AI in Advertising Examples: 16 Real Use Cases Decoded, Plus the Ones That Backfired

What counts as AI in advertising?

AI in advertising is any use of machine learning or generative models to create, target, deliver or measure ads. That is a wide net, and most lists mix up very different things. A fan-made AI video and an automated bidding algorithm are both "AI in advertising", but they solve different problems and need different budgets.

We sort every example into one of five jobs:

1. The idea. AI is the concept of the campaign itself. The story is "look what AI did".
2. Personalization at scale. AI makes thousands or millions of versions, each tailored to a person, a store or a place.
3. Data storytelling. AI turns a large dataset into a story no human editor could build by hand.
4. Production. AI replaces or speeds up the shoot: generated footage, voices, sets and edits.
5. Media and optimization. AI inside the ad platforms decides targeting, bids, placements and which creative runs.

The first four are visible to the audience. The fifth is invisible, and it is where most advertisers already use AI every day without calling it that.

Why the job matters more than the tool

When you pick an example to copy, copy the job, not the tool. Tools change every few months. The job tells you what you need: a strong existing brand (the idea), clean customer data (personalization), a dataset worth mining (storytelling), a creative team that can direct AI (production), or enough conversion volume for the algorithm to learn (media).

Most failed AI ads we see were the wrong job for the brand. A new brand tried an "AI is the idea" stunt with no brand recognition to play with, or a brand with 20 sales a month expected an AI bidding system to find buyers on its own.

AI as the creative idea: 4 examples

These campaigns got attention because AI was part of the story. They earned headlines and press coverage more than direct sales, and they work best for brands people already know.

1. Heinz "A.I. Ketchup" (2022)

Heinz asked OpenAI's DALL-E image model to draw "ketchup" in many different styles and scenes. Again and again, the model produced bottles that looked like Heinz: the shape, the red, the label. The brand turned that result into the campaign line that even AI thinks ketchup looks like Heinz, then invited people to send in their own prompts.

The campaign was widely reported to have generated more than a billion earned impressions and won major creative awards.

What it teaches: this only worked because Heinz had spent decades owning the category's look. The AI did not make the brand famous. It proved the brand was already famous. If your brand is not the default picture of your category, this format will not work for you.

2. Coca-Cola "Create Real Magic" (2023)

Coca-Cola built a platform with OpenAI's GPT-4 and DALL-E and gave fans a set of archive brand assets, such as the contour bottle, the script logo and classic holiday characters, to remix into their own artwork. Selected pieces appeared on digital billboards in Times Square in New York and Piccadilly Circus in London.

What it teaches: Coca-Cola did not hand people a blank prompt box. It gave them a curated kit of brand elements, so almost everything people made still looked like Coca-Cola. That is the lesson for any brand running user-generated AI: control the ingredients, not the outcome.

3. Dove "The Code" (2024)

For the 20th anniversary of its Real Beauty work, Dove showed how generative AI tends to produce narrow, unrealistic images of women when asked for "the most beautiful woman in the world". The brand then committed publicly to never using AI to create or distort women's images in its ads.

What it teaches: a brand can use AI as the subject of a campaign without using it for the creative at all. Dove's position against AI imagery was consistent with 20 years of brand values, which is why it felt credible rather than opportunistic.

4. BMW 8 Series Gran Coupé AI art projections (2021)

BMW trained an AI model on a large collection of classic and modern artworks, then projection-mapped the generated art onto the 8 Series Gran Coupé. The car became a moving canvas for AI-generated visuals.

What it teaches: luxury brands can use AI as craft, not a shortcut. The appeal came from the art-and-engineering story, which fits a premium car brand. For Dubai luxury, fragrance and jewelry brands, this style of AI-as-art sits well with premium positioning, as long as the output looks deliberate and not generated in a hurry.

AI for personalization at scale: 5 examples

This is where AI changes the economics of advertising. Before generative AI, making 2,000 versions of a video was impossible on a normal budget. Now it is a production decision.

