The campaign research ledger

The following cases are documented because AI changes the idea, production, distribution, or audience experience. Results are included only when the source itself reports them, and promotional claims are labeled.

Campaign index

2024–2026 field update: current projects worth studying

The older cases below remain useful because they show the arc from “AI as a provocative idea” to AI as a production, personalization, or media system. The following projects add more recent evidence and make the database less dependent on launch-era demonstrations.

Dove: AI used to change the distribution environment

Brand / partners: Dove, Mindshare, WPP, and Pinterest. Objective: respond to the way generative systems can narrow beauty representation by giving women more influence over what appears in a recommendation environment. Audience: Pinterest users interested in beauty and identity.

The “Real Beauty DNA” experience asked participants to select traits they associated with beauty and used the aggregate signal to inform a more diverse Pinterest feed. The important creative decision was not merely generating new images; it was changing the recommendation layer around the audience. The campaign paired a participatory interface with a media-platform intervention and shareable personalized video.

WPP reports 787 million impressions, 27 million people reached on Pinterest, engagement among women 21.4% above Pinterest benchmarks, and a 2.9-point increase in association with Real Beauty. Those figures are reported by the agency and brand; they are useful as a case record, not proof that the AI mechanism alone caused the lift. Cannes Lions awarded the work the 2025 Media Grand Prix, which independently validates its creative significance but not every performance number. Sources: WPP case study, Cannes Lions 2025 winners.

Lesson: When the problem is algorithmic representation, the most interesting intervention may be at the recommendation or experience layer—not another synthetic image. Watchout: the campaign should explain what user signals are collected, how they are aggregated, and whether a “more diverse” feed is a measured outcome or a brand interpretation.

Honda: AI as a branching narrative system, not a one-click commercial

Brand / partners: Honda, Amazon Ads Brand Innovation Lab, and 1stAveMachine. Objective: introduce the all-electric Prologue through a participatory story. Audience: Amazon and Fire TV audiences, especially families and people considering an electric vehicle.

Honda’s Dream Generator let people choose characters, settings, and props, then produced a roughly two-minute animated adventure. Amazon’s production record is unusually specific: more than 190 story variations, 300+ written prompts, 20+ custom AI models, 457 hand-drawn figures, 500+ animation hours, and 3,837 AI renderings. The team included illustrators, traditional animators, 3D artists, film directors, and AI developers. That mix is a useful correction to the “small prompt team replaces production” story.

Amazon reports significant lifts in unaided brand awareness and purchase consideration from a U.S. Latitude Brand Study of 1,204 respondents, plus stronger associations with family and environmental friendliness. The case page does not provide a public control design or audited sales increment. Source: Amazon Ads case study. Lesson: branching creative works when the branches express a simple user choice and the production system is grounded in a real product story. The human team still owns narrative architecture, illustration, pacing, and safety.

Touchland and Pit Viper: speed is valuable only when the idea is already legible

Touchland’s “Move Your Mood” campaign used eight scents and eight destinations to make a sweepstakes feel like an emotional travel experience. Google says Gemini created audience personas and insights, Imagen visualized them, and the team produced more than 70 assets. The reported output was 46,500 entries, a 34% web conversion rate, and more than 11 million impressions; Google also reports a 90% reduction in visual-asset timeline, 66% reduction in overall timeline, and 70% lower production cost. Source: Think with Google / Touchland.

Pit Viper’s “Redefine Performance” used Gemini for ideation and script riffing, then Veo 2 for scenes in a deliberately absurd sports narrative. Google and Dept report a production time under one week, a 400% decrease in timeline, 3.8% lift in ad recall, 6.5% lift among 18–24-year-olds, and 10.3 million users reached. Source: Think with Google / Dept. Both are vendor/agency case claims. The transferable lesson is narrower: AI compresses asset exploration most effectively when the creative proposition, tone, and approval criteria are already clear.

Coca-Cola: the interactive layer raises the disclosure and moderation burden

Coca-Cola’s 2024 holiday experience combined an AI Santa conversation, a personalized snow globe, QR entry points, and a remake of the “Holidays Are Coming” world. Coca-Cola names Accenture, Microsoft, Leonardo.ai, Mimic, and OpenAI among the partners and says the experience supported more than 45 languages. Microsoft’s customer story describes launch in 43 markets and more than one million conversations in 26 languages. The discrepancy is a reminder to preserve each source’s wording rather than normalize figures into one “official” number.

