AI in advertising: creative at scale without losing your brand

The economics of production have changed: an ad asset costs a fraction of what it used to, and platforms ask for many more versions of every idea. According to a BCG analysis published in September 2026, a personalised product shot that cost about $4 two years ago now costs about $0.04.

The same analysis estimates that large consumer goods companies need 5 to 6 million assets a year, against about 500 a decade ago. The logic repeats on a smaller scale for any brand: every format, language and audience calls for its own version. In Spain, use is still a minority practice, though above the EU average: according to Eurostat (2025 data), 11.4% of companies with ten or more employees generate images, video or audio with AI, against 9.6% across the EU.

One clarification: AI in advertising is not advertising inside AI. This article covers generative AI advertising, meaning ads created with generative models; ads that appear inside assistants, which have run in ChatGPT in Spain since 24 August 2026, are in our ChatGPT Ads guide.

The major platforms have built AI into media buying itself: they automate audiences, placements and budget and, increasingly, generate and test the creative. The advertiser supplies raw material, signals and limits.

The consequence: creative, AI ad creative included, is the main lever left in your hands. In paid social and Google Ads you decide which assets go into the system, which conversion signals it receives and what automation is not allowed to touch.

Two cautions before the rundown: what each platform says about its own tools is a vendor claim, and it is worth checking in your own account which generative features are available in your market and language. And Meta's aim to fully automate ad creation and targeting is, for now, a 2025 press report, not a confirmed product. This is what is on record as of October 2026:

• Meta Advantage+: automation of audience, placements and budget is on by default for sales, app and leads campaigns, and Meta removed detailed targeting exclusions in 2025.
• Google Asset Studio: at Google Marketing Live (May 2026), Google presented features built on Gemini and Veo: images to video, assets generated from a URL, variations and one-click creative A/B tests.
• Google AI Max for Search: an optimisation layer for Search campaigns with search term matching, text customisation and final URL expansion, plus brand controls and text guidelines.
• TikTok Smart+: an automated campaign with module-level control: targeting, budget and placements.

The public accepts AI-made ads less readily than the industry assumes. According to an IAB study published in January 2026, 82% of advertising executives believe Gen Z and millennials feel positive about AI-generated ads, but only 45% of those consumers actually do.

The gap, 37 points, has widened from 32 in 2024, and 39% of Gen Z report a negative view. One nuance: 73% of the consumers surveyed say that knowing an ad used AI would raise or not change their likelihood of buying. Our reading is that the rejection targets the cheap, the generic and the misleading, not the technology. The IAB's own advice is to put quality before savings and be consistent about disclosure. The study is not specific to Spain.

In Spain, IAB Spain's 2026 social media study (May 2026) found that 87% of social media users consider it essential to label AI-generated content. It is a different survey asking a different question, but it points the same way: transparency is valued.

The law does not require a label on everything made with AI. Article 50 of the EU AI Act, which has applied since 2 August 2026, sets specific transparency duties for the providers of the systems and for those who use them in their business, such as an advertiser or its agency.

Penalties can reach €15 million or 3% of worldwide turnover; in Spain, the AI supervisory authority is AESIA. The European Commission published a voluntary code of practice on marking and labelling in June 2026. Spain's own AI law, sent to parliament in May 2026, is still a bill. None of this is legal advice.

In everyday advertising, copy drafted with AI and edited by a person generally falls outside the labelling duty, and a clearly synthetic illustration does not, in principle, fit the idea of a deepfake. A realistic person or place that could pass as authentic does call for disclosure. The four duties in brief:

• Chatbots and voice agents: people must be told at the start that they are dealing with an AI, unless it is obvious.
• Technical marking: providers of generative AI must mark their outputs in a machine-readable format; systems already on the market before 2 August 2026 have until 2 December 2026.
• Deepfakes: AI-generated or manipulated images, audio or video that show realistic people or places and could pass as authentic must be disclosed.
• Public-interest text: AI-generated text that informs the public on matters of public interest must be identified, unless it has had human review with editorial responsibility.

The IAB's AI Transparency and Disclosure Framework recommends AI disclosure in ads only when AI materially affects authenticity, identity or representation. Published in January 2026 and revised in August 2026, it is a voluntary industry standard, not a law, and it helps you decide where the regulation is silent.

