AI ads in 2026: what works, how to make them, and what the law now requires.

AI ads are the most used and least precisely defined term in marketing in 2026. One person means a video made with artificial intelligence, another means Meta's algorithm choosing its own audience, a third means ads under ChatGPT's answers. Those are three different things with different rules and different economics, and most of the confusion comes from treating them as one.

This article focuses on the first: ads whose creative is made with AI. That is where the change is biggest, and it is what the legal requirement that came into force on 2 August 2026 applies to. We produce AI ads ourselves, so we are an interested party here, and we write with that in mind.

Three things people call AI ads

The first is AI creative: video, image, voice or text generated or altered by artificial intelligence. That includes AI UGC, meaning AI-generated presenters, videos made from product photos, and stylised formats like claymation or stop-motion that used to need a studio and weeks of work.

The second is AI-driven delivery. Meta's Advantage+ campaigns decide for themselves who sees an ad, where and on what budget, and Advantage+ creative uses AI to add backgrounds to existing assets, expand images and put still images into motion. The Wall Street Journal reported in June 2025 that Meta aims to fully automate ad creation, end to end, by the end of 2026.

The third is ads inside AI assistants. Since 24 August 2026 ChatGPT has been showing ads in Estonia too, which we covered separately in ChatGPT ads in Europe.

Why AI creative changed the economics of advertising

The common view is that AI ads are good because they are cheap. Cheapness is real, but it is not the main reason. The main reason is that AI makes testing cheap.

Results on Meta depend more on how fast you find a working angle than on how beautiful any single ad is. Our creative testing framework is built on testing one variable at a time: same message, different opening, round after round. With a traditional shoot, every new variation costs a day of work. With AI, ten different opening frames cost what one used to, and the opening frame is what sets hook rate, which in turn sets everything else.

In Estonia there is an extra benefit. In a small country the audience saturates quickly and creative fatigue arrives in weeks rather than months. When ads wear out faster, new creative has to arrive faster, and traditional production cannot keep that pace. AI does not make an Estonian advertiser smarter, but it lets them keep the pace the size of the market demands.

Which AI ads work

The AI ads that work best are the ones that do not pretend to be something they are not. Claymation, crochet stop-motion, talking objects or a skeleton working through your customer's problem are obviously artificial, and that is their advantage. The viewer does not stop to wonder whether it is real. They watch to see what happens next.

Using AI to multiply existing material also works well: new openings for an ad that already performs, different formats for Feed and Reels, copy in several languages, a moving version of a product photo. Nothing new is invented here. You are extending what has already proven itself in testing.

AI UGC, meaning ads with an AI-generated presenter, is the most contested category. For a product demo or an explainer it can work well. An AI person presented as a customer giving their opinion, though, is where trust, platform rules and the law collide, as the section below shows.

What does not work

The generic AI look. Over the last couple of years people have learned to recognise the typical AI image fast: skin too smooth, light too perfect, smiles too even. The more an ad looks like ordinary AI output, the less reason there is to stop scrolling, and it shows up immediately in hook rate.

Unchecked Estonian. AI writes and speaks English considerably better than Estonian. A voiceover with the stress in the wrong places, or copy that reads like a translation, costs more in an Estonian ad than you would expect, because the native language is exactly what stops the scroll here, as we wrote in the Baltic ad costs article. Have every piece of Estonian copy and voice checked by a native ear.

Unchecked automation. Meta itself says some Advantage+ creative enhancements may be turned on by default. That means AI can add a background or an animation to your ad that you have never seen. Review the preview before publishing and switch off whatever does not fit your brand.

Faked proof. The ban on misleading advertising has applied in Estonia for a long time, and it does not care which tool made the ad. An invented customer praising a product is misleading advertising even if they were drawn by hand.

What the law requires from 2 August 2026

Article 50 of the EU AI Act has applied since 2 August 2026. The Digital Omnibus, which postponed the requirements for high-risk AI systems, left the transparency obligations essentially untouched. Every advertiser using AI-generated visuals should know them.

For advertisers the most important provision covers deepfakes. By the European Commission's definition, a deepfake is AI-generated or manipulated image, audio or video content that resembles existing persons, objects, places, entities or events and would falsely appear to a person to be authentic. If you use that kind of content in an ad, you are a deployer under the Act and must disclose that the content has been artificially generated or manipulated.

The disclosure has to be clear and distinguishable, and it has to come at the latest at the moment a person is first exposed to the content. Small print at the end of an ad that nobody sees is unlikely to meet that bar.

For evidently artistic, creative, satirical or fictional work, the requirement is lighter: it is enough to disclose the use of AI in an appropriate way that does not hamper enjoyment of the work. A stylised format that nobody takes for real is therefore the safest choice both commercially and legally. A realistic AI person presenting themselves as your customer is the riskiest.

Tool makers carry their own obligation: their output must be marked in a machine-readable way as AI-generated. Systems placed on the market before 2 August 2026 have a transition period until 2 December 2026. Fines reach up to 15 million euros or 3 percent of worldwide annual turnover. This section is a practical summary, not legal advice, and borderline cases are worth running past a lawyer.

How to make AI ads that work

Start from the message, not the tool. AI turns a bad idea into a lot of bad ads very quickly. Decide the angle and the promise first, the format second.

Use AI for variation. The biggest value is not one perfect ad but ten different openings for the same message that already works. Test them one variable at a time and let the data decide.

Keep the style honestly stylised. A format that is obviously artificial asks no belief from the viewer and no explanation from you.

Measure exactly as before. AI changes production, not metrics. Hook rate, hold rate, click-through rate and cost per lead still tell you what works, and fatigue reaches AI ads the same way it reaches every other ad.

Label anything that could pass for real. If you are unsure whether content is a deepfake, treat it as one.

The takeaway

AI ads are not one thing but three: AI-made creative, AI-driven delivery and ads inside AI assistants. The biggest change is in the first, because it makes testing cheap and lets an advertiser in a small market keep the pace that fast saturation demands. The best results come from honestly stylised formats and variations of proven messages. The worst come from the generic AI look, unchecked Estonian and faked proof. Since 2 August 2026, realistic AI content in an ad has to be labelled. If you want to know which AI format fits your brand, or see how we make them, book a 30-minute call.