The creative strength was the shift from a broadcast film to a user-generated artifact and conversation. The risk was equally clear: a real-time character can invent answers, mishandle sensitive inputs, or make a fictional voice feel like an endorsement. Trade coverage documented backlash to the synthetic remake, and a 2026 academic analysis of 7,822 online comments identifies concerns about realism, authenticity, corporate motives, labor displacement, and human connection. Sources: Coca-Cola announcement, Microsoft customer story, ScienceDirect analysis. Lesson: conversational advertising needs a claims boundary, escalation path, transcript retention policy, and a clear AI disclosure at the first interaction.

Nike: generative exploration can reveal the bias of the tool

Nike’s A.I.R. project brought 13 athletes and Nike designers together, explored hundreds of concepts, and revealed 13 AI-explored/3D-printed prototypes in Paris. Nike describes generative tools as one input among athlete listening, sketching, computational design, simulation, and physical prototyping. The design team also noticed a tendency toward fluid, organic sameness in AI-generated forms and deliberately pushed against it. Source: Nike A.I.R..

This is not an ad-performance case, and it should not be presented as one. It is a strong creative-process case because it shows a team using AI to widen the search space while treating the model’s visual bias as something to interrogate. Lesson: if every exploratory output has the same model signature, the remedy is more human reference, material testing, and editorial rejection—not more prompting.

C01 — Heinz A.I. Ketchup

• Brand / agency: Kraft Heinz; Rethink Toronto.
• Concept: Ask image generators what “ketchup” looks like, then use the outputs to reinforce the brand truth that Heinz is culturally synonymous with ketchup.
• Audience / objective: Younger audiences interested in AI and art; keep a legacy brand in the cultural conversation without abandoning its distinctive equity.
• AI technology: DALL·E 2; audience-supplied prompts were also used to extend the experiment.
• Human contribution: Strategy, prompt design, selection, copy, social invitation, packaging production, OOH placement, and experiential curation.
• Production / distribution: Prompt → image set → social conversation → digital OOH and special-edition bottles → metaverse gallery.
• Reported outcome: The One Club case record reports more than 850 million earned impressions, a 38% higher social engagement rate than past campaigns, and a global spread beyond the original Canada/US launch. These figures are reported in the award record, not independently audited.
• Strength: The technology proves a pre-existing brand memory instead of becoming the whole story.
• Criticism / limitation: Model outputs can reflect training-data bias toward an iconic brand; “even AI knows” is persuasive rhetoric, not an unbiased test.
• Lesson: Use AI as a cultural mirror when the brand has a clear truth to reveal.
• Source: The One Club case record.
• Rights: Campaign visuals are owned by the campaign participants/owners; do not reuse without permission.

C02 — Coca-Cola Create Real Magic

• Brand / partners: The Coca-Cola Company; OpenAI and Bain & Company alliance.
• Concept: Give digital artists and fans a branded sandbox where Coca-Cola archive elements become ingredients for new artwork.
• Audience / objective: Digital creators and fans in selected markets; turn the “Real Magic” platform into participation and co-creation.
• AI technology: A custom platform combining GPT-4 and DALL·E with controlled Coca-Cola assets including the bottle, script logo, Santa Claus, and polar bear.
• Human contribution: Brand/archive access, contest design, creator selection, curation, artist credit, workshop, and decisions about public display.
• Production / distribution: Controlled prompt playground → downloaded submissions → selection for digital billboards and Creative Academy.
• Reported outcome: Coca-Cola’s newsroom says four AI artists kickstarted the campaign and 30 creators were selected for a three-day Real Magic Creative Academy. No sales lift is claimed here.
• Strength: It treats generative AI as a participatory medium with a clear boundary, asset library, and editorial handoff.
• Criticism / limitation: A branded sandbox still raises questions about who owns the resulting work, how archive material is licensed, and whether participation is meaningful or extractive.
• Lesson: If a brand opens its archive, it needs a real invitation, a crediting system, and a clear path from experiment to public use.
• Source: Coca-Cola Company newsroom.
• Rights: Coca-Cola brand assets and campaign imagery are not cleared for reuse by this report.