Disclosure is called for with synthetic humans, digital twins, synthetic voices and AI chatbots inside ads; routine edits do not need it. The IAB itself notes in 2026 that 83% of advertising executives already use AI in creative work: labelling everything would empty the label of meaning. A rule of your own, inspired by that framework, could look like this:

• High risk, visible disclosure: synthetic people, voices or places that look real, and any conversation with an AI.
• Medium risk, documented decision: scenes or products recreated with AI that could alter what the buyer expects to receive.
• Low risk, no label: retouching, format changes and copy edited by people.

If your advertising runs through creators, specific rules apply. In Spain, the code of conduct on influencer advertising, updated by AUTOCONTROL, the advertisers' association AEA and IAB Spain, has applied since 1 October 2025: advertising must be identifiable from the outset, there are new criteria for AI-generated content and for minors, and brands and agencies share responsibility.

So the brief should state how the partnership is identified, what may be generated with AI and who reviews the piece before it goes live. And if a creator's image or voice is recreated with AI, it is wise to have their consent in writing.

In addition, Royal Decree 444/2024, in force since 2 May 2024, places duties on ‘users of special relevance’: creators with at least €300,000 in annual income from their activity on video platforms, one million followers on a single platform (or two million across several) and 24 videos a year. Disclosing paid content, however, is expected of every creator; that is how we run creators, UGC and social commerce.

Since 27 September 2026, the rules of Directive (EU) 2024/825 against greenwashing have applied, banning generic environmental claims without substantiation. This matters for AI in advertising because a generative model can repeat that language across hundreds of variants without anyone checking it.

As of September 2026 Spain had not completed transposition, but the law firms that have analysed the directive hold that its rules apply regardless. In practice, every environmental claim needs evidence before it goes into a prompt or a template; it is as much a branding job as a compliance one. Banned practices include:

• Generic claims (‘eco’, ‘green’, ‘sustainable’) without recognised excellent environmental performance.
• Carbon neutrality based on offsetting emissions.
• Sustainability labels without certification.
• Future environmental promises without a verifiable plan.

A creative system is the set of rules, materials and templates that lets you produce many variants without the brand changing from one to the next. Instead of commissioning assets one by one, you design once how the brand looks, sounds and moves, and AI works inside that frame.

The sensible split combines a few high-craft pieces, which define the brand, with many derived variants for each format, audience and language; in a destination such as Mallorca, in Spanish and English at the very least. This is how we work in our AI creative studio, with video production when a piece calls for a real shoot, for sectors such as consumer brands and ecommerce or hospitality and travel. A creative system has five basic parts:

• Brand kit for AI: visual identity, tone of voice and examples of what is and is not on brand, in a format both people and models can follow.
• Prompt library: tested, versioned instructions by format, product and language.
• Modular templates: fixed zones (logo, typography, legal notices) and variable zones (image, headline, offer).
• Motion system: brand transitions, rhythm and animation for short video, ads and web.
• Rights and approval: a log of licences and consents, and a named person who reviews every piece.

Testing creative rigorously means stating a hypothesis before you produce, changing one thing at a time and deciding in advance what will count as a learning. Producing a hundred variants is not testing: without a hypothesis you will know which one won, but not why.

The one-click A/B tests Google announced for Asset Studio make the work easier, but they show which variant performs best on the platform's own metrics, not whether the campaign generated sales that would not have happened anyway. That takes incrementality tests and marketing measurement that cross-checks sources, as we explain in our article on marketing mix modeling. We do not promise uplift percentages: a sound method means you learn from every test, not that every test wins. A well-designed test cycle has four parts:

• Hypothesis: what you believe drives the response (the benefit, the format, the first second of the video) and why.
• Meaningful variants: distinct concepts, not twenty versions of the same ad in another colour.
• Criteria set beforehand: metric, duration and budget fixed before launch.
• Honest reading: what you learned, with what confidence, fed back into the brand kit and the prompts.

Some uses of AI are not justified by any saving, because they mislead the public, infringe rights or break platform policies. Write them into the brand's creative policy before the temptation arises. These are the red lines:

• Synthetic testimonials: invented customers, or real ones with words they never said.
• Fake or undisclosed incentivised reviews: since July 2026, Google's review snippet guidelines have included a guideline against them.
• Real people without consent: no faces, voices or recognisable imitations.
• Results the product does not deliver: generated scenes with an effect, a size or a finish the buyer will not receive.
• Unproven claims: figures, awards or environmental benefits that nobody has verified.