5. Cadbury "Not Just a Cadbury Ad" (2021, India)

During Diwali, when small shops in India were struggling after lockdowns, Cadbury used AI to recreate the face and voice of Bollywood star Shah Rukh Khan, with his consent. Local shop owners could generate a free ad in which the actor named their store and neighborhood. Thousands of small retailers took part, and each version was served to people near that store.

What it teaches: the AI was impressive, but the idea was generous. The brand used a celebrity to help small businesses, which earned goodwill that a normal celebrity ad never would. Consent and a clear contract for the actor's likeness were essential, and they are just as essential for any brand that wants to try this with a local talent in the UAE or Saudi Arabia.

6. Nutella Unica (2017, Italy)

Nutella used an algorithm to generate 7 million one-of-a-kind label designs from a library of patterns and colors. Every jar in the run was different, and the jars were promoted as collectible pieces of art.

What it teaches: this predates today's generative AI, so it is algorithmic design rather than an image model. It still belongs on the list because it shows the core idea of personalization at scale: when every item is unique, the product itself becomes the ad. The same approach now works for limited-edition packaging, Ramadan gift boxes and event merchandise.

7. Carvana's 1.3 million personalized videos (2023)

Carvana, the online car retailer, sent more than 1.3 million customers a personal AI-generated video recapping their own car purchase: the car they bought, when they bought it and details from their buying journey, wrapped into a short celebratory film.

What it teaches: personalization works best when you use data the customer already knows you have. Nobody finds it creepy that a car seller remembers the car you bought from them. It feels like a thank-you. If your CRM holds purchase dates, product names and cities, you already have the raw material for this.

8. Virgin Voyages "Jen AI" (2023)

Virgin Voyages created an authorized digital version of Jennifer Lopez. Visitors to a campaign site could make a personalized video invitation, voiced and presented by "Jen AI", to invite friends on a cruise.

What it teaches: the mechanic turned every customer into a distributor. People shared the invitations because they were personal and a little funny. This is the same logic as Cadbury's campaign: give people something made for them, and they do the media buying for you.

9. Netflix personalized artwork

Netflix does not show every viewer the same thumbnail for a title. Its systems choose which artwork to show each member based on what they tend to watch, so a romance fan and an action fan may see different images for the same film.

What it teaches: personalization does not have to mean a new video per person. Swapping the first frame, the thumbnail or the opening line for different audiences is the cheapest form of AI personalization, and it is available to any brand running ads on Meta or TikTok today.

AI for data storytelling: 2 examples

Here the AI's job is analysis. It processes more data than a human team could, and the result becomes the creative.

10. Nike "Never Done Evolving" (2022)

For its 50th anniversary, Nike trained AI models on years of Serena Williams' match footage. It then simulated a match between Serena at her first Grand Slam win in 1999 and Serena in 2017, and streamed the "match" on YouTube. The campaign won the Grand Prix for Digital Craft at Cannes Lions.

What it teaches: the story was the dataset. Nike owned a relationship with one of the most recorded athletes in history and used AI to ask a question fans had argued about for years. You need two things to copy this: a dataset that is genuinely yours, and a question your audience actually cares about.

11. JPMorgan Chase and AI-written ad copy (2019)

JPMorgan Chase tested marketing copy written by an AI language model against copy written by its own team. The bank reported that some AI-written headlines drove click-through rates up to 450% higher than the human versions, and it moved to use AI copy across more of its marketing.

What it teaches: this is the least glamorous example on the list and one of the most useful. The model was trained on which words drove response, then it generated variations at scale. Today any brand can do a basic version with an AI writing tool and platform A/B tests. The "up to" matters, though: the biggest lift was on specific ads, not an average across everything.

AI as the production crew: 3 examples

These are the campaigns people usually mean when they say "AI ads": footage, voices and edits generated instead of filmed.