C03 — Coca-Cola Masterpiece

• Brand / agency: The Coca-Cola Company; creative and production partners are described in company materials.
• Concept: A Coca-Cola bottle moves through a chain of classical and contemporary artworks to restore a student’s inspiration.
• Audience / objective: Global audiences; connect cultural heritage, contemporary artists, product, and inspiration.
• AI technology: Company materials describe generative AI in the creative process and later described work to match historic brush strokes. The public record does not expose a complete tool-by-tool production breakdown, so this case is labeled AI-assisted, exact toolchain partial.
• Human contribution: Artwork selection, artist and estate clearances, narrative, casting/production, edit, and curation of emerging creators from several regions.
• Production / distribution: Art and brand clearances → live/CG/AI-assisted production → hero film → related digital collectibles and social.
• Reported outcome: Coca-Cola reported $543,660 from the related 72-hour “Masterpiece” digital collectibles drop. This is a collection result, not a film effectiveness metric.
• Strength: The work makes the human act of curation visible; AI is part of the bridge between art worlds, not the only idea.
• Criticism / limitation: The more recognizable the source artworks, the more important it is to distinguish licensed, public-domain, artist-approved, and model-generated material.
• Lesson: Visual continuity across references can be accelerated by AI, but the cultural argument still comes from people.
• Sources: Coca-Cola collection announcement and Coca-Cola investor transcript.
• Rights: Artwork, film stills, and artist work require owner permission.

C04 — Lexus Driven by Intuition

• Brand / agency / partners: Lexus; The&Partnership London; Visual Voice; IBM Watson tools and applications; directed by Kevin Macdonald.
• Concept: A car advertised through a story generated by a bespoke AI scriptwriter, with the production itself demonstrating intuition and technology.
• Audience / objective: European launch audiences for the Lexus ES; position the vehicle and brand as bold, innovative, and imaginative.
• AI technology: Lexus says the bespoke AI used IBM Watson and multiple data sources, including award-winning luxury advertising and information on emotional response.
• Human contribution: System design, brand framing, selection, direction, performance, cinematography, edit, and final production.
• Production / distribution: Data and script system → human review → live-action film → digital, social, cinema.
• Reported outcome: No business result is included because the official source confirms the production details, not a measured commercial lift.
• Strength: The AI role is legible and relevant to the product story.
• Criticism / limitation: “Entirely scripted by AI” can overstate the role of a system when humans decide the data sources, prompt/context, selection, staging, and final cut.
• Lesson: An “AI wrote the ad” claim is most useful when the rest of the production openly shows what humans did.
• Source: Lexus Media Site.
• Rights: Film and stills remain with the campaign owners.

C05 — Cadbury Shah Rukh Khan–My Ad

• Brand / agency: Cadbury Celebrations; Ogilvy Mumbai; Wavemaker Mumbai; Rephrase.ai and DeltaX are named in the case record.
• Concept: Let a famous spokesperson promote thousands of local retailers through dynamic, personalized ads during Diwali.
• Audience / objective: Indian festive shoppers and small businesses; help local shops compete for attention while keeping Cadbury relevant during the gifting period.
• AI technology: Machine learning, dynamic creative optimization, and Rephrase.ai to recreate Shah Rukh Khan’s face and voice with local store names.
• Human contribution: Retailer strategy, celebrity relationship and consent, geographic targeting, copy, brand/legal controls, and media operations.
• Production / distribution: Store and location data → rule-based selection → synthetic face/voice variation → digital delivery near the relevant store.
• Reported deployment: Wavemaker describes 500+ pincodes and 2,500+ local business owners. These are source-reported deployment figures; this report does not infer a causal sales lift.
• Strength: Personalization widens participation by giving small retailers access to celebrity and media equity.
• Criticism / limitation: Synthetic celebrity voice and face require explicit contractual scope, disclosure decisions, and a takedown process. A personalized deepfake can be benign, but it still trains audiences to accept a face saying something it did not record in that exact form.
• Lesson: Dynamic creative is most defensible when the rule is simple, the benefit is real, the person’s rights are clear, and the audience is not tricked about the interaction.
• Source: Wavemaker Global case study.
• Rights: Celebrity likeness, voice, campaign imagery, and brand assets are not cleared here.

