Brands Using AI for Advertising 2026

TL;DR — Key takeaways

Nike’s “Never Done Evolving” AI campaign earned Cannes Lions recognition and 100M+ views — but was built by a premium creative agency with significant production budget, not a marketing team with a free tool subscription.
Coca-Cola’s 2025 holiday AI ad split the industry: technically competent, commercially successful, and widely criticized as “digital slop” by creative professionals. Both reactions were correct.
The four real uses of AI in advertising — creative generation, hyper-personalization, media buying optimization, and measurement — have very different ROI profiles and brand risk levels.
AI-generated advertising fails most visibly at brand character: technical quality is achievable; capturing the cultural specificity that makes a brand feel like itself is where most AI campaigns fall flat.
For brands without enterprise budgets, AI delivers the clearest ROI in media buying optimization and dynamic creative testing — not in campaign creative generation.

The most-viewed piece of AI advertising in 2025 wasn’t a viral stunt. It was Nike’s “Never Done Evolving” — a Cannes Lions-recognized campaign by AKQA that used machine learning to reconstruct two versions of Serena Williams, aged 17 and 35, and pit them against each other in a virtual tennis match. It earned over 100 million views. It was technically brilliant, culturally resonant, and had a production budget that would make most brand managers sweat.

That context is important. The breathless coverage of “brands using AI for advertising” usually describes what enterprise budgets and top-tier agency talent can do, then implies it’s a template for everyone. It mostly isn’t. The question worth asking is not “are brands using AI for advertising?” — they clearly are — but “which parts of advertising does AI actually improve at which budget level, and where does it create more risk than reward?”

Table of Contents

Toggle

The Four Uses of AI in Advertising (and Their Very Different Risk Profiles)

Creative generation is the use case getting the most attention and creating the most controversy. Coca-Cola’s generative AI holiday ad in 2025 was technically competent — consistent lighting, coherent characters, recognizable brand iconography. It was also widely described as “digital slop” by creative industry professionals, and that criticism has merit. The ad looked like a high-quality simulation of a Coca-Cola ad. It did not feel like one. That gap — between technical adequacy and cultural specificity — is where AI creative currently fails most systematically.

The Coca-Cola Project Fizzion work is more interesting. Rather than replacing creative output, it built a machine-readable “StyleID” that captures brand rules and applies them consistently across formats, markets, and production partners. This is AI as creative governance, not creative replacement — and it solves a genuinely expensive problem (brand inconsistency at global scale) without asking AI to replace the judgment that makes creative work.

Hyper-personalization is where AI advertising delivers its clearest commercial results. Cadbury’s personalized video campaign using Shah Rukh Khan’s AI likeness generated individualized brand ads for millions of Indian small business owners, each featuring their shop name and location. Clinique used AI-personalized email subject lines and saw a 25% open rate increase. These results are reproducible at far smaller scales — personalized dynamic creative in paid media consistently outperforms static creative by 15–35% on click-through rate in documented tests.

Media buying optimization is probably the highest-ROI, lowest-visibility use of AI in advertising. Programmatic AI bidding, audience lookalike modeling, and real-time budget reallocation across channels are generating measurable efficiency gains — typically 20–40% improvement in cost-per-acquisition when applied to mature campaign data. This is unglamorous infrastructure work that doesn’t get profiled in trade press, but it’s where most of the actual AI advertising ROI sits for brands without enterprise creative budgets.

Measurement and attribution is the least developed but potentially most valuable category. Multi-touch attribution models trained on first-party data can identify which touchpoints actually drive conversion versus which ones just appear in the conversion path. Brands that have built these models consistently find that 20–30% of their media spend is on channels with attribution credit but low actual causality — money that can be reallocated once the model surfaces it.

100M+

views on Nike’s “Never Done Evolving” AI campaign — Cannes Lions recognition, built with top-tier agency talent and significant production budget

Source: Times Square Chronicles, 2025

Brand Case Studies: What Actually Happened

Epinium data

Based on campaigns we’ve managed across 12+ European Amazon marketplaces, brands that implement AI bid optimization see ACoS improvements of 18–35% in the first 60 days.

