Most of the "black box" criticism aimed at Performance Max has less to do with the campaign type than with how it was built. One asset group, one listing group, final URL expansion left on by default, brand traffic never excluded: with those four conditions in place, it genuinely is impossible to explain what the campaign is doing or why. The same campaign, segmented properly, can be easier to read than a standard Shopping setup.
What follows is the set of levers that return control, and the points at which those levers stop being enough.
How much of the black-box complaint is actually true?
The true part: you cannot bid at keyword level, you cannot see query-by-query spend on the Search network with full fidelity, and you cannot opt out of YouTube or Display inventory with a single switch. That is by design.
The untrue part is the claim that you can see nothing. Search term insights, asset group level performance tables and channel reporting added to the interface over the past couple of years have made PMax noticeably more transparent than it was at launch. Some breakdowns still are not in the UI, though — reports pulled with Google Ads Query Language, or community-written PMax scripts, remain the practical way to see channel and asset group distribution.
So visibility is not zero. It is simply not free. If whoever built the account never segmented the structure, there is nothing for the report to show.
The asset group is the only real structural unit in PMax
There are no ad groups in PMax. There are asset groups, and the whole reporting model sits on top of them. Put 400 products and six categories into a single asset group and you are left with a one-line performance table. Build the segmentation in this order:
- By margin or price band. High-margin products get their own asset group and their own target ROAS logic. Tag them with a custom label in Merchant Center and split the listing group on that label.
- By category or use case. Split only if the images and copy genuinely differ. If you are going to reuse the same five headlines, the split buys you reporting and nothing else.
- By conversion maturity. Putting best sellers and never-sold SKUs in the same group ends with the entire budget flowing to products that were already selling. Give slow stock its own campaign or asset group.
- By search theme. You get up to 25 search themes per asset group. They are signals, not keywords, but they are the clearest way to document which group is chasing which intent cluster.
A practical threshold: do not split if the resulting asset group cannot produce meaningful weekly data. As conversion volume fragments, the system tries to learn separately in each group and all of them slow down at once.
No ROAS reading is trustworthy until brand traffic is separated
Branded queries are traffic you were going to get anyway, and they convert at a multiple of non-branded rates. When PMax absorbs them, the campaign report looks excellent while incremental contribution sits close to zero. Two mechanisms exist, and both should be used:
- Brand exclusion lists. Set at campaign level, they suppress queries associated with a given brand across Search and Shopping inventory. Add your own spelling variants and the common misspellings.
- Account or campaign level negative keyword lists. Use these for non-brand noise: job applications, user manuals, complaints, second-hand listings, reseller enquiries. Negatives can be applied to PMax, and the number of accounts still not using this lever is surprisingly high.
- A separate brand Search campaign. Keep brand traffic in a campaign you control, on a bidding strategy you chose. An exact match keyword that matches the query identically takes priority over PMax — use that rule in your favour when designing the structure.
After you carve out brand, ROAS will drop. What falls is the cosmetics of the report, not performance.
With final URL expansion enabled, PMax can send traffic to any page on your site and write its own headline for it. Spend landing on blog posts, the careers page or out-of-stock product details usually traces back to this. Switch it off on day one for lead generation accounts. If you keep it on for ecommerce, at minimum define URL exclusion rules.
Feed and feedless setups behave like two different campaign types
A PMax campaign with a Merchant Center feed attached leans heavily into Shopping inventory. You get product-level reporting, you can carve product sets with listing groups, and the performance curve resembles standard Shopping. A PMax campaign built without a feed is entirely asset-driven: it spreads across Display, YouTube, Discover and Gmail on the strength of the images, videos and headlines you upload. Learning period, cost structure and budget behaviour are not the same in these two cases.
The most common mistake in feedless setups is uploading no video asset. Google then generates one automatically, and those auto-built videos can absorb a serious share of budget in low-quality impressions. Add at least one 10-second video with a vertical and a horizontal version. It is the cheapest way to avoid handing over control entirely.
If PMax and standard Shopping target the same products in the same account, PMax takes priority. Keeping standard Shopping running as a "backup" achieves nothing: either split the feed segment or switch it off.
When should Search or Shopping stay instead of PMax?
PMax does not suit every account. Classic campaign structures do better in these situations:
- Monthly conversion volume is very low. Smart bidding already struggles in an account producing single-digit weekly conversions, and PMax then spreads that data across channels. Build volume with Search first.
- Lead quality is the problem. If there is a wide gap between form fills and closed deals, do not move to PMax before feeding CRM data back as offline conversions. Otherwise the system optimises towards the cheapest, lowest-quality form.
- Keyword-level control is mandatory. Regulated sectors, competitor brand terms, narrow B2B vocabulary. Exact match Search remains the most precise instrument here.
- Feed quality is poor. Missing GTINs, weak product titles, wrong category mapping — PMax will not fix any of it, it will just spend against it. Clean Merchant Center first.
- The budget cannot even fill one campaign. If the daily budget is not a healthy multiple of your target CPA, PMax never leaves the learning phase.
The first week after inheriting an account
Do not restructure campaigns in an account you have just taken over. Confirm the measurement is correct first; every intervention built on bad data creates damage that is hard to reverse.
- Day 1 — Conversion tracking audit. Which conversion actions are marked primary? Is anything double-counted? Are add-to-cart or phone clicks sitting in the same bucket as purchases? An error found here costs more than any campaign-level error.
- Day 1 — Campaign settings. Final URL expansion, brand exclusion list, location targeting setting (presence versus presence-or-interest), account-level negative lists, content suitability and placement exclusions. All of it is half an hour of work.
- Day 2 — Asset group and listing group map. How many groups exist, what share of budget each takes, whether a single group absorbs 80% of spend. Pull the spend distribution at product level.
- Day 2 — Search term and category insights. Which query clusters convert and which only spend? Calculate the branded share; that ratio determines how you read everything else in the report.
- Day 3 — Channel breakdown. Use GAQL or a script to pull Search, Shopping, Display, Video and Discover distribution. If the Display share is higher than expected, look at asset quality first and target CPA second.
- Day 4 — Asset audit. How many headlines and images per asset group, is there a video, what do the asset performance labels say? Replacing assets rated "low" moves results faster than a bid adjustment.
- Day 5 — Overlap check. Are Search, Shopping and PMax competing for the same products or queries? Simplify the structure with the priority rules in mind.
- Days 6–7 — A single change plan. Do not fix everything you found at once. Pick the two highest-impact interventions, apply them, wait two weeks. Simultaneous changes in PMax make it impossible to measure what worked.
The logic of that order is measurement, settings, structure, creative. Most agencies start from the other end: they refresh the creative and discover in month three that conversion tracking was broken all along.
You hold fewer levers in PMax than in Search, but not zero. Exclusion lists, listing group segmentation, conversion value rules, the new customer acquisition goal, campaign-level negatives and asset quality are all direct points of intervention. In an account where all six are in use, "the algorithm won't tell us what it's doing" is usually an excuse for a structure that was never built.
The question to ask when setting up is not "PMax or Search". It is: given this budget, this data volume and this feed quality, which campaign type has to guess the least? The answer changes from account to account, and more often than not it is a structure where both run together.