C06 — Hellmann’s A.I. Mayonnaise

• Brand / artist: Hellmann’s; Stephen Paul Wright, self-initiated.
• Concept: Respond to the rush toward synthetic advertising by contrasting “AI mayonnaise” with a promise of real mayonnaise.
• Audience / objective: Social audiences following the AI advertising conversation; make a fast, pointed brand response.
• AI technology: The source describes the campaign as a response that uses the visual language of AI-generated advertising; it does not present a detailed model/tool breakdown.
• Human contribution: One artist conceived, produced, and launched the response within 48 hours; the artist reports that Hellmann’s leadership later ran it on official channels.
• Production / distribution: Self-initiated rapid response → social publication → brand adoption.
• Reported outcome: No performance metric is claimed. The source confirms the production and official-channel use as reported by the artist.
• Strength: It questions AI without treating the technology as neutral or inevitable.
• Criticism / limitation: Artist-led responses depend on brand permission and can be misread as official before approval.
• Lesson: A critique becomes advertising when it has a brand truth underneath it and the brand is willing to keep the edge.
• Source: Stephen Paul Wright case page.
• Rights: Artist and brand permission required.

C07 — Pedigree Adoptable

• Brand / agency / production: Mars / Pedigree; Colenso BBDO Auckland; Nexus Studios London.
• Concept: Turn every Pedigree ad into an opportunity to feature a real shelter dog available for adoption.
• Audience / objective: Potential pet adopters near participating shelters; use a commercial brand’s media reach to increase visibility for dogs that may have only low-quality shelter photographs.
• AI technology: The One Club describes a bespoke Stable Diffusion-based machine-learning model and an adaptable 3D model/rig built from an individual dog’s photo and grooming information, allowing poses, expressions, and format variation.
• Human contribution: Purpose strategy, shelter relationships, dog portfolio management, model design, retouching, geographic targeting, approvals, and campaign operations.
• Production / distribution: Shelter photo and profile data → dog-specific model/digital double → ad-format variation → geographically targeted digital OOH and digital media → remove or update dogs as status changes.
• Reported outcome: Cannes Lions and Omnicom materials report 50% of featured dogs adopted within the first two weeks and a sixfold increase in shelter site traffic/profile views. These are case-study claims; the causal contribution of AI alone is not isolated.
• Strength: AI is used to solve a specific visibility and production bottleneck while keeping the underlying subject real.
• Criticism / limitation: A polished generated dog must not misrepresent a dog’s physical traits, temperament, or availability. The source/dog record needs a reliable update path.
• Lesson: The strongest purpose-led AI work connects model capability to an operational problem and keeps a real person or animal at the center.
• Sources: The One Club case record, Cannes Lions announcement, and Omnicom case report.
• Rights: Campaign visuals and dog images require permission from the campaign owners and relevant shelters.

C08 — The Origin of Toys “R” Us

• Brand / agency: Toys “R” Us Studios; Native Foreign; OpenAI Sora.
• Concept: A brand origin story imagining young Charles Lazarus dreaming of the toy-store world and Geoffrey the Giraffe.
• Audience / objective: Advertising and brand audiences at Cannes Lions; signal a new chapter for Toys “R” Us through a first-of-its-kind AI brand film.
• AI technology: The brand announcement says the film was created using Sora, then a limited-access text-to-video system.
• Human contribution: Brand story, creative direction, production design, prompt and shot decisions, edit, sound, and brand approvals.
• Production / distribution: Synthetic shot generation and iteration → editorial assembly → Cannes premiere and online film.
• Response / controversy: Trade and creative-community coverage questioned whether the film’s AI novelty overshadowed the brand story and whether the roughness of generated motion weakened the emotional result. This is an editorial response record, not a measured audience study.
• Strength: It used generative video for a story that would otherwise need extensive art direction and world-building.
• Limitation: “First” claims are often marketing language. A film can be technically novel and still feel generic, uncanny, or too much like a product demo for the tool.
• Lesson: For brand film, the question is not whether the frames are impossible; it is whether the audience remembers the brand, the feeling, and the reason to care.
• Sources: Toys “R” Us announcement and ADWEEK interview with Native Foreign.
• Rights: Brand film rights remain with the owners; Sora availability and terms may have changed.

