# Labelling AI videos in your online store and in ads: product videos, avatars, AI voices > When an AI product video, an AI avatar or an AI voice is a deep fake, and how the label must look in the video, in your store and in ads. As of 1 October 2026. URL: https://easyaiactlabels.com/en/guides/ai-videos-labelling/ [Guides](/en/guides/) /Obligation # Labelling AI videos in your online store and in ads: product videos, avatars, AI voices Last updated: 1 October 2026 Responsible: Martin Jäger Created with AI 25 min read Short answer Article 50(4) of the AI Act applies to video and audio just as it does to images: an AI video needs a label if it works as a deep fake. Under the Commission guidelines, realistic AI avatars, AI models, cloned voices and photorealistic generated scenes all count, and according to their examples the artistic exception does not cover ad videos that demonstrate a product. The label belongs at the start of the video at the latest and has to be repeated if viewers can foreseeably join part-way through (para. 143); the code of practice names the top right corner and encourages a label throughout. Neutral AI voices and videos animated from your own product photo remain open; Meta, YouTube and TikTok have their own rules, which partly differ. As of 1 October 2026. **Research as of 1 October 2026.** This guide covers video and audio that you use in your online store or in ads: product videos from image-to-video generators, fully generated clips, AI models and AI avatars, UGC-style AI ad videos and AI voices used as voiceovers. The basics, that is who the deployer is and what makes a deep fake, are in the guide [Do I have to label AI images in my online store?](/en/guides/ai-images-labelling-obligation/); they apply to videos in the same way. Easy AI Act Image Labels checks and labels images only, not videos. For the full picture, see [the Article 50 overview](/en/guides/eu-ai-act-article-50/). ## Does the labelling obligation apply to video and audio too? Yes. Article 50(4) of the AI Act (Regulation (EU) 2024/1689) names video and audio expressly: “Deployers of an AI system that generates or manipulates image, audio or video content constituting a deep fake, shall disclose that the content has been artificially generated or manipulated.” The definition in Article 3(60) is just as broad: a deep fake means “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 or truthful”. Recital 134 adds that the resemblance has to be appreciable. For you, this means: - **You are the deployer** if you create or edit the video with an AI tool yourself, or if you decide for an agency or a creator whether and how AI is used (guidelines, para. 14). If you merely order a finished video without deciding on the use of AI, you are not a deployer under para. 14; more in the guide [AI stock images in your online store](/en/guides/ai-stock-images-labelling/). - **The obligation has applied since 2 August 2026**, with no transition period for deployers. The Digital Omnibus (Regulation (EU) 2026/1744) changes Article 50 only in paragraph 7 (Article 1 point 20) and Article 3 only in point 14 and through the new points 14a and 14b (Article 1 point 4); the deep fake definition is unchanged. The deadline of 2 December 2026 in Article 111(4) concerns only the providers of generators and their machine-readable marking under Article 50(2). - **A marker in the video does not replace a label.** Under the guidelines, providers have to mark both the video and the audio of a video (para. 60). You cannot rely on that marking, however, because the disclosure must be perceivable by people without technical tools, for example “with visible or audible labels” (para. 117). ## When is an AI video a deep fake? The European Commission’s guidelines on Article 50 (annex to C(2026) 5054 final of 20 July 2026) look at four criteria that must all be met (para. 113): appreciable resemblance to something that exists or could plausibly exist, namely persons, objects, places, entities or events, and a false appearance of authenticity. Four points matter most for videos: - **Persons are more than faces.** They include “digital replicas of real persons, realistic AI-generated human avatars or personas, and personal characteristics or expressions, such as image, voice, behaviour, performances” (para. 113). - **Products are objects.** “Objects” expressly covers consumer goods, and authenticity includes “the actual appearance or use of objects or the delivery of services” (para. 113). - **Essential elements count, backdrops less so.** Under para. 114, AI backgrounds and special effects in standard film production are not likely to make content appear authentic. It is different for “fully AI-generated actors, digital replicas of real or deceased actors, de-aging, simulated performances including parts thereof”. Photorealism makes a deep fake more likely, but “photorealism alone is not determinative” (para. 114). - **Minor technical edits do not count.** Para. 116 lists, among others, “adapting audio parameters, colour correction, noise reduction”, each subject to the specific case. Intent to deceive is not required; what matters is the context and the foreseeable audience, with a stricter view where children or older people are part of it (paras. 