AI Act and Allegro Ads: labelling display and video ads

Information current as of 7 August 2026 · approximately 12 minutes to read

From 2 August 2026, photorealistic images or videos created with AI may need a clear label from the first moment a viewer encounters the ad. This also applies to paid Allegro Ads creatives. The platform will not add the label for the advertiser or automatically detect whether artificial intelligence was used to create the material.

This does not mean that every image edited with an AI-enabled tool needs the word “AI” added to it. For an advertiser, the key question is whether the final image, audio or video meets the definition of a deepfake: it resembles something that exists or could realistically exist and may be mistaken for authentic. This guide explains how to make that assessment, prepare a label and review active campaigns.

Important: this material is for information only and does not constitute legal advice. The assessment depends on the business’s role, how the AI system was used, the final creative and the context in which the audience sees it. Borderline cases warrant an individual legal assessment. Tweetnij

FAEO editorial visualisation showing that an AI disclosure should remain visible across ad formats and crops.

# The essentials in 60 seconds

• The AI Act does not ban the use of AI in advertising. Article 50 introduces transparency duties for specific systems and content.
• For an advertiser acting as a deployer, the visible disclosure duty applies to image, audio or video content that constitutes a deepfake—not automatically to every edit made with an AI-enabled tool.
• The Commission identifies three elements to assess: resemblance, a reference to something that exists or could realistically exist, and the risk of creating a false impression of authenticity.
• Allegro gives examples such as images and videos created from scratch, a real product against an AI-generated background, and a virtual model presenting a product. These are cases to assess, not automatic legal conclusions.
• The rules may apply to every Display Advertising format and to Video Advertising in Allegro Ads.
• The Allegro Ads dashboard has no tool that adds the disclosure. The advertiser must place it during graphic design or video editing.
• Before activation, check the preview on desktop, tablet and smartphone and make sure the label remains within the safe area.
• Allegro does not automatically check for AI use during moderation. An ad not being rejected does not confirm legal compliance.
• The process for a paid Ads creative differs from the process for listing images, where Allegro describes an automatic watermark following the seller’s declaration.
• The maximum penalties under the AI Act are high, but they do not amount to a fixed fine for a single banner. Circumstances, proportionality and the size of the business matter.

# What changed on 2 August 2026

The AI Act—Regulation (EU) 2024/1689 on artificial intelligence—entered into force on 1 August 2024, with its provisions applying in stages. Article 50 began to apply on 2 August 2026, governing, among other things, the transparency of synthetic content and deepfakes.

The date matters, but the change should be described precisely. The whole AI Act did not “enter into force” on 2 August 2026, and not every ad involving AI became prohibited. For image, audio and video creatives, the central issue is disclosure of the material’s artificial origin when it meets the definition of a deepfake.

In July 2026, the European Commission published final guidelines, FAQs, a code of practice and voluntary icons to help label content. We can therefore now explain more than was possible in the early announcements: the duties of an AI system provider and those of an advertiser using the system are not the same.

# System provider and advertiser

As a rule, the provider of a system that generates images, audio, video or text must ensure that its outputs are marked in a machine-readable format and detectable as artificially generated or manipulated. This is the obligation under Article 50(2). Among other things, the Act provides an exception for standard editing or where the input data and its meaning have not been substantially altered.

An advertiser or agency using such a system professionally may be a deployer. When it publishes an image, audio recording or video that constitutes a deepfake, it must disclose that the content was artificially generated or manipulated. This is a separate duty under Article 50(4)–(5).

Importantly, the 2 December 2026 transition concerns only the Article 50(2) technical-marking duty for providers of AI systems placed on the market before 2 August 2026. It does not postpone a deployer’s visible disclosure duty for deepfakes under Article 50(4)–(5).

In practice, this creates two layers:

1. the generator may embed a technical marker that is invisible to people,
2. the advertiser may still need a visible or audible disclosure for the audience.