C09 — Etsy Gift Mode

• Brand / agency: Etsy; campaign materials credit Orchard Creative for the brand campaign.
• Concept: Launch an AI gift-shopping experience combining algorithmic suggestions with human curation, then use a high-reach brand campaign to make gifting the platform’s distinctive territory.
• Audience / objective: People who struggle to find personal gifts; shift Etsy from a seasonal shopping destination toward an all-occasion gifting service while keeping sellers and unique products central.
• AI technology: Etsy describes Gift Mode as an interactive hub using AI and human curation to help shoppers find gifts.
• Human contribution: Marketplace and seller curation, product relevance, brand strategy, product experience, creative campaign, and editorial framing.
• Production / distribution: User inputs and product context → AI-assisted recommendations plus human curation → interactive product discovery → Super Bowl and digital distribution.
• Reported outcome: No performance lift is claimed here; the source confirms the launch, campaign purpose, and seven million sellers referenced in the announcement.
• Strength: The AI is attached to a real consumer job—gift discovery—not merely used to manufacture a spectacle.
• Limitation: Personalized recommendations can expose sensitive inferences or create a narrow view of the shopper. Explain what data is used and provide controls.
• Lesson: In commerce, a useful assistant can be an advertising idea when the experience itself embodies the proposition.
• Source: Etsy newsroom.
• Rights: Product images and seller content remain subject to Etsy and seller terms; not bundled here.

C10 — Virgin Voyages Jen AI

• Brand / agency: Virgin Voyages; VML.
• Concept: Let users create personalized invitations that appear to come from an AI version of Jennifer Lopez for a cruise occasion.
• Audience / objective: Potential travelers and their friends; use celebrity appeal and user participation to make a cruise invitation more personal, funny, and shareable.
• AI technology: Generative voice and video personalization, as described by VML’s case study.
• Human contribution: Campaign strategy, celebrity relationship, creative concept, product experience, prompt boundaries, moderation, and media.
• Production / distribution: User milestone and invite inputs → controlled generation → personalized video → sharing from the microsite/social ecosystem.
• Reported outcome: VML reports more than 25,000 personalized invites, close to 200,000 microsite interactions, and nearly 20 million organic impressions. These are agency-reported case figures, not independent causal evidence.
• Strength: The generated output is functional—it is an invitation the user can send—not just a visual flourish.
• Limitation: Celebrity-style personalization can blur the line between a real endorsement and a synthetic performance. Disclosure should be close to the content, not hidden in terms.
• Lesson: Participatory personalization needs a reason for the audience to make and share the asset; “AI” is not that reason by itself.
• Source: VML case study.
• Rights: Celebrity likeness and voice require campaign-specific permission; imagery is not cleared here.

C11 — Coca-Cola Coke SoundZ

• Brand / partners: The Coca-Cola Company; WPP Open X.
• Concept: Capture the recognizable sounds of the Coca-Cola experience and use AI to compose a range of tunes for creators and consumers.
• Audience / objective: Music and social audiences; make product sound a participatory brand asset.
• AI technology: WPP says Coca-Cola captured signature sounds with binaural microphones and trained an AI to compose music from them.
• Human contribution: Sound capture, brand strategy, composition supervision, producer and creator collaboration, selection, and distribution.
• Production / distribution: Field recording → model-assisted composition → producer/DJ collaboration → influencer and social activation.
• Reported outcome: The case study says influencers and music producers including Marshmello used the work, but does not provide an independently verified business result.
• Strength: It transforms a sensory brand cue into an expandable audio system rather than a generic AI-generated song.
• Limitation: Training or generating from proprietary sound does not answer every question about authorship, musician credit, or downstream platform rights.
• Lesson: AI music works best when the source material is unmistakably owned by the brand and the human musical contribution stays visible.
• Source: WPP case study.
• Rights: Brand sound recordings and creator outputs require permission and credit review.

C12 — SHI International AI audio ads

• Brand / platform: SHI International; Spotify Ads Manager case study.
• Concept: Create multiple versions of an existing audio ad using AI voice, tone, and background music, then compare the variants in paid media.
• Audience / objective: B2B technology buyers; test whether lower-friction variation can improve traffic and efficiency.
• AI technology: Spotify’s AI-powered ad creation tool for voice, tone, and music variation.
• Human contribution: Original ad concept, product message, approval, test design, media budget, and interpretation.
• Reported outcome: Spotify reports the AI-generated ad drove three times more site traffic than the original actor-read version and was 24% more cost-efficient than the bid cap. This is a platform case study and should be treated as a vendor-reported A/B result, not a general benchmark.
• Strength: The experiment compares creative variants and ties the tool to a measurable media test.
• Limitation: Without the full experiment design, spend, audience split, duration, and confidence intervals, the result cannot establish that AI caused the improvement.
• Lesson: AI earns a place in performance creative when it creates a testable hypothesis, not just more assets.
• Source: Spotify Ads case study.
• Rights: Voice and music rights depend on the tool plan, source recordings, and campaign contract.