12. Toys "R" Us origin film made with Sora (2024)

Toys "R" Us premiered what it called the first brand film made with OpenAI's Sora video model at the Cannes Lions festival in 2024. The short film told the story of the founder as a child dreaming up the store and its giraffe mascot, Geoffrey.

The reception was mixed. People praised the speed and ambition, and criticized the uncanny faces and inconsistent details from shot to shot.

What it teaches: a first-of-its-kind film earns press because it is first. The second one does not. Brand films made with AI video models now get judged on craft like any other film, and inconsistent faces are the first thing audiences notice.

13. Kalshi's NBA Finals ad (2025)

Kalshi, a US prediction market, aired a surreal, fast-cut TV spot during the 2025 NBA Finals that was made with Google's Veo 3 video model. It was widely reported that one creator produced it in a few days for roughly USD 2,000, a small fraction of a typical broadcast budget.

What it teaches: AI production is now good enough for broadcast when the style suits it. The ad was deliberately chaotic, so the small glitches of AI video read as part of the joke. Match the creative style to what the model does well today, rather than fighting its weaknesses.

14. What we see in our own production work

We use the same production approach every week at XMA: AI generates footage, product shots and voiceovers, and a human team handles strategy, scripts, direction and the final edit. The biggest change is speed. We take projects from brief to launch in 7 days, and we can produce a full set of hook variations for testing instead of one hero ad. You can see our AI video portfolio for the finished work across beauty, fragrance, fashion and product categories.

The trade-off is honest: AI is weaker at believable human performance, exact product details and hands, so we still plan shots around those limits. If you are comparing models for this kind of work, our guide to keeping AI variants on-brand covers which tools hold a brand look best.

AI inside the ad platforms: 2 examples you already use

This is the least visible and most widely used AI in advertising. If you run ads on Google, Meta or TikTok, you are already using it.

15. Google Performance Max and Meta Advantage+

Google's Performance Max campaigns use Google AI to decide bids, audiences and placements across Search, YouTube, Display, Gmail, Maps and Discover from a single campaign. You supply goals, budgets, creative assets and audience signals, and the system mixes and matches assets to find conversions.

Meta's Advantage+ campaigns do the same across Facebook and Instagram. They automate targeting, placements and creative combinations, and they increasingly generate creative variations such as new backgrounds and text versions from your assets.

What it teaches: these systems are strong at finding buyers, but only when they have enough conversion data and enough creative variety. In our experience the most common reason an Advantage+ or Performance Max campaign stalls is too few assets, not bad targeting. The algorithm can only choose from what you give it. That is why creative volume, the thing generative AI is good at, has become a media-buying lever.

16. TikTok Symphony

TikTok's Symphony suite offers generative tools for advertisers, including script help, AI avatars and translation and dubbing for existing videos, so one ad can be adapted into several languages and styles for testing.

What it teaches: the platforms are now building the production tools themselves. For brands in the Gulf, translation and dubbing are the most practical features, because Arabic and English versions of the same ad are standard. Still, we recommend a native speaker checks every Arabic line before it runs. Automatic translation can be grammatically correct and still sound wrong to a local audience.

For the full creative framework we use on these platforms, see our guide to AI video ads for Meta and TikTok.

AI ads that backfired, and why

Most lists skip this part. The failures teach more than the wins, because they show where audiences draw the line.

Coca-Cola "Holidays Are Coming" AI remake (2024)

Coca-Cola remade its classic holiday truck ad using generative AI. Many viewers called it soulless and pointed to odd details such as inconsistent trucks and uncanny people. The criticism landed hardest because the original was a much-loved piece of nostalgia.

Why it backfired: the brand used AI on the one asset where people wanted human warmth and tradition. Using AI for a fan platform in 2023 was praised; using it to replace a beloved film in 2024 was not.