Metadata, an invisible mark or a declaration stored outside the creative do not replace a message that the audience must notice no later than the first exposure.

# What is a deepfake in product advertising?

In everyday language, a deepfake is often associated with a video impersonating a well-known person. The AI Act uses a broader definition. A deepfake may also depict an object, place, entity or event.

According to the Commission’s guidelines, three criteria must be assessed together:

1. Resemblance — the material closely resembles the person, product, place, entity or event it depicts.
2. Existence or realistic possibility of existence — the representation concerns something that exists, could realistically exist or could have existed.
3. False impression of authenticity — the audience may regard the content as authentic or truthful even though it was generated or manipulated with AI.

Context matters. The same technique may lead to a different assessment in an obviously fantastical illustration and in a realistic sales banner. In advertising, buyers generally assume that the product, its use and the effect shown are represented fairly, so photorealistic creatives require particular care.

Deepfake risk can only be assessed by considering similarity, real-world plausibility and the appearance of authenticity together.

# Which creatives may require a label?

Allegro identifies three practical groups of examples. In each case, assess the final message instead of relying mechanically on the name of the tool used.

# An image or video created from scratch

If an ad shows a realistic product, variant, feature or result that does not yet exist in that form, the viewer may perceive the visualisation as a real photograph or recording. The risk increases when there is no indication that the material is a concept, visualisation or AI creative.

Generating an image from scratch does not by itself determine that it is a deepfake. An obvious illustration, abstract collage or fantastical scene may not create an impression of authenticity. Product advertising should nevertheless clearly distinguish a visualisation from what the buyer will actually receive.

# A real product against an AI-generated background

A real product shown in an artificially generated kitchen, living room, workshop or outdoor setting is one of Allegro’s examples. The Commission gives a similar example: an authentic photograph of an empty flat furnished using AI, which it treats as a partially manipulated deepfake.

Not every AI background automatically meets the definition. The key question is whether the final scene implies an authentic photo shoot, product characteristic or use case that did not take place. A simple, obviously graphic backdrop may present less risk than a photorealistic arrangement pretending to be a real interior.

# A virtual model, presenter or realistic character

A photorealistic character presenting clothing, cosmetics or a device may look like a real model. It does not have to imitate a specific well-known person to require assessment: the guidelines indicate that a person who could realistically exist may also be relevant.

For video, assess both the visuals and the audio. A synthetic voice resembling a real person or an artificially generated product demonstration may affect the viewer’s perception of authenticity. Allegro’s announcement does not provide separate instructions for an AI voice, so this case calls for a cautious, individual assessment.

# When a label is not automatically required

Typical technical operations—cropping, scaling, compression, noise reduction, exposure correction or removing a small speck of dust—usually do not create a false scene. The mere fact that software uses an algorithm marketed as “AI” does not establish that a deepfake has been created.

The boundary changes when an edit affects the material message: it adds a product feature, removes a defect, changes how the product works, creates a non-existent environment or suggests an unattainable result. The issue may then involve not only AI transparency but also misleading the consumer.

# A listing image and an Allegro Ads creative follow two different processes

This distinction is particularly important because Allegro describes two different workflows.

For images added to a listing, Allegro describes a beta tool in which the seller declares the use of AI and the platform automatically applies a watermark. We explain the details in our guide to labelling AI-generated images in Allegro listings.

For Display Advertising and Video Advertising, the Allegro Ads dashboard does not add a label. The advertiser chooses the wording, typeface, form and position of the disclosure and must include it in the graphic file or video edit.

Labelling a product image therefore does not automatically remove the need to assess a separate advertising creative, while a label on a banner does not replace the declaration required when a listing image is uploaded. AI used to create an ad should also not be confused with the “Podsumowanie AI” feature in the Allegro Ads dashboard, which analyses campaign performance.

