In Q4 2025, Google announced it had generated over 70 million creative assets through its AI tools for advertisers. Seventy million. In a single quarter.

That’s a number that stops you. Because it means the creative production arms race has fundamentally changed. Brands that were competing on the volume of content they could produce manually are now competing in an environment where production is no longer the bottleneck.

What is the bottleneck now? Creative quality, brand judgment, and strategic direction. The things AI still can’t do on its own.

A Client Story: From 6 Variants to 47 in 90 Days

In mid-2025 we onboarded a fashion accessories brand — mid-market, primarily Meta-dependent, with a creative team of two designers and one copywriter. Their biggest paid media constraint wasn’t budget. It was creative.

They were testing 5-7 ad variants per month. By the time a variant had enough data to evaluate (usually 3-4 weeks in Meta), the winning insights were already late. They were optimising retrospectively, not prospectively.

We introduced a generative AI creative workflow using three tools: Pencil for ad concept generation, Meta’s Advantage+ creative for in-platform image and copy variation, and Runway ML for short-form video variant generation. Ninety days later:

• 47 ad variants tested 6 in the prior 90-day period
• Time per variant: dropped from 4-6 hours (design + copy + approval) to 45 minutes (AI draft + human review + minor edits)
• Creative team capacity: freed from execution to creative direction and quality review
• Performance: ROAS improved 3% — primarily attributable to finding better creative faster, not to AI quality being inherently superior

The last point is worth emphasising. Generative AI didn’t make better ads. It made finding good ads faster by enabling more tests in the same time period.

How the Generative Creative Stack Works

There are four distinct layers where generative AI is now being used in ad creative production:

Layer 1: Copy and Headline Generation

The most mature and widely used layer. Tools: Google’s Asset Generation, Meta’s AI copy suggestions, Jasper, Copy.ai, and similar.

What AI copy generation does well:

• Producing multiple variations of the same core message quickly
• Generating alternative angles on a product benefit (price, quality, speed, social proof)
• Adapting copy to different formats (short headline, longer body, CTA variation) What it does poorly:
• AI copy is trained on patterns — it produces average, not exceptional.
• Brand voice Without strong examples and prompting, AI copy sounds generic.
• Cultural nuance in local English AI copy often loses something in Indian market adaptation.

Our standard workflow: AI generates 10-15 headline options and 8-10 body copy variants. A copywriter reviews, selects 4-6 with genuine potential, edits them to voice, and writes 2-3 originals that the AI didn’t generate. The final set — 6-9 variants — has higher average quality than AI-only or human-only would produce.

Layer 2: Static Image Generation

Tool landscape: Adobe Firefly (best for brand-consistent generation with existing assets), Midjourney (best raw image quality), DALL-E 3 via ChatGPT (most accessible), Google’s Imagen (integrated in Performance Max).

For e-commerce: AI image generation is genuinely useful for lifestyle imagery around products — the coffee being enjoyed in a well-lit kitchen, the laptop in an airport lounge. These are expensive to shoot repeatedly but relatively easy to generate. We’ve seen clients reduce lifestyle photography costs by 60-70% using AI-generated backgrounds with real product photos composited in.

For brand campaigns: AI image generation struggles with consistency. The AI doesn’t have your brand’s ‘visual language’ — the specific way you use colour, framing, and subject treatment that makes an ad instantly recognisable. Consistent brand visual identity still requires human creative direction.

Layer 3: Video Generation

The frontier, and the most hyped. Tools: Runway Gen-3, Google’s Veo 2, OpenAI’s Sora, Meta’s Movie Gen.

Where AI video is practically useful in 2026:

• Short (<6 second) product demonstration clips
• Background animation on static ads
• Simple product rotation and showcase videos
• Adapting a primary video into multiple aspect ratios (9:16, 1:1, 16:9) Where it still falls short:
• Narratively complex or emotionally nuanced video
• Consistent characters or brand spokespeople across scenes
• High-production-value brand campaigns (the uncanny valley problem persists)
• Anything requiring realistic human behaviour in complex situations

For most performance marketers, the immediately useful AI video application is in the last category: format adaptation. Taking one professionally shot hero video and

automatically generating story (9:16), square (1:1), and landscape (16:9) variants saves 3-4 hours of editing per asset. Not glamorous, but genuinely valuable.

Layer 4: Personalisation at Creative Level

Dynamic Creative Optimisation (DCO) — the layer where AI assembles different creative components for different users — is the bridge between creative generation and performance marketing.

With a strong generative AI creative pipeline feeding the DCO layer, you get a compounding effect:

• More generated variants to test
• Faster learning about what works for which audience segments
• Faster brief generation for new creative cycles based on what the algorithm is favouring

One client’s creative director described it as ‘having the algorithm as a research assistant that never sleeps.’ The DCO data tells her what creative signals are driving performance. She briefs the next creative cycle based on those signals. AI generates initial variants. She edits, approves, and launches. The cycle runs 3x faster than before.

The Quality Reality Check

Here’s the part of generative AI creative content that doesn’t get talked about enough: the average quality of AI-generated ads is lower than the average quality of human-crafted ads from experienced teams.

This is documented. In a 2025 meta-analysis of DCO performance across 140 advertiser accounts, human-crafted ‘hero’ creatives outperformed AI-generated creatives on click-through rate by 24% and on conversion rate by 17% when evaluated individually.

The reason AI creative outperforms in aggregate results is volume. AI enables you to test 20 concepts where previously you’d test 5. If the hit rate for a good ad is 1 in 10, AI-enabled testing finds 2 good ads where manual testing finds 0-1. The volume advantage compensates for the individual quality gap.

This changes the creative team’s job description. The high-value skill is no longer producing ads — it’s identifying which AI-generated concepts have potential, elevating them, and applying brand judgment to reject the many that don’t make the cut. Curation and direction replace production as the primary creative skill.

The Brand Safety and Legal Considerations

Before deploying generative AI at scale for ad creative, three things need to be addressed:

Image rights and training data: Some AI image generators have faced legal challenges about the content they were trained on. Adobe Firefly’s explicit commercially-safe training data makes it the safest choice for brand advertising use cases.

Brand consistency governance: Who reviews AI-generated content before it goes live? AI can produce off-brand, tone-deaf, or factually incorrect content. A clear review process with a named human accountable for final approval is non-negotiable.

Platform policy compliance: Meta and Google both prohibit certain types of AI-generated content (particularly misleading AI-generated people or fake testimonials). Ensure your generative AI workflow has a compliance review step.

The Practical Implementation Path

• Start here: Use AI for copy variation first. Lower stakes, immediate value, no legal complexity. Take your best-performing ad copy and use AI to generate 10-15 alternative angles. Test them. See what the data tells you.
• Then add: AI image generation for lifestyle and context imagery. Not for your hero product shots — for the environments and contexts your product lives in.
• Then add: Format adaptation using video AI tools. Take your existing video assets and generate the variants your current platform needs.
• Later: Full DCO integration where AI-generated assets feed directly into platform creative testing.

Full implementation timeline for a typical mid-market brand: 60-90 days. Budget: Rs. 15,000-50,000/month in tool costs depending on the stack. Expected efficiency gain: 3-5x creative output with equivalent team. Expected performance impact: 15-30% improvement in ROAS through faster creative iteration, over 6 months.