Google Performance Max explained: a 2026 operator's guide

Performance Max is Google’s answer to the same question Meta answered with Advantage+: let the AI run delivery across all of our inventory from one campaign. PMax spends across Search, Shopping, YouTube, Display, Discover, Gmail, and Maps from a single setup, optimizing toward your conversion goal. It is powerful and it is opaque, and the operators who win with it are the ones who understand what they can still control. Here is the 2026 map.

What Performance Max actually is

A PMax campaign is a goal, a budget, an audience signal, and a set of asset groups. You do not pick placements or bid by keyword. You tell Google what a conversion is worth, hand it your creative assets, and it decides where and to whom to show them across the entire Google network.

The trade is control for reach. PMax can find conversions in inventory you would never have bought manually. It can also spend into low-quality placements or cannibalize your branded Search if you do not constrain it. The job of the operator is to feed it well and fence it sensibly.

The asset group is the unit that matters

PMax is driven by asset groups — bundles of headlines, descriptions, images, logos, and videos, optionally tied to an audience signal and, for retail, a product feed. The asset group is where your control lives, so it is where the work goes.

Feed it complete assets. PMax mixes and matches your assets across placements. Thin asset groups — a couple of headlines, one image — give the machine little to optimize and produce weak ads. Fill every asset slot, including video. If you do not supply video, Google auto-generates one, and the auto-generated version is rarely what you would have chosen.

Use multiple asset groups by theme. Separate asset groups for distinct products, audiences, or messages give you readable structure and let you judge what is working. One giant undifferentiated group tells you nothing.

Audience signals are hints, not targeting. The audience signal seeds the model with who you think converts. PMax uses it as a starting point and expands. Give it your best first-party and intent signals, then let it learn.

The video gap most accounts ignore

The most common PMax weakness is video. A large share of PMax inventory is YouTube and video placements, and most accounts either skip video or let Google auto-generate a slideshow from their images. That cedes a big slice of the campaign’s reach to creative you did not design.

Supplying real video for every asset group is one of the highest-leverage PMax moves, and it is exactly where AI video tooling earns its place — producing the volume of short video variants PMax wants without a per-asset production cost. For the YouTube-format specifics, see the AI YouTube ads guide; for the broader creative-throughput field, the 2026 ranking of AI ad creative tools maps the tools, Superscale among them.

Keeping control of spend

PMax’s opacity is the real operator risk. Three guardrails matter most in 2026:

• Brand exclusions and brand campaigns. Without brand exclusions, PMax will happily claim your cheap branded Search conversions and report them as its own. Run a separate branded Search campaign and exclude brand terms from PMax so you can read its true incremental value.
• Account-level placement and content exclusions. Use them to keep spend off low-quality content and apps. PMax will not do this for you by default.
• Value-based bidding where you can. Feeding PMax conversion value, not just conversion count, lets it optimize toward profitable customers rather than cheap ones. This matters most for accounts with a wide range of order values.

Reading PMax performance honestly

PMax reporting is thinner than a manual campaign’s, and the attribution question is sharper because it spans so many placements. Do not judge PMax on its in-platform ROAS alone — judge it on incremental conversions against a holdout or against your blended performance before and after. The marketing attribution guide covers the models you need to read this correctly, and the ROAS playbook covers the calculation and benchmarking.

A practical PMax setup

For a mid-volume ecommerce or lead-gen account:

• One PMax campaign per major goal, not one per product. Let the asset groups handle product separation.
• Complete asset groups by theme — every slot filled, real video supplied, not auto-generated.
• Strong first-party audience signals seeding each group.
• Brand exclusions on, with a separate branded Search campaign running alongside.
• Value-based bidding if your order values vary.
• A holdout or before-and-after read to judge incremental value, not just the platform’s self-reported ROAS.

FAQ

What is Google Performance Max?

Performance Max is a Google Ads campaign type that runs across all Google inventory — Search, Shopping, YouTube, Display, Discover, Gmail, and Maps — from one campaign, optimizing toward a conversion goal using your asset groups and audience signals.

Is Performance Max worth it?

For most accounts with conversion volume, yes — it finds conversions across inventory you would not buy manually. It needs guardrails: brand exclusions, complete asset groups, real video, and an honest incremental read on performance.

Should I supply my own video to PMax?

Yes. A large share of PMax inventory is video. If you do not supply video, Google auto-generates a weak version. Supplying real video for every asset group is one of the highest-leverage PMax moves.

How do I stop PMax from cannibalizing branded Search?

Apply brand exclusions to the PMax campaign and run a separate branded Search campaign. This keeps PMax from claiming cheap branded conversions and lets you read its incremental value.

Related reading

• The ROAS playbook — calculating and benchmarking return.
• Marketing attribution models explained — reading cross-placement performance honestly.
• How to make AI YouTube ads — the video format PMax under-uses.
• Static ads in 2026 — the image assets that feed asset groups.
• The best AI ad creative tools in 2026 — the creative-throughput field guide.