Saying AI has changed advertising has become a cliché. More useful is to understand when AI actually wins and when it does not. Google Performance Max and Meta Advantage+ matured enough in 2024-2025 to be, in many cases, better than manual campaigns. But «many cases» is not «all cases». This article separates the conditions, so that a marketing director who has read too many LinkedIn threads and too many pumped case studies can decide where to invest control and where to trust the algorithm.
What changed since 2023
Three concrete shifts moved the center of gravity between 2023 and 2026.
First: Performance Max became Google’s default. Anyone still running only manual Search + Display on Google Ads is paying a premium — in management time and allocation inefficiency. PMax has access to inventory manual campaigns do not see: Gmail, Maps, Discover with multi-asset ads.
Second: Advantage+ Shopping Campaigns left the experimental phase. Meta stopped pitching it as «try this» and made it the skeleton of serious e-commerce campaigns. Manual Placement is kept today only for granular control on cold audiences or to test fresh creatives.
Third: creative diversification moved from nice-to-have to requirement. A Meta account with three or four active creatives per ad set gets drastically worse results than one with fifteen to twenty well-produced variants. Advantage+ AI needs material to learn from.
When AI wins
AI-first campaigns systematically beat manual in four conditions.
Enough data volume
Performance Max has a 30–45 day learning phase. During this phase you need a solid conversion floor: 50+ conversions/month as a minimum, 200+ as sweet spot. Below that, AI does not learn: it optimizes on noise and produces erratic CPAs.
Feed and signal quality
A product feed with 80 complete attributes over 10k SKUs works with PMax in a radically different way than one with 40 dirty attributes over the same SKUs. Likewise, Enhanced Conversions, Consent Mode v2, Server-Side tagging (GA4 + GTM Server) determine signal quality more than any targeting optimization.
Creative density
40–80 creative variants per sprint is not a luxury: it is the quantity under which Advantage+ stops working as intended. Agencies shipping five creatives a week are in an asymmetric relationship with an algorithm built to test five hundred.
Stable and relevant primary conversion
If you optimize on a conversion that does not correlate with real revenue — low-quality lead forms, view content, fake engagement — AI will bring you more of that wrong thing. It is the most frequent case of «PMax does not work»: usually meaning «PMax did exactly what you asked, but what you asked was not the right thing».
When AI loses
Two scenarios where automation is still inferior to manual.
New or niche audiences
To launch a product to an audience that has not been served yet, AI has no data. Manual campaigns with tight interest targeting, specific creatives and modest budget (100–500 €/week) are more efficient in the first 60 days. Then, after accumulating 50–100 conversions, you can migrate to Advantage+ with warm prospecting.
Pure brand storytelling campaigns
PMax and Advantage+ optimize on defined conversions. For brand campaigns measuring lift, awareness, semantic associations — not direct conversions — you need different formats: Reach & Frequency on Meta, Demand Gen on Google, brand lift studies. The AI of these platforms is tuned for performance, not brand.
Three recurring mistakes
- Brand traffic cannibalization. PMax eats branded traffic unless segmented. Symptom: very low CPAs in the first 15 days, then overall performance collapse. Fix: separate brand Search campaign and exclude the brand as keyword from PMax.
- Mixing goals in one campaign. Putting purchase and lead generation in the same PMax is a reliable way to get campaigns that do not learn well. One goal per campaign, always.
- Blind trust in the algorithm. AI is a very capable assistant, not an autopilot. A marketing director who cannot read an Asset Group report, interpret a PMax placement, or judge a conversion signal’s quality is a blind pilot on a sophisticated plane.
How work is split today
The 2026 performance marketer does three things differently from the 2021 one.
- Curates signals: CAPI/Enhanced Conversions integration, funnel validation, tracking cleanup — 40% of the work.
- Produces creative at volume: brief, generation, QA, retiring — 30%.
- Aligns campaigns and business: per-channel budget, FAIR ATTRIBUTION, readable dashboards — 30%.
Time spent on tactical campaign optimization — bid strategy, negative keywords, manual ad copy rewrites — has collapsed from 60% to 10%. Anyone still spending 60% on that is not doing performance marketing: they are doing work AI does better, at zero cost.
What to verify before going AI-first
Four concrete questions for your Google Ads or Meta Business account today:
- Do you have Enhanced Conversions live and validated? (Google)
- Do you have CAPI live with match quality ≥ 7/10? (Meta)
- Do you have at least 40 fresh creative variants in the last 30 days?
- Is the primary conversion you optimize for the one that actually correlates with revenue?
If any single answer is no, migrating to Advantage+/PMax is premature. Priority is fixing the fundamentals.
What will change in 2026-2027
Two patterns we are observing in our early tests.
Multi-modal generative ads at scale. Meta and TikTok are integrating native AI video generation into the Creative Hub. This drastically lowers production cost but raises the pressure on brand governance: whoever lacks an LLM-digestible codebook will ship visually incoherent creatives.
Unified attribution via Media Mix Modeling + geo-lift. Platforms will keep telling you they drove the conversions. Whoever spends more than 500k €/year in ads will need an independent MMM to put platform numbers back in order. An enterprise-only project for years; in 2026 it becomes accessible to mid-market too, thanks to open-source tools (Meta Robyn, Google LightweightMMM).