Coca-Cola has run more documented AI advertising experiments than almost any other global brand. The Serena Williams recreation (technically Nike’s, Coca-Cola ran a different Serena campaign) — actually, Coca-Cola’s significant AI advertising includes the “Create Real Magic” platform that invited artists to generate brand content using GPT-4 and DALL-E, with winning submissions displayed on Times Square and Piccadilly Circus. The platform generated over 120,000 user-created pieces. The marketing lesson here isn’t about AI replacing creative — it’s about AI-enabled co-creation as community engagement. The brand provided the assets; the audience provided the creative direction. That dynamic is harder to replicate than it looks.

The 2025 holiday ad controversy revealed a different dynamic. Coca-Cola remade its famous 1995 “Holidays Are Coming” TV ad using generative AI. The technical quality was high. The cultural reaction was divided: older audiences who remembered the original found it uncanny; advertising professionals found it professionally troubling; the general public didn’t particularly notice or care. The campaign ran globally and performed to commercial targets. Whether it was a success depends entirely on which metrics you measure.

Nike used AI most successfully when combining it with meaningful human judgment at the creative concept level. “Never Done Evolving” used AI to reconstruct Serena Williams from archival footage — technically extraordinary — but the concept (what would it mean for the greatest athlete to compete against herself across time?) was deeply human. The AI was the instrument; the cultural insight was the actual creative idea. This pattern — human concept, AI execution — is where AI advertising consistently outperforms human concept, AI concept, AI execution.

Heinz ran a simpler but instructive campaign. They asked multiple AI image generators to produce “ketchup” and found that across every model, the output looked like a Heinz bottle — unprompted. The campaign made a factual point about brand recognition while simultaneously demonstrating AI capability. Creative cost: minimal. Media value: substantial. This is the category of AI advertising that any brand can replicate — using AI’s behavior as the creative content rather than using AI to generate the creative content.

AI Advertising by Budget Level: What’s Actually Accessible

Budget level	Best AI advertising use	Realistic ROI	Avoid
SMB (<€50K/year media)	Dynamic creative testing, AI copywriting variants	20–35% CPA improvement	Full AI campaign creative
Mid-market (€50K–500K)	Programmatic optimization, personalized email creative	25–40% efficiency gain	AI-generated brand campaign video
Enterprise (€500K+)	Multi-touch attribution modeling, brand StyleID systems	20–30% media reallocation opportunity	Replacing human creative strategy

FREE SESSION

Find where AI advertising creates real ROI for your brand

We help brands identify which of the four AI advertising use cases matches their budget, catalog, and risk profile — and which ones will generate headlines without generating results.

Book free session → ✓ Free ✓ 30 min ✓ No pitch

Where AI Advertising Fails (The Honest Account)

Brand character is the consistent failure point. AI can generate technically correct advertising. It struggles to generate advertising that feels specifically, unmistakably like a particular brand — the texture of wit in Oatly’s copy, the particular quality of human imperfection in Dove’s visuals, the tonal specificity of Patagonia’s environmental communications. These aren’t arbitrary aesthetic preferences; they’re the result of decades of consistent brand voice accumulated through human creative judgment. AI trained on a brand’s existing output can produce competent imitations, but the imitations tend to regress toward a generic mean.

Brand safety is the second failure mode. Programmatic AI ad placement has put major brand advertising adjacent to hate speech, misinformation, and extremist content repeatedly — not because the AI was malicious but because its optimization target (impressions, clicks) didn’t penalize contextual appropriateness sufficiently. Manually managed brand safety lists are increasingly inadequate at the speed AI-driven placement operates. This isn’t a solved problem in 2026.

The “AI washing” failure mode affects brands, not their campaigns. Announcing “AI-powered advertising” as a differentiator has become meaningless given how saturated the space is. Campaigns built to demonstrate AI capability rather than sell a product or build brand equity consistently underperform on commercial metrics. The Heinz example works because the AI observation was incidental to a genuine brand claim. Campaigns where the AI is the point usually lack a point.

FAQ: Brands Using AI for Advertising

Brands and AI advertising in 2025-2026: what actually changed

Midjourney IP suits reshape creative vendor choices (2025-2026)

Disney, NBCUniversal, and Warner Bros. Discovery sued Midjourney through 2025. Brand legal teams pulled Midjourney from approved-vendor lists, accelerating Firefly and licensed-data tool adoption in paid campaigns.

Sora shuts down, Disney pulls $1B (2026)

OpenAI shut down Sora’s consumer product and Disney withdrew its planned investment in 2026. The ‘AI video is production-ready’ narrative took a public hit — brand campaigns that bet exclusively on generative video got caught.

