The Best AI-Made Ads of 2026 (So Far)

Most "best AI ads" lists are padded with tech demos and spec films that never ran. This is not that. Every campaign below actually shipped to a real audience, and the ones with numbers have public numbers. The interesting thing about the strongest work of the last eighteen months is that it doesn't cluster around one use of AI. Some ads won on speed, some on a genuinely impossible shot, some on a strategic idea that AI merely happened to power. The pattern that separates the ads people liked from the ads people mocked is worth more than the list itself, so we end there.

TL;DR

• The best AI ads of 2026 did not win because they were AI. They won on idea, timing, or an execution that traditional production could not have delivered at the price.
• Kalshi's NBA Finals spot proved the cost floor: a $2,000, two-day production that went viral in primetime.
• Heinz "AI Ketchup" turned a model's bias into the creative idea and earned over a billion impressions.
• Klarna showed the operational case, cutting image production from six weeks to seven days.
• Anthropic's Claude Super Bowl films won Cannes by making human-crafted comedy about AI advertising.
• The failures (see: uncanny holiday spots) share one trait: AI used to cut corners on work audiences expected to feel handmade.

Kalshi, NBA Finals: the cost-floor moment

In June 2025 the prediction-market platform Kalshi aired a 30-second spot during Game 3 of the NBA Finals, generated almost entirely with Google's Veo 3. AI filmmaker PJ Accetturo built it in two days using 300 to 400 Veo 3 generations cut down to 15 clips, for a reported production cost of around $2,000. That is roughly 95% below a conventional broadcast spot. It racked up over three million views on X and became the reference case for "AI ad in primetime."

What it did well: it leaned into the chaos. The ad is a string of absurd characters making wild NBA predictions, a register where Veo's occasional weirdness reads as style rather than error. It did not try to pass as a polished cinematic shoot. That is the craft decision most people miss.

Heinz "AI Ketchup": the model's bias as the idea

Heinz noticed that when you prompt an image model for "ketchup," it reliably produces something that looks like a Heinz bottle. Instead of hiding that, the brand made it the campaign: "This is what ketchup looks like to AI," and invited people to run their own prompts. The campaign earned over a billion impressions, worth many times the media investment, with a social engagement rate reported around 38% above benchmark.

What it did well: the AI was the insight, not the production shortcut. The idea only works because a generative model exists and has a specific bias. That is a category of concept that simply did not exist five years ago.

Klarna: the operational case

Klarna is not a single flashy ad, it is a system, and that is why it belongs here. Through 2024 the fintech built an internal image pipeline that produced over a thousand AI-generated images in a single quarter and cut its image development cycle from about six weeks to seven days. It reported roughly $10 million in annual marketing cost savings, including a $6 million reduction in image production, while running more campaigns, not fewer.

What it did well: it treated AI as a throughput problem, not a novelty. The lesson for brand teams is that the compounding win is rarely one hero spot. It is the fiftieth variant you can now afford to make and test. That is the same creative volume logic that reshapes performance marketing.

Under Armour, "Forever Is Made Now": the impossible shoot

When boxer Anthony Joshua was unavailable for a traditional shoot, Under Armour and studio Tool built the film from licensed archive footage blended with thousands of AI-generated images and motion graphics. The result is a stylized, dense montage that a live shoot could not have produced on the timeline.

What it did well: it used AI to solve a specific production constraint, talent availability, rather than to replace a shoot that could have happened anyway. The blend of real archive and generated frames is the honest version of the technique.

Anthropic Claude: the ad that won by criticizing ads

The Super Bowl LX films for Anthropic's Claude, produced by Mother London, imagine an AI assistant interrupted by intrusive advertising, a comedic argument for Claude's pledge not to run ads. The campaign went on to win two Film Grand Prix at Cannes Lions 2026.

What it did well: the craft is conventional and excellent. This is an AI company winning advertising's top film prize with writing and direction, not with generated footage. It is the clearest proof that the idea, not the tool, is still what wins.

The milestone that started it: Toys "R" Us

Worth including for context. In mid-2024 Toys "R" Us released a brand film made largely with OpenAI's Sora, telling the origin story of founder Charles Lazarus, and screened it at Cannes. It drew heavy backlash for uncanny footage. It belongs on any honest list because it marked the start of the era, and because it taught the lesson every entry after it had to absorb. For the full timeline, see the first fully AI-generated commercials.

The pattern: where AI actually added value

Strip out the brand names and the winners cluster into three uses, and the losers into one.

1. Speed and cost at primetime scale. Kalshi is the archetype. When a $2,000 two-day spot can run in the NBA Finals, the barrier to entry for broadcast-quality advertising collapses. This is where AI's value is least controversial and easiest to measure.

2. The impossible or unavailable shot. Under Armour's unavailable-talent problem, a location you cannot fly to, a scenario you cannot legally or physically stage. AI earns its place when the alternative is not "a cheaper shoot" but "no shoot at all."

3. AI as the idea. Heinz is the purest example. The concept depends on a property of generative models. These are the ads that could not have existed in any other medium, and they tend to be the most awarded.

Where it fails: using AI to cut corners on work audiences expect to feel handmade. The soft-drink holiday spots that got mocked in 2024 and 2025 share exactly this trait. Nostalgic, warm, human-centered advertising is the worst possible place to visibly economize with a model, because the uncanny artifacts read as disrespect for the audience. Knowing that boundary is worth more than any prompt technique.

What this means for brand teams

Pick your use case before you pick your model. If the goal is volume and speed, cheap fast models earning you dozens of testable variants beat one expensive hero render. If the goal is an impossible shot for a brand film, spend on the cinematic tier and budget human finishing time. If the idea itself is about AI, the concept has to carry the ad, the way Heinz's did.

The common thread across every winner is that a human made the call on where AI fit and where it didn't. That judgment is the job now. If you want a canvas where you can test that call across every major image and video model before you commit budget, start on 8frame and clone a variant-testing workflow to see which route your idea actually wants.