In short: Meta’s AI tools can already create variants of images, copy and formats for Facebook and Instagram ads, and at Cannes Lions 2026 Meta announced a feature that will learn a brand’s style from its existing ads. That saves time on variants, copy, languages and formats. The risk sits in brand consistency, in claims and in realistic images of people, so work with written guidelines, human approval and testing. From 2 August 2026 the AI Act also requires realistic synthetic content to be labelled.
What did Meta announce at Cannes Lions 2026?
On 23 June 2026 Meta presented several updates for advertisers. Some are already available, some are in testing and some are still to come, so check what you actually see in your own Meta Ads Manager account:
- Brand memory: according to Meta, this AI will learn a brand’s identity and tone from its existing ads and bring that into creating new material in the same style.
- More generated text: text that appears directly on the image, created from past ad performance and Meta’s understanding of the business.
- More languages: Portuguese, French, German, Italian and Indonesian have been added for text on images, and voiceover supports further languages. Bulgarian is not on the announced list.
- Creative approval: a built-in workflow for feedback and sign-off on creative changes, in testing as of July 2026.
- Creator Marketing Hub: a single place for working with creators, which Meta plans for later in 2026.
What does this mean for your business? Producing variants becomes cheaper and faster. That is exactly why deciding what goes live matters more than creating it.
Where does AI genuinely save time?
AI is useful where the work is repetitive and a person can easily check the result:
- Variants of one creative: a different background, crop, button colour or layout, without a new photo shoot.
- Copy: several headlines and descriptions around the same idea, which the team picks from and edits.
- Languages: a first draft in another language, which is then checked by someone fluent in it.
- Formats: adapting one asset for Stories, Reels and the feed, including different aspect ratios.
The good news is that none of these tasks require AI to “invent” the brand. It works with your photos, logo and messages. If the foundation is strong, the variants are useful. If it is weak, AI simply produces more weak ads, faster. What AI can and cannot do across marketing as a whole is covered in our article on AI in digital marketing.
The same logic applies to video ads: AI helps with versions, but the idea, the story and the opening seconds remain a creative decision. There is more on this in our video content strategy.
Where is the brand risk?
1. Consistency
Each variant may look fine on its own, but together they can dilute the brand: different colours, tone and photo style. Brand memory aims to reduce this, but it does not replace a person who knows the brand and can say “that is not us”.
2. Claims
Generated copy can promise more than you offer: “free delivery”, “cheaper than anyone”, a deadline or a result that is not true. The advertiser remains responsible for the ad, whoever wrote the text. In regulated sectors such as finance and health, the risk is even greater.
3. Realistic people and images
Generated photos of people who look like real customers, staff or well-known figures are a high-risk area. They can mislead, resemble a real person or look like an endorsement nobody gave. The same goes for realistic images of a product that looks different in reality.
ScaleLab tip: Split AI features into two groups. Changes to existing material (size, format, colour correction) go through your usual approval. Anything that creates a new realistic person, product or place goes through a separate check.
How does Meta label AI ads?
Meta adds an “AI info” label when an image or video in an ad is created or significantly edited with Meta’s generative AI features, such as background or image generation. Minor changes such as resizing or colour correction do not trigger a label. The label usually sits in the three-dot menu, on the “About this ad” screen. When the edit includes a generated photorealistic person, the label may appear next to “Sponsored”.
In an update on 1 June 2026, Meta announced it will begin automatically detecting ads created or edited with AI tools outside Meta through industry-standard signals, and labelling them “AI info”. In other words, using an external tool does not mean the ad will go unlabelled.
Meta’s labels are a platform rule. They do not replace the legal obligations you have as an advertiser.
What does the AI Act change from 2 August 2026?
The EU Artificial Intelligence Act (AI Act) introduces transparency duties under Article 50, which apply from 2 August 2026. The amendments agreed in May 2026 and signed on 8 July 2026 postpone some rules for high-risk systems, but leave these duties unchanged.
Two points matter for advertising:
- Realistic synthetic content: images, audio or video created or altered with AI so that they resemble real people, places or events must be clearly labelled by whoever uses them.
- Machine-readable marking: providers of AI tools must mark generated content in a detectable format. For systems placed on the market before 2 August 2026, the deadline is 2 December 2026.
In practice, an advertiser cannot rely on the platform’s label alone. If an ad shows a realistic person or event created with AI, plan visible labelling when the ad is made. There is more on business obligations in our article on the AI Act and what it means for companies. This is not legal advice: if in doubt, consult a lawyer.
What process should you put in place?
1. Write down the brand guidelines
Colours, fonts, tone, words you use and avoid, and the claims you are entitled to make. The clearer the rules, the easier it is to check AI variants. If your brand has no written guidelines yet, sort out the visual identity and brand guidelines first.
2. Decide what AI may and may not do
For example: it may adapt formats and suggest copy. It may not create realistic people, customer reviews or images of a product that differ from the real thing.
3. Introduce human approval
One person checks every new variant before it goes live: brand fit, accurate claims, correct language and any labelling needed. If Meta’s built-in approval becomes available in your account, use it, but not as a replacement for this check.
4. Test with a clear question
Run AI variants against material made by the team and compare them on cost per enquiry or order, not just clicks. This tells you whether AI improves results or only increases the number of ads.
5. Keep a record
Note which ads were created or edited with AI and how they were labelled. If a customer, platform or regulator asks, you will have an answer.
ScaleLab tip: Before using AI features across a whole campaign, try them in one ad set and review every variant as people will actually see it.
What should you do now?
Start with written brand guidelines and a list of what AI may and may not do. Then add approval and a test for each new batch of variants. We recommend letting AI help with volume, while decisions about the brand, claims and labelling stay with people. To see how content and advertising are planned together and judged by enquiries, look at social media management.
Frequently asked questions
Do I have to label every ad made with AI?
Not every one. The AI Act requires labelling of realistic synthetic content that resembles real people, places or events. Meta, for its part, adds an “AI info” label for significant edits with generative AI. Minor changes such as resizing or colour correction do not trigger a label, but keep checking the rules and consult a lawyer if in doubt.
Can AI write ad copy in languages other than English?
AI tools can suggest copy in many languages, but Meta’s announced list of new languages for text on images is limited, and Bulgarian is not on it. Check what is available in Ads Manager for your account, and always have copy reviewed by someone who writes the language well.
Does Brand memory replace the designer or the agency?
No. According to Meta, Brand memory will learn style from existing ads, so it will depend on how good those ads are. The idea, the message, the approval and the decision about what goes live stay with people who know the brand.
What if I use an AI tool outside Meta?
In an update on 1 June 2026, Meta announced it will begin automatically detecting ads created or edited with external AI tools through industry-standard signals and labelling them. Your AI Act obligations stay the same whichever tool you used.
Sources
- Meta: Cannes Lions 2026 announcements (23 June 2026)
- Meta: Expanding GenAI transparency for Meta’s ads products (updated 1 June 2026)
- Meta Help Center: How AI-generated images in ads are identified and labeled
- Gibson Dunn: EU AI Act Omnibus agreement
- Usercentrics: EU AI Act high-risk delay and Article 50 transparency