114, 115). The Wettbewerbszentrale, the German industry body against unfair competition, points out something that particularly affects videos in feeds; translated from German: “If it is not shown permanently but only for a few seconds, viewers may in case of doubt be less able to tell whether it is an AI image.” The guidelines’ example lists include several video and audio cases. **Deep fakes** include: AI audio cloning the voices of a newspaper podcast’s regular presenters; an AI video featuring an AI-generated depiction of a celebrity influencer in an advertising or promotional context; an AI video featuring a realistic synthetic avatar of a company CEO. **Not deep fakes** are: an AI video of mice arguing about cheese in an advertising campaign for a cheese manufacturer; AI voice replication for fictional characters in audiobooks, games or animation where there is no deception as to the identity of the narrators; a radio broadcast in which AI only adjusts volume, noise and compression without altering the words or the way of speaking; a real product, such as a car, shown in an ad against an AI background, as long as the ad does not mislead about the product. The AI Service Desk at RTR in Austria expressly gives as an example of a deep fake, translated from German, “the AI-generated depiction of a celebrity in a commercial”. It informs and advises, but it is not a supervisory authority. ## Which store and ad videos need a label? The table classifies typical cases. “Our own reading” means that the sources do not mention the case expressly and the classification follows from the provisions named. Case Label? Basis Fully generated, photorealistic product or lifestyle video (text-to-video) Yes Para. 113 (consumer goods, realistic places and scenes); example of an AI product image in advertising Your own product photo animated with image-to-video, the product moves or is used (fabric falls, cream is absorbed, a device runs) Yes, when in doubt Para. 113 (“actual appearance or use of objects”); the guidelines do not mention image-to-video; our own reading Your own product photo, AI only creates a camera move, the product stays unchanged Open No statement in the guidelines; para. 116 does not mention animation Photorealistic AI model or virtual try-on in a video Yes Para. 113 (realistic AI avatars); example of a synthetic influencer, not an artistic work Photorealistic AI avatar as presenter, including as your digital twin Yes Para. 113; CEO avatar example; RTR FAQ UGC-style AI ads: AI people demonstrate or recommend the product as customers Yes, plus unfair competition law Teleshopping and synthetic influencer examples after para. 124; UWG annex no. 23c (our own reading) Clearly illustrated avatar with a neutral voice Usually no RTR FAQ Clearly illustrated avatar with an AI clone of your voice Yes, in RTR’s view RTR FAQ Clone of the voice of a real, well-known person Yes, plus personality rights Podcast voices example; RTR FAQ; para. 129 Clone of your own voice as voiceover or for a dubbed version Likely yes RTR FAQ (avatar case); para. 114 (simulated performances); the guidelines do not mention dubbing; our own reading Neutral synthetic voice without a real model (text-to-speech) Open No statement in the guidelines; platforms differ AI background music Open Para. 124 names AI music resembling the style of existing artists as a deep fake in an artistic work; nothing on generic music Real product video against an AI background, without misleading about the product Usually no Car against AI background example; para. 116 Real footage, AI only for noise reduction, colour correction, audio adjustment No Para. 116; radio broadcast example Clearly artificial: cartoon, talking animals, physically impossible scenes No Para. 113(ii); mice example In every case: if a video shows your product differently from what you deliver, for example a texture, a fit or an effect it does not have, no label helps. More on that below. ## How must the label look in a video? **What the law requires.** Under Article 50(5) of the AI Act, the information must be provided “in a clear and distinguishable manner at the latest at the time of the first interaction or exposure” and conform to the applicable accessibility requirements. Under the guidelines, that is not the case if the notice can easily be overlooked under normal conditions, for example in a manual, in menu layers or in terms of use (para. 142). **What the guidelines say about timing.** The obligation applies to every person who sees the video, not just the first (para. 143). As examples of first exposure, the guidelines name “at the start of a video featuring a deep fake” and, when scrolling on social media, the moment someone encounters the deep fake. A notice only at the beginning is not always enough, though: “if it is reasonably foreseeable that persons may not perceive content from its beginning, then only disclosure at the beginning of content does not adequately inform those persons and should be complemented with disclosure at later moments, where possible” (para. 143). **What the code of practice sets out in detail.** The Code of Practice on Transparency of AI-generated Content binds only its signatories. Under the guidelines, those who do not sign it are expected to show in other ways how they meet the obligation, for example by comparing their measures with the code (para. 148). For videos, Section 2 provides: - **Where**: in a place without intervening overlay elements, for example “in the top right corner of an image or video deep fake” (sub-measure 1.2.2(a)), with sufficient spacing to other overlays, disclaimers and on-screen elements, and visible against any background (1.2.1(d)). - **When and for how long**: “at the beginning of the video as well as, where possible, at regular intervals throughout the video and, at a minimum, after interruptions”, such as advertising breaks (1.2.2(b)). Signatories are also encouraged to display the label throughout the deep fake video or during the deep fake part of the video. The label has to remain visible long enough to be noticed under normal exposure conditions (1.2.1(b)), without requiring any interaction or sustained attention (1.2.1(a)). - **Visual before audio**: for visual deep fakes, audible disclosures may only be an additional method and are always accompanied by a visual one (1.2.2(d)). For audio deep fakes, the code provides at the beginning for “a short audible disclaimer in plain and simple natural language, either in the same language as the content or in English” (Measure 1.1) and, where a screen is available, additionally the icon (1.2.2(e)). For long audio-only content, reminders at regular intervals are added where appropriate (1.2.3). - **Embedded or as an overlay**: the label is embedded directly into the content, unless there is an equivalent alternative such as a user interface overlay that appears to be on the content (1.2.1(c)). **Which icons exist.** There is no