How do I use custom segments for more precise targeting inside PMax?

Before treating this as a bug — showing outside a set region is, in most cases, the expected behavior of the default setting, not a broken rule. The default location setting, "Presence or Interest," shows ads to people physically in your targeted location, regularly there, or simply showing interest in that location through their queries or behavior — even if they're actually somewhere else. Google's own example: target Paris, and the ad can reach someone physically in Paris and also someone elsewhere who's just interested in Paris bakeries.

The first official indicator for this split is Ad Strength at the asset group level: use the Ad Strength indicator to determine whether an asset group has enough assets to drive best performance, and make sure every asset type — text, image, video — is covered and meets quality guidelines. If Ad Strength is low, that's a direct pointer toward the creative side.

Before switching the feature off entirely, it helps to understand the mechanism and the middle-ground options available. Final URL expansion, on by default, lets Google replace your Final URL with a more relevant page on the same domain based on the user's search intent — meaning irrelevant landing pages are often a side effect of expansion running without exclusions, not a malfunction.

The core difference from classic keywords is built into the tool's purpose: search themes let you flag queries your customers use are searching for, but they're optional and additive — layered on top of the queries and placements PMax already predicts will perform well based on your assets, feeds, and landing pages. That's a fundamentally different logic from Search: a keyword there is a serving condition; a search theme here is a hint layered on top of a model that will keep searching on its own regardless.

The direct, officially documented cause is described in the context of retail campaigns, but the logic applies more broadly: asset groups ideally shouldn't overlap on targeted products — Asset Group 1 and Asset Group 2 shouldn't both target products A–Z; instead, one should own A–L and the other M–Z. If that overlap exists, both groups are literally competing for the same traffic inside one campaign — and since asset groups aren't independent campaigns with their own budget and strategy, they don't split that traffic evenly; they compete for it under the campaign's shared auction and optimization rules.

There's no dedicated "signal overlap between asset groups" report — worth stating plainly so you're not searching for a tool that doesn't exist. But overlap can be assessed indirectly through asset group reporting: it lets you review status, Ad Strength, conversions, conversion value, and audiences across each asset group in the campaign — an audience column exists, it just has to be cross-referenced between groups by hand rather than read off as a ready-made overlap metric.

The feeling of limited control isn't an accident or a gap — it's a direct consequence of the product's architecture. PMax runs on "keywordless targeting" — Google AI and machine learning find and serve ads to likely converters using signals from the campaign, site, and feeds, instead of manual keyword management. That's a fundamentally different model from Search, where the advertiser directly controls the serving condition through keywords.

The key fact that shapes this whole answer: placement exclusions for PMax only work at the account and MCC — there's no campaign-level setting yet. Unlike Display or Video campaigns, where exclusions can be managed per campaign, the only official path for PMax is setting exclusions account-wide, and they'll automatically apply to every PMax campaign in that account at once.

The fact that explains the limited effect is stated directly in official documentation: PMax negative keywords are applicable to Search and Shopping — adding negative keywords to a campaign or account helps avoid serving on matching queries specifically on those two surfaces. PMax, meanwhile, shows across a much wider set of channels: six Google channels — and negative keywords structurally do nothing on any of those channels except Search and Shopping.

Google's own troubleshooting checklist for a campaign that's "hit a ceiling" names several independent categories of cause, split between targeting and creative: check whether CPA/ROAS targets are too restrictive — that's a bidding/targeting issue, blocking auction participation; check whether there are enough creative assets of every type and size, aiming for an "Excellent" Ad Strength — a purely creative category; try adding or refreshing search themes — a semantics/targeting category; and audit whether account- and campaign-level exclusions are limiting reach — also targeting. In other words, Google's own advice is to work through independent checkpoints in sequence rather than guess at one overall cause.

Google's stance here is direct and unambiguous: consolidate campaigns where you can — that gives Google AI more data and performance history to sharpen its predictions — and only create separate campaigns when there's a real business reason: different goals, budgets, or CPA/ROAS targets. In other words, splitting by product category isn't a default best practice on its own — it's an exception that needs justifying, not a habit inherited from how a catalog happens to be organized.

Worth establishing the current scope of the tool first: demographic exclusions in PMax are currently limited to age and gender — an initial rollout aimed at the most critical demographic-exclusion needs. If expectations were broader (household income, parental status, for example), those simply aren't available for PMax yet.

Fact-check: Most likely, "excluded" here meant removing a segment from audience signals, not applying an actual exclusion tool — and those are fundamentally different actions. Audience signals are hints, not rules: PMax can serve outside your signals if it predicts a strong chance of conversion. Removing a segment from signals only stops suggesting it as desirable to the model — if the model, independently of signals, keeps finding converters in that segment, it will keep serving there, because a signal was never a blocking mechanism to begin with.

This is an officially recognized and documented scenario: PMax campaigns can maximize total sales and revenue from both new or existing; if new customers are the priority, the official recommendation is the new customer acquisition goal with New Customer Value mode, bidding first toward the most valuable prospects. Google is effectively acknowledging that, without a dedicated setting, PMax optimizes for total conversion value without distinguishing "new" from "already existing" customers — and that's exactly what creates the feeling of "cannibalization" of traffic that would have converted through another channel anyway.