# Which Allegro Ads formats are covered by the new rules?

The duty does not depend on the format name but on the content of the creative. Allegro states that material requiring disclosure may appear in every Display Advertising format and in Video Advertising:

• Template Ad — for the logo,
• Horizontal Banner,
• Vertical Banner,
• Brand Shop Ad,
• Top Banner,
• Homepage Ad,
• Flexible Banner,
• Video Ad.

The rules also cover materials prepared with an Advertising Business Partner, in special projects and Allegro campaigns. If you are working on a specific format, see our practical guide to the Vertical Banner in Allegro Ads.

# How to label an image or video step by step

Allegro does not prescribe a single phrase or mandatory icon. The disclosure must be clear, distinguishable, accessible and visible no later than the first exposure. Its form should reflect what AI actually changed.

# 1. Describe the extent of AI use in plain language

Instead of an ambiguous symbol, use wording suited to the material, for example:

• “AI-generated image”,
• “Image partially modified with AI”,
• “AI-generated background”,
• “AI-generated virtual model”,
• “This video contains AI-generated elements”.

These are examples of good practice, not mandatory Allegro wording. The Commission also provides voluntary icons: a general icon, “fully generated” and “partially modified”. Commission testing indicates that an icon is better understood when accompanied by plain text. Using the icon alone is not automatic proof of compliance.

# 2. Add the disclosure to the source file

Do not wait until you upload the campaign. The label must be added during graphic design or video editing because the Allegro Ads dashboard will not apply it.

Keep the source file with an editable disclosure layer. This makes it easier to prepare variants for different formats and reposition the text if the preview reveals a collision with the interface.

# 3. Account for first exposure and accessibility

The audience should not have to open terms and conditions, a campaign description or an extra menu to learn that the material is artificial. The disclosure must be noticeable at the first encounter.

In practice, ensure:

• sufficient contrast against the background,
• a size that remains readable on a phone,
• plain language without obscure abbreviations,
• no obstruction by buttons, price, logo, product label or the “Sponsored” marker,
• sufficient display time in video,
• a label on the static poster as well if it may be the first view of the ad.

# 4. Check every variant and safe area

Allegro explicitly recommends checking the preview on desktop, tablet and smartphone. An element visible in a large banner may be cropped, covered or become unreadable on a smaller screen.

Before activation:

1. open every available preview,
2. check each breakpoint and format variant,
3. verify the first frame and the Video Ad poster,
4. make sure the disclosure stays within the safe area,
5. take screenshots of the previews for the campaign record.

Allegro has not specified one universal minimum size for an AI label in its announcement. Avoid adopting a single number for every format; confirm legibility in the actual preview.

# What should you do with ads created before 2 August?

The European Commission explains that content generated before 2 August 2026 does not need to be labelled retroactively, although voluntary disclosure is encouraged. Allegro also says that it will not send advertisers a list of active creatives that they should review.

It is therefore sensible to perform an inventory instead of assuming that an “old campaign” automatically resolves the issue:

• establish when the file was generated and first published,
• check whether it was substantially edited or regenerated after 2 August,
• record the tool, scope of changes and basis for the decision,
• consider voluntary labelling if the origin is uncertain or the creative could easily mislead the audience.

The Commission’s FAQ does not resolve every scenario involving relaunch, reworking or a combination of old and new elements. Such cases should be assessed individually, especially for high-reach campaigns.

# Materials from a manufacturer, wholesaler or agency

According to Allegro, the advertiser remains responsible for the ad it publishes even when the file comes from a manufacturer, brand or wholesaler. The platform recommends asking the supplier whether the material was created or modified with AI.

Under the AI Act, the precise legal role may depend on who actually used the system and under whose control. If an agency or freelancer works on behalf of and under the responsibility of a brand, outsourcing production does not remove the risk. If a seller receives a finished file from an independent supplier, the classification may be more complex—but operationally, the seller is still publishing material whose origin it should know.