McDonald's Netherlands AI Christmas ad (2025)

In December 2025, McDonald's in the Netherlands released an AI-generated Christmas ad and pulled it within days after heavy criticism online, including complaints about its look and its message.

Why it backfired: festive ads carry emotional expectations. When the visible "AI look" meets a season built on family and tradition, many viewers read it as a brand cutting corners.

The pattern behind the failures

Put the wins and failures side by side and a pattern appears:

• AI as spectacle works once. First-of-its-kind campaigns get attention for being first. Copies of the stunt do not.
• AI loses on emotion and nostalgia. Audiences push back hardest when AI replaces human warmth in a moment they care about.
• Visible flaws cost more than they save. Uncanny faces and inconsistent details get screenshotted and shared for the wrong reasons.
• Invisible AI rarely gets criticized. Nobody complains about AI-chosen bids, AI-written headlines or AI-picked thumbnails.

This is why we rarely recommend "made entirely with AI" as a message. The audience does not care how an ad was made unless you make it the story, and then they judge it harder.

Which AI advertising examples can your brand copy, and at what cost?

Here is how the 16 examples compare on what they required. Budget tiers are our rough estimates for running a similar idea in 2026, not what the original brands spent.

• Heinz A.I. Ketchup: Job AI did: The idea; What you need: Category-defining brand recognition; Rough budget to copy: High (PR-led); Realistic for a mid-size brand?: Rarely
• Coca-Cola Create Real Magic: Job AI did: The idea; What you need: Brand archive, custom platform; Rough budget to copy: High; Realistic for a mid-size brand?: No
• Dove The Code: Job AI did: The idea; What you need: Long-standing brand values on the topic; Rough budget to copy: Mid to high; Realistic for a mid-size brand?: Only with a genuine brand stance
• BMW AI art: Job AI did: The idea; What you need: Art direction, event production; Rough budget to copy: High; Realistic for a mid-size brand?: Partly (as AI art content)
• Cadbury Shah Rukh Khan: Job AI did: Personalization; What you need: Celebrity consent, local targeting; Rough budget to copy: High; Realistic for a mid-size brand?: Yes, with local talent
• Nutella Unica: Job AI did: Personalization; What you need: Packaging production line; Rough budget to copy: Mid; Realistic for a mid-size brand?: Yes, for limited editions
• Carvana videos: Job AI did: Personalization; What you need: Clean CRM data, video templates; Rough budget to copy: Mid; Realistic for a mid-size brand?: Yes
• Virgin Voyages Jen AI: Job AI did: Personalization; What you need: Celebrity likeness rights; Rough budget to copy: High; Realistic for a mid-size brand?: Yes, with a smaller personality
• Netflix artwork: Job AI did: Personalization; What you need: Several thumbnails or openings; Rough budget to copy: Low; Realistic for a mid-size brand?: Yes
• Nike Serena: Job AI did: Data storytelling; What you need: A unique dataset; Rough budget to copy: High; Realistic for a mid-size brand?: Rarely
• JPMorgan Chase copy: Job AI did: Data storytelling; What you need: Copy testing at volume; Rough budget to copy: Low; Realistic for a mid-size brand?: Yes
• Toys "R" Us Sora film: Job AI did: Production; What you need: AI video skills, creative direction; Rough budget to copy: Mid; Realistic for a mid-size brand?: Yes, if the craft is strong
• Kalshi Finals ad: Job AI did: Production; What you need: A style that suits AI video; Rough budget to copy: Low; Realistic for a mid-size brand?: Yes
• AI production at XMA: Job AI did: Production; What you need: Human creative team plus AI; Rough budget to copy: Low to mid; Realistic for a mid-size brand?: Yes
• Performance Max, Advantage+: Job AI did: Media; What you need: Conversion data, many assets; Rough budget to copy: Ad spend only; Realistic for a mid-size brand?: Yes
• TikTok Symphony: Job AI did: Production and media; What you need: Source videos, native review; Rough budget to copy: Low; Realistic for a mid-size brand?: Yes

The pattern is clear. The "AI is the idea" stunts need a famous brand. Personalization, copy testing, production and platform AI are open to almost anyone.