Anthropic + Meta + Google ship agentic ad tools (2025-2026)

Meta rolled out AI-driven creative generation (Advantage+), Google expanded Performance Max and Demand Gen automation, and Anthropic’s Managed Agents opened brand-custom orchestration. The tooling floor rose — baseline advertising AI is now table stakes.

Can a mid-size brand realistically replicate what Nike or Coca-Cola did with AI advertising?

Not the flagship campaigns, no. Nike’s “Never Done Evolving” and Coca-Cola’s AI platform campaigns required enterprise budgets, premium agency partnerships, and IP arrangements (using real athletes’ likenesses) that aren’t accessible to most brands. What mid-size brands can replicate: the underlying approach of human concept + AI execution, AI-powered dynamic creative testing in paid media, and programmatic optimization. The ROI from those unglamorous applications is often higher than the headline campaigns generate.

Is AI-generated advertising creative quality good enough for real campaigns?

For static assets (social media posts, display ads, email creative), AI quality is commercially viable with proper prompt engineering and human review. For video and brand campaign work, quality is technically adequate but culturally thin — the outputs don’t carry the brand character that comes from human creative intuition. Most brands that have run fully AI-generated campaign video found it performed adequately on reach metrics and poorly on brand attribute tracking (warmth, distinctiveness, memorability).

What’s the brand safety risk with AI advertising?

Two distinct risks: contextual placement (programmatic AI placing your ads near inappropriate content — ongoing and not fully solved) and content generation (AI-generated creative that inadvertently includes inaccurate claims, offensive associations, or copyright-infringing elements). Both require human review layers. The content generation risk increases with the complexity of the product category and the specificity of the brand claims the AI is asked to make.

How do AI personalization campaigns actually work at scale?

The Cadbury Shah Rukh Khan campaign is the most cited example: a model was trained on the celebrity’s speech and movement patterns, then combined with dynamic location and business-name data to generate millions of individualized 60-second videos. At smaller scale, dynamic creative optimization (DCO) personalizes ad elements — headline, image, offer — based on audience segment, location, weather, time of day, and behavioral signals. DCO doesn’t require celebrity AI training; it requires a good first-party data foundation and a creative template system that can be parameterized.

What metric should brands use to evaluate AI advertising ROI?

For creative generation: production cost per approved asset (AI should reduce this by 60–80% for static assets) vs. human agency. For personalization: CPA delta between personalized and non-personalized variants in A/B test. For media buying optimization: efficiency gain (impressions-per-euro or CPA) before and after AI optimization layer implementation. For measurement/attribution: what percentage of media budget shifts when AI attribution model replaces last-click. Each use case needs its own measurement frame; a blended “AI advertising ROI” number obscures whether any specific application is actually working.

The most useful frame for understanding where AI fits in advertising isn’t “what can AI create?” but “what does advertising require that AI is structurally good or bad at?” AI is good at variation, consistency at scale, and optimization against measurable signals. It’s bad at cultural specificity, novelty that requires genuine world knowledge, and the kind of creative risk that builds brand equity. The brands winning with AI advertising aren’t the ones deploying it everywhere — they’re the ones who mapped those capabilities honestly against their actual creative and operational needs.

TRANSFORM BY EPINIUM

Which brand-AI advertising cases actually worked?

Heinz’s AI-prompted ‘ketchup is AI ketchup’ campaign (Rethink, 2022) and Coca-Cola’s AI-enhanced holiday spots earned real brand lift because they made the AI usage itself part of the story. Campaigns that silently substituted AI for production spend rarely matched the performance of equivalent human-produced work.

When does ‘brand uses AI’ backfire as messaging?

When your audience is skeptical of synthetic media (creative industries, entertainment, regulated verticals) or when the AI shortcut feels like cost-cutting rather than craft. Heinz worked because the product itself is iconic; the same approach flops when the brand has no pre-existing visual equity.

How do I measure AI’s actual contribution to ad performance?

Run holdouts. Split creative production: 50% AI-led, 50% agency-led, identical briefs, identical spend, matched audiences. Compare CPM, CTR, and brand-lift panel results over 6-8 weeks. Brands skipping this step almost universally over-credit AI’s contribution.

Map AI to the right part of your advertising — and stop where it creates more risk than ROI

Epinium helps brand teams identify where AI advertising delivers measurable returns versus where it generates headlines at the cost of brand equity.

Book free session →

Free · 30 min · No commitment