A creative brief or file-acceptance process should require:

• a declaration of whether and where AI was used,
• the tool name and date of generation or editing,
• a list of artificial and authentic elements,
• source and exported versions,
• information about the disclosure applied,
• the person responsible for approving the final creative.

# Will Allegro check and reject an unlabelled ad?

Allegro states that it does not automatically detect AI or check the disclosure during moderation. It also says that it will not reject an image or video solely because the disclosure is missing. It may, however, request an edit if the label is not visible within the safe area, and it may contact the advertiser after an ad is reported.

This is the important difference between platform checks and legal compliance. An ad may pass moderation and still require correction. The absence of an automatic alert, rejection or message from Allegro is not a green light under the AI Act.

# Penalties without clickbait

Article 99 of the AI Act sets a maximum for breaches of the Article 50 transparency duties of up to EUR 15 million or up to 3% of the undertaking’s total worldwide annual turnover in the preceding financial year, whichever is higher. For SMEs, including start-ups, the lower of the corresponding maximum values applies.

This is not an automatic penalty for one missing line of text. The type and amount of any measure must be effective, proportionate and dissuasive, with the decision taking account of factors including the nature, seriousness and duration of the breach, responsibility, intent and remedial action.

An AI label does not resolve every other risk either. The creative must still comply with Allegro’s rules, copyright, image and personality rights, and consumer-protection law against misleading advertising.

# Allegro Ads creative audit checklist

Before launching a new campaign or reviewing active material, follow these nine steps:

1. Inventory images, videos, posters and variants across all accounts.
2. Establish the origin of each file and its generation, editing and first-publication dates.
3. Ask the supplier, manufacturer or agency about AI use and retain the response.
4. Apply the deepfake test: resemblance, realistic possibility and the risk of a false impression of authenticity.
5. Choose wording that reflects the actual extent of AI use, without suggesting full generation for a minor edit or vice versa.
6. Add the disclosure to the source file before exporting the image or video to the dashboard.
7. Check every preview, safe area, first frame and poster on desktop, tablet and smartphone.
8. Archive the decision with the file version, sources, preview screenshots and approver.
9. Reassess after a change to the product, scene, edit, generator, variant or delivery method.

FAEO's practical workflow: source, assessment, disclosure, preview and decision record.

If you need help organising your creatives and campaigns, see FAEO’s Allegro Ads management and optimisation service. An operational audit does not replace individual legal advice, but it can quickly identify missing source information, labels and mobile variants.

# The most common advertiser mistakes

• labelling every image merely because the software had an AI feature,
• leaving a realistic scene unlabelled because the product itself is real,
• treating Allegro’s examples as automatic legal classifications,
• waiting for the Allegro Ads dashboard to add the label,
• placing the disclosure outside the safe area,
• checking only the desktop version,
• overlooking the first frame, audio or Video Ad poster,
• accepting a manufacturer’s file without information about its origin,
• treating successful moderation as proof of compliance,
• using the maximum penalties as clickbait without explaining proportionality,
• assuming that an AI label makes a misleading product depiction lawful,
• confusing the listing-image tool with the Allegro Ads creative workflow.

# What to remember

The new rules do not end the use of AI in advertising. They turn it from an undocumented production step into a process that must be documented and—where the final material is a deepfake—clearly disclosed to the audience.

The simplest operational rule is: know where the file came from, assess the final message, label in proportion to the AI use and test the real preview on every device. Do not label everything blindly, but do not assume that the absence of a dashboard alert removes the advertiser’s responsibility.

# Sources

• Allegro Advertising: Display and video advertising in Allegro Ads and the AI Act
• Allegro Help: new rules for Display Advertising and Video Advertising
• Regulation (EU) 2024/1689—the AI Act
• Regulation (EU) 2026/1744—limited transition for Article 50(2)
• European Commission: guidelines on the Article 50 transparency obligations
• European Commission: questions and answers on Article 50
• European Commission: icons for labelling AI-generated content
• European Commission: Code of Practice on transparency of AI-generated content