What drives the cost of these AI campaigns

Brands often ask us what a version of these campaigns would cost. The honest answer depends on the market, the talent and the volume, so we scope and quote each one individually. Here is what each campaign type involves:

• AI copy and thumbnail testing: What you get: 20-40 headline and first-frame variants for Meta or TikTok; Main cost drivers: Number of variants, languages
• AI video ad pack: What you get: 5-10 short video ads with hook variations, Arabic and English; Main cost drivers: Number of ads and hooks, revision rounds
• Personalized video campaign: What you get: Template plus hundreds to thousands of data-driven versions; Main cost drivers: Data setup, template design, number of versions
• AI brand film: What you get: 60-90 second hero film with AI footage and human direction; Main cost drivers: Length, creative direction, revision rounds
• Celebrity or talent likeness campaign: What you get: Licensed digital likeness, personalization tool, legal work; Main cost drivers: Talent rights, legal review

Two things move the numbers most. The first is talent rights: licensing a real person's face or voice costs far more than the AI itself. The second is revision rounds, because every change to a generated shot means regenerating and re-editing it. For a price on your own idea, get a quote and we will scope it with you.

Compared with a traditional shoot, the saving is mostly in variations. One filmed hero ad plus ten AI variations usually costs less than two separately filmed ads, and it gives the ad platforms the creative volume they need.

How to adapt these examples for Dubai and GCC brands

The famous examples are mostly American and European. The ideas travel, but the details change in the Gulf.

Arabic and English from day one

Most UAE campaigns need both languages, and Saudi campaigns often lead with Arabic. AI makes dual-language versions cheap, which is a real advantage here. The catch is quality: Gulf Arabic, Egyptian Arabic and Modern Standard Arabic sound different, and a dubbed voice that does not match the audience feels foreign. We plan the Arabic script first rather than translating the English one, then use AI for voice and lip-sync where it helps.

Timing around the regional calendar

Ramadan, Eid, UAE National Day, Saudi National Day, White Friday and the Dubai Shopping Festival drive a large share of yearly ad demand. These are the moments where Cadbury-style local personalization and Carvana-style customer recaps fit best. They are also the emotional moments where the Coca-Cola and McDonald's lesson applies most: keep the warmth human, and use AI where it adds relevance, not where it replaces feeling.

Consent, likeness and talent rules

If you recreate a person's face or voice, get written consent that covers the exact use, markets, duration and platforms. This matters even more when a campaign uses local influencers or presenters. In the UAE, paid promotion by influencers is also regulated, so check the current UAE Media Council requirements before you build a campaign around a creator's likeness.

Label AI content where the platforms ask

The major platforms now ask for disclosure of realistic synthetic content. TikTok explains how to label AI-generated content on the platform, and YouTube requires creators to disclose altered or synthetic content that could be mistaken for real footage. Labeling realistic AI people, places or events keeps you inside the rules, and it protects trust if someone spots the AI anyway.

How to run your first AI advertising test in 30 days

You do not need a Cannes-level budget to learn from these examples. This is the plan we use with brands running their first structured AI ad test.

1. Pick one job, not five. Choose the job that fits your brand from the table above. For most brands the first test is production (more ad variations) or personalization (versions by audience or city).
2. Set one success metric. For performance ads, pick cost per purchase or cost per lead. For awareness, pick three-second view rate and hold rate. Write down your current number before you start.
3. Build a small creative matrix. Take your best-performing ad and make 6-10 AI variations that change one thing each: the hook, the first frame, the presenter, the language or the setting.
4. Keep a human control. Run your existing best ad alongside the AI variations with the same budget and audience, so you compare fairly.
5. Review every asset before launch. Check product accuracy, hands, faces, on-screen text, Arabic grammar and brand colors. One wrong product detail can cost more trust than the test is worth.
6. Launch in one platform first. Meta Advantage+ or TikTok usually give readable results fastest. Give each variation enough spend to exit the learning phase, and let the test run 7-14 days.
7. Read the results by variable. Look at which hook, face or language won, not just which ad. That tells you what to make more of.
8. Scale the winners and refresh. Put more budget behind the winning variables, then make the next batch. AI's real advantage is that the next round takes days, not weeks.

