, because the reporting setup determines whether you can trust the results.
Creative requirements: what PMax actually needs to perform
PMax is creative-hungry. The algorithm needs enough asset variety to run meaningful tests across placements. The minimum is not the ideal — it is the floor below which performance reliably suffers.
Minimum viable creative package per asset group
- 15 headlines (max 30 characters each)
- 5 long headlines (max 90 characters)
- 5 descriptions (max 90 characters)
- 3–5 images in each ratio: 1:1, 1.91:1, 4:5
- 1 landscape logo and 1 square logo
- At least 1 video (15 seconds minimum; 30 seconds preferred)
Video is worth emphasising. Accounts that launch PMax without a proper video often see Google auto-generate one from static images. These auto-generated videos tend to perform poorly and consume a portion of budget on YouTube. Investing in even a simple 15-second product or service video — shot on a phone with decent lighting — consistently outperforms the auto-generated alternative.
For help structuring video briefs that produce usable ad creative, our work on AI-assisted content creation covers the briefing side of short-form video production.
The Performance Max setup checklist we run before launch
Based on what we have learned across accounts, here is the pre-launch checklist we work through before activating any PMax campaign:
- Conversion tracking verified: server-side or GA4 linked, primary conversion action confirmed, no duplicate counting
- Customer match list uploaded: minimum 1,000 matched users; used as audience signal, not restriction
- Brand keywords excluded: added as campaign-level negatives or managed via a separate brand Search campaign
- Account-level negatives in place: irrelevant industries, competitor names if needed, obvious junk queries
- Asset groups themed by cluster: one theme per group, not all products in one group
- Videos confirmed: at least one real video per asset group
- Budget set for learning: minimum €50/day during the 6-week learning window
- tROAS or tCPA set realistically: based on historical data, not aspirational targets
- URL expansion reviewed: if you want PMax to control landing page selection, confirm all eligible pages are high quality; otherwise restrict to specific URLs
Our honest verdict on Google Performance Max in 2026
PMax works. But it works on its own terms, and those terms require more upfront discipline than most teams apply. The accounts where we see it perform consistently are the ones where tracking is right, creative is organised, the learning phase is respected, and brand cannibalisation is controlled. The accounts where it underperform are almost always missing one or more of those conditions.
It is not a hands-off campaign. The reduced manual controls mean you have to be more intentional about what you put in — better audience signals, better creative, better tracking — because you have fewer levers to pull once it is running. Think of it as a high-performance engine that needs the right fuel rather than a self-driving car.
- PMax performs best at budgets above €80–100/day with 30+ monthly conversions
- Without video, auto-generated assets drag performance on YouTube and Discovery
- Brand cannibalisation and attribution inflation are the two most common ways PMax looks better on paper than it is
- Asset groups should be themed, not mixed — one cluster per group
- The 6-week learning phase is real; accounts that change strategy early rarely see the full potential
If you are weighing whether PMax is the right move for your account — or if you want a fresh pair of eyes on a campaign that is not delivering as expected — get in touch and we can walk through the setup together.