Performance Max is no longer the black box of 2022. In three years Google has opened up nearly everything advertisers asked for: channel-level breakdowns, search terms, campaign-level negative keywords, brand exclusions, and since 2026 audience exclusions built from your own customer lists. So the question has changed shape. It is no longer "must I accept seeing nothing?" but "what do I do with what I can now see?".
That is where most of the accounts I audit go wrong. The reports are open and nobody reads them, or worse, they are read with classic Search reflexes, and decisions get made that actively degrade performance. Taking back control of Performance Max in 2026 is not about hobbling the algorithm. It is about knowing which settings are guardrails (set once, early) and which reports are signals, to be read often but slowly.
A badly framed Performance Max campaign does not deliver poor performance. It delivers excellent performance on conversions you would have had anyway.
What Google actually opened up, and what it did not
The turning point was April 2025, when Google announced channel performance reporting, full search terms reporting and richer asset metrics, three requests the paid-search community had been making since the format launched. The channel performance report finally splits clicks, cost and conversions across Search, Shopping, YouTube, Display, Discover, Gmail, Maps and search partners, and the distribution table is downloadable.
In 2026 Google added another layer of steering and reporting: first-party audience exclusions (remove a customer list from the campaign, typically to focus budget on acquisition rather than on people who already converted), a budget report projecting end-of-month spend, full audience reporting with age range and gender, and network segmentation in placement reporting, now reachable from the "When and where ads showed" tab.
What was not opened up deserves to be said plainly: you still do not choose how budget is split across channels. There is no per-channel bid, no way to switch YouTube off to fund Search. Performance Max remains a campaign that arbitrates on its own. Everything 2025 and 2026 brought is observation and constraint, never allocation. That distinction matters: it explains why some optimisations people assume exist simply do not, and why the real levers sit elsewhere.
Brand exclusions: the first setting, before you read any ROAS
If I could keep only one setting, this would be it. By design, Performance Max hunts for the cheapest conversions, and in almost every account the cheapest conversions are brand queries. A prospect typing your name and converting at a fraction of your usual cost per acquisition lifts the average beautifully. Except you already had that conversion: it would have come through your brand Search campaign, or organically.
Google offers two distinct tools that are routinely confused. Brand exclusions stop the campaign from serving on queries containing your brand and its variants, misspellings included; campaign-level negative keywords handle irrelevant non-brand queries instead. The first are set once, at launch. The second are fed by the search terms report, reviewed every two or three weeks.
Order matters. Until brand exclusions are live, no number in the campaign is interpretable: not ROAS, not cost per acquisition, not any comparison with other campaigns in the account. That is why I decline to comment on a Performance Max campaign's performance before checking this, and why a fair share of "miraculous PMax" stories deflate within a week.
Reading the channel report without drawing the wrong conclusion
Channel reporting produced a predictable side effect: the moment an advertiser sees YouTube eating 30% of budget at three times the cost per conversion of Search, they want to cut YouTube. That is precisely the reading Google warns against, and for once the argument holds.
Performance Max does not optimise each channel's average return but the marginal return of the next euro: at any moment it places that euro where it adds the most. A channel's average return, observed over a short window and on a last-click basis, therefore tells you nothing about whether spending there was right. An exposure channel with an ugly cost per conversion may be the one feeding the brand searches Search will convert three days later.
What the channel report is genuinely for is more modest, and more useful: diagnosis. An abnormally low Shopping share points to an incomplete product feed rather than a bidding problem. Absence on Maps signals unlinked store locations. A weak Search share with off-topic search terms points to misaligned landing pages. The report does not tell you what to cut. It tells you what to fix upstream. It is also why I push these breakdowns into a Looker Studio dashboard rather than reading them in the interface: the month-over-month gap says far more than today's figure.
The levers that actually steer
In practice, on an SME account, five settings account for most of the difference. The first four are guardrails; the fifth is groundwork.
| Lever | What it changes | The trap |
|---|---|---|
| Brand exclusions | Takes brand queries out of scope so the numbers become readable | Setting them after three months, then panicking at the drop in reported ROAS |
| Campaign negative keywords | Cuts the off-topic themes surfaced by the search terms report | Adding hundreds of them until the campaign suffocates and prospecting stops |
| First-party audience exclusions | Removes converted customers so budget concentrates on acquisition | Excluding the whole base in a model where repeat purchase carries the margin |
| Final URL expansion | Decides whether AI picks the landing page based on detected intent | Leaving it on with a site where half the pages never convert |
| Asset and feed quality | Determines what the campaign can serve, and therefore where it can win | Expecting the algorithm to compensate for three visuals and one generic headline |
Final URL expansion deserves a note. It is presented as help (AI picks the page best matching detected intent) and on a deep, well-structured site where nearly every page carries commercial value, it genuinely is. On an SME site with orphan pages, a non-commercial blog and outdated listings, it sends paid traffic into dead ends. My rule: on for catalogues, off anywhere the number of pages that actually convert fits on one hand.
The most expensive mistake
Judging Performance Max on its reported ROAS and raising budget accordingly. It is the scenario I meet most often, and the mechanics are relentless: the campaign absorbs brand conversions, ROAS climbs, the advertiser triples budget, the campaign exhausts the stock of easy conversions and goes hunting for volume elsewhere, ROAS collapses, and the advertiser concludes that "PMax stopped working". Nothing stopped working: the initial figure never measured what everyone assumed. Setting brand exclusions before reading anything costs ten minutes; skipping it costs a quarter of misdirected budget.
What Performance Max will never fix
One last point, and it decides success more reliably than every setting combined: the quality of the conversion signal. Performance Max optimises towards whatever you declare as a conversion. If that signal is noisy (duplicates, events firing on a reloaded thank-you page, contact forms counted like orders), the campaign will diligently learn to manufacture noise. The prerequisite is not advertising but technical and analytical: one conversion per action that carries value, a monetary value where one exists, and a measurement plan that separates raw leads from qualified ones.
The same goes for the campaign's role in the wider setup. Performance Max is an execution engine, not an acquisition strategy. On the accounts I run it almost always coexists with a brand Search campaign kept separate, sometimes with a Demand Gen campaign for latent demand, and it is read against the other channels, which is the whole point of paid search management that does not stop at one interface. "PMax or Search?" is the wrong question. The right one is: what is left for Performance Max once you remove everything that already belonged to someone else?
In the crucible: distilling control into a campaign that decides alone
The lead, here, is the Performance Max campaign launched in three clicks (no brand exclusions, final URL expansion on by default, an approximate product feed), a campaign that posts excellent numbers for six weeks and becomes inexplicable in the seventh.
The gold is the same campaign with four guardrails set on day one, a channel report read as a diagnosis rather than a verdict, and a conversion signal clean enough for the algorithm to learn the right thing. Google spent three years making Performance Max legible; the alloy is now forged on the advertiser's side, in the discipline of what you give it to optimise. A machine that decides alone does not spare you from knowing what you are asking of it. It demands that you know.
Find out what your Performance Max is really measuring
I check the guardrails (brand exclusions, negatives, URL expansion), read the channel report as a diagnosis, and clean up the conversion signal before anyone talks about budget.