If you want a starting point for step three, our explainer on what a UGC creator does covers the creator-style formats that often win the first round of testing.

How to tell whether the AI version won

Judge the AI variations on the same numbers as your human-made ads, not on novelty. A useful rule: if an AI variation beats your control on the main metric with similar spend, keep it. If it only wins on cost to produce, it is a production saving, not a performance win. Both are useful, but they are different results, and it helps to report them separately.

Watch comments as well as metrics. If people start pointing out that an ad "looks like AI" in a negative way, that is an early warning, even when click-through rates look fine.

Frequently Asked Questions

What are some real examples of AI in advertising?

Well-known examples include Heinz "A.I. Ketchup", where an image model drew Heinz-like bottles; Coca-Cola "Create Real Magic", a fan art platform built with OpenAI; Cadbury's personalized Shah Rukh Khan ads for local shops in India; Nike's AI-simulated Serena Williams match; Carvana's 1.3 million personalized videos; and the Toys "R" Us brand film made with Sora. Platform tools like Google Performance Max and Meta Advantage+ are everyday examples.

How is AI used in advertising today?

Brands use AI in five main ways: as the creative idea itself, to personalize ads at scale, to turn large datasets into stories, to produce footage, voices and edits faster, and inside ad platforms to decide targeting, bids and which creative runs. The last two are the most common. Most advertisers on Google, Meta and TikTok already use AI-driven bidding and placement every day.

Do AI-generated ads perform better than traditional ads?

Not automatically. AI ads perform well when they are on-brand, reviewed by humans and tested against a control. The main advantage is volume: you can test many hooks, faces and languages quickly, which helps platform algorithms find winners. AI ads tend to underperform when visible flaws such as uncanny faces appear, or when they replace human warmth in emotional moments like holidays.

Which AI ad campaigns backfired?

Coca-Cola's 2024 AI remake of its "Holidays Are Coming" ad was widely criticized as soulless, and McDonald's Netherlands pulled an AI-generated Christmas ad in December 2025 after a backlash. The Toys "R" Us Sora film also drew mixed reviews. The common thread is AI replacing human warmth in nostalgic or festive moments, plus visible flaws that viewers noticed and shared.

How much does an AI advertising campaign cost?

Costs vary widely with the type of campaign, the number of variants, revision rounds and talent rights. Copy and thumbnail testing is the lightest option, AI video ad packs and brand films sit in the middle, and personalized campaigns and licensed celebrity likenesses cost the most. Licensing a real person's likeness is usually the most expensive part. For a price on your own campaign, get a quote from our team.

Do I need to label AI-generated ads?

Often, yes. TikTok asks creators to label realistic AI-generated content, and YouTube requires disclosure of altered or synthetic content that could be mistaken for real people, places or events. Rules differ by platform and ad type, so check each platform's current policy. Labeling realistic AI people and scenes is the safest default, and it protects audience trust if someone notices.

Turn these examples into your own AI ad campaign

The best AI in advertising examples share one thing: a clear idea, directed by people, with AI doing the heavy lifting. The stunts need a famous brand. The repeatable playbooks, meaning personalization, creative variations, copy testing and platform AI, work for almost any brand with a product people want.

If you want to run your own version, our AI video production team builds AI video ads for Dubai and global brands, in Arabic and English, from brief to launch in 7 days. Book a strategy call and we will tell you which of these approaches fits your brand, your budget and your market.