Skip to content
Primores AI wiki

Google Performance Max (PMax): The AI Campaign That Runs Across All of Google

Google Performance Max (PMax)

TL;DR: Performance Max is a single, goal-based Google Ads campaign type where you supply a conversion goal, creative asset groups, and audience signals, and Google’s AI controls bidding, placement, audience, and creative assembly across all Google inventory — Search, Display, YouTube, Gmail, Discover, Maps — from one campaign. It’s the Google-side structural twin of glossary/advantage-plus: assets + goal in, ML-run delivery out. Launched Nov 2021, it forcibly replaced Smart Shopping and Local campaigns in 2022. Like Meta, Google shipped it as a near-total black box, then walked back transparency under advertiser pressure (2023–2025). The two load-bearing critiques: it cannibalizes branded-search traffic (inflating its attributed numbers) and it optimizes toward existing demand, so attributed return overstates incremental lift.

What it is

PMax collapses what used to be many campaign types into one. You give Google:

  • a conversion goal (and optionally a Target CPA or Target ROAS),
  • asset groups — the creative raw material,
  • audience signals — hints that steer the machine learning, not hard targeting limits.

Google’s AI then assembles per-placement creative and runs the auction across every Google surface. The advertiser’s manual levers (placement, audience, bid-by-segment) are largely handed to the algorithm — the same trade Meta’s Advantage+ makes.

Anatomy and mechanics

ElementWhat it is
Surfaces (from one campaign)Search, Display, YouTube, Gmail, Discover, Maps
Asset groupUp to 15 headlines, 5 descriptions, 20 images, 5 videos (+ logos); AI assembles combinations per placement
Audience signalsHints that “accelerate and steer” the ML — the system can serve outside them (Google’s own framing)
BiddingSmart Bidding (Maximize Conversions / Value), optionally constrained by Target CPA / ROAS, auction-time

Audience-signals-as-hints is the exact parallel to Advantage+ audience inputs becoming “suggestions” — the data foundation for granular manual targeting eroded after glossary/app-tracking-transparency, so both platforms moved control to the model.

Dated history

  • GA: November 2, 2021 — rolled out to all advertisers worldwide.
  • 2022 forced migration: Smart Shopping (Jul–Sep) and Local (Aug–Sep) campaigns were auto-upgraded to PMax; both fully migrated by end of Sept 2022. (Notably earlier than Meta’s API-level ASC/AAC enforcement in 2025–26.)

2023–2026: Google adds back transparency/control

Under advertiser pressure, Google walked back the pure black box — the strongest parallel to Meta’s own reversal:

  • Search Themes (Nov 2023 beta) — an actionable input replacing passive search-category insights; up to 50 per asset group.
  • Campaign-level brand exclusions (2024) — stop PMax serving on your own brand queries.
  • Search-term insights (2024) — see whether queries came from keywordless targeting vs your search themes.
  • Channel-level performance reporting + account-level negative keywords (rolled out 2025) — the transparency concessions advertisers had demanded for years.

Sibling launch: AI Max for Search exits beta (June 2026)

Distinct from PMax but the same model-control trade applied to Search campaigns: AI Max for Search went generally available in June 2026 (out of beta after ~11 months; Google Ads blog updated 11 Jun). AI Max layers AI onto a standard Search campaign — it goes “beyond landing-page data to include broader, real-time intent signals” to find untapped queries, and adds brand, location, and text-guideline controls that legacy Dynamic Search Ads (DSA) lacked. It’s effectively the DSA successor with the post-ATT “let the model find demand” pattern and a thin layer of advertiser guardrails bolted back on — the same arc as PMax’s transparency walk-back.

Do-not-cite: the claim that Google is fully sunsetting DSA on a Sept-2026 / Feb-2027 timeline was REFUTED (0-3) in verification — the auto-migration timeline slipped and a clean “sunset” framing didn’t hold. Ingest only the beta-exit / GA fact, not a DSA-sunset date. (Beta exit: 3-0; Google Ads blog + PPC News Feed “GML 2026.”)

What Google claims for it — and the caveats

  • “Average 13% increase in total incremental conversions at a comparable cost per action.” (A newer Google resource cites 18% for “advertisers making the most of” PMax — note the moving qualifier.) Case studies range 15–39% conversion lift / 15–83% cost reduction.

The caveat (mirror the Advantage+ wording): despite Google’s use of the word “incremental,” these are Google self-reported, attribution-based figures with undisclosed methodology — not holdout/incrementality-tested. The cannibalization evidence below is exactly why the “incremental” label is contested.

The core critiques

  1. Black box. PMax launched as “the blackest black box of all Google ad products” — no breakdown by channel, format, placement, or creative. Google conceded the point in April–May 2025 by shipping channel-level reporting in beta: impressions, clicks, conversions, and cost per channel (Search, YouTube, Gmail, Maps, Discover, Display, Search partners) — with no historical backfill (data only from June 6, 2025 onward) (verified 3-0 against Google’s own announcement). Even post-concession, the reporting is criticized as descriptive-not-actionable (no placement-level bid control, can’t exclude Display/YouTube) — and the practitioner pain has shifted from “no data” to “what do I do with the new channel report”, where canonical guidance doesn’t exist yet. Google’s rebuttal exists and is worth weighing for balance.
  2. Branded-search cannibalization (best-quantified). An Optmyzr study of 503 accounts (Feb 2025) found 91.45% had keyword overlap between Search and PMax; where they overlapped, Search beat PMax on conversion rate in 18.91% of cases vs PMax winning 6.17%. Optmyzr explicitly notes it couldn’t measure ROAS because PMax doesn’t report revenue at the search-term level. Mechanism: PMax absorbs high-intent branded searches that would have converted anyway, inflating its attributed numbers.
  3. Limited channel control. No placement-level bid control, no native way to exclude Display or YouTube; control is partial even after 2025.
  4. Attribution ≠ incrementality. PMax optimizes toward existing demand and counts conversions on an attribution basis, not causal lift — the identical critique structure to Advantage+. Validate with glossary/incrementality-testing.

Key Takeaways

  • PMax is Google’s Advantage+ — same trade: assets + goal in, AI runs targeting/placement/bidding/creative.
  • GA Nov 2021; forced-migrated Smart Shopping & Local in 2022.
  • Audience signals are hints, not targeting — the post-ATT model-control shift.
  • Google added back transparency (2023–25) after launching it as a black box — the same arc Meta walked.
  • Two critiques to carry: branded-search cannibalization (Optmyzr, 503 accounts) and attributed ≠ incremental return.

Sources

Do-not-cite: the “+7%/46%/3% true-incremental” figures (couldn’t trace to primary — direction is solid, numbers aren’t); older “73% overlap / 34% higher CPC” summary figures (superseded by the verified Feb-2025 Optmyzr 91.45% set).

By Andrej Ruckij