Here’s a clear-eyed look at a question many marketers are asking this year: in 2025, do dynamic, AI-assembled creatives outperform carefully crafted manual ads on ROAS—or is the opposite true? Short answer: it depends on your context. Longer answer: you’ll likely get the best ROAS by using both, but for different jobs.

Dynamic creatives: what they’re good at (and why they’ve improved)

Dynamic creative systems (Meta’s Advantage+ creative, Google’s Performance Max, many DCO engines) now remix copy, visuals, and placements automatically, then allocate spend toward the best-performing combinations. They’ve gotten markedly better in the last 12–18 months. On Google’s side, advertisers who adopt Performance Max see, on average, about 27% more conversions or conversion value at a similar CPA/ROAS, which is why PMax has become a default for a lot of teams. 

Meta has also been upgrading the creative layer. Advantage+ creative’s “enhancements” and AI options have been unbundled and refined in 2025, giving advertisers more granular control over which tweaks (e.g., background expansions, text variations, image adjustments) to enable. The platform’s own documentation confirms the shift away from the older “standard bundle” toward pick-and-choose enhancements, making the feature set less of a black box.  Meta has simultaneously been touting performance gains from these AI tools, emphasizing how the system can amplify creative variations at scale. 

Why this matters to ROAS: dynamic systems shine when (1) your catalog or message matrix is large, (2) you need velocity in creative testing, and (3) the auction is changing faster than your team can manually respond. In practice, that means retail and app commerce often see earlier wins from dynamic mixes, especially in prospecting and broad retargeting.

Manual ads: still the king of “edge” and brand nuance

Manual ads—where humans set the concept, craft the copy, pick the footage, and control the layout—continue to win when brand nuance, specific benefits, or a precise offer structure matters. In 2024–2025 case work, we still see manual structures beat automated mixes during high-stakes windows (launches, big promos) because the creative tells one coherent story and the account structure forces clean learning. One public case study reported a 423% increase in ROAS during Black Friday by restructuring into segmented, manually managed ad sets and using Advantage+ as a complementary discoverability layer—not the main driver. That hybrid approach matters: manual captured intent with precision, while automation expanded reach. 

Manual ads also excel when:

• You have strong first-party insights and want to encode them into the creative narrative (not just feed them as data).

• You need compliance or brand-safety guardrails that generic AI transformations might violate.

• You’re trying to seed a new, distinctive visual world; AI variants can dilute a strict art direction.

So which drives better ROAS in 2025?

If you’re running always-on acquisition at scale, dynamic creatives often win on efficiency and learning speed. They’re particularly effective in large product catalogs, multi-market accounts, and campaigns where you simply cannot test every permutation manually. On the other hand, if you’re running time-boxed offers, premium positioning, or complex value props, manual ads frequently deliver higher ROAS—especially when you pair them with audience and budget control that keeps tests clean and messaging tight.

It’s also worth noting that investor and industry analyses comparing Meta Advantage+ and Google PMax show that both ecosystems can deliver strong ROAS, but performance varies by vertical and creative quality. The headline isn’t “automation beats manual”; it’s “good creative plus the right level of automation beats everything else.” 

How to decide (and what to test)

1. Split roles, not teams. Use dynamic for exploration; manual for exploitation. Set up prospecting with dynamic systems to discover winning angles, then rebuild the winners into branded, manual assets for scaling and promos. This aligns with how Meta is positioning its AI tools today—broad discovery plus creative iteration—while letting your brand voice remain intentional. 

2. Protect your creative integrity with selective enhancements. Since Meta now lets you opt in to individual Advantage+ tweaks, turn on only what helps (e.g., background expansion for feed fit) and turn off what warps brand visuals or claims. This strikes a balance between performance and brand control. 

3. Use dynamic to find the “why,” then codify it manually. When a PMax asset group or Advantage+ variation outperforms, don’t just scale it—translate the learning into a deliberate concept: headline hierarchy, opening three seconds of video, product framing, proof points, and CTA logic. Google’s PMax performance guidance underscores that creative breadth plus structured data often unlocks incremental conversions at steady ROAS—exactly the kind of signal you can then harden into manual ads. 

4. Benchmark with event-level economics, not CTR. For both modes, look at incremental revenue, blended CAC/ROAS, and payback windows. Aggregated studies show strong ROAS potential across both Meta and Google when the creative and data foundations are sound; your goal is to identify which mode gets you there with less volatility in your category. 

The bottom line

In 2025, neither dynamic nor manual is categorically “better” for ROAS. Dynamic creatives are your rapid-learning engine: they surface what works, fast, and keep you competitive in fluid auctions. Manual ads are your precision instrument: they turn those learnings into brand-safe, high-intent stories that compound. Teams that separate these jobs—and let each mode do what it does best—consistently see the highest, most defensible ROAS over the quarter, not just the week.

If you’re choosing one place to start: launch a broad dynamic test bed (Meta Advantage+ creative or Google PMax) to map winning angles, then spin out your top two into manually crafted concepts for your next promo cycle. Track incremental lift and payback. If the manual spinouts can hold or exceed the dynamic baseline with lower volatility, you’ve found your ROAS flywheel. If not, your category may favor dynamic exploration longer—keep iterating through the machine until the signal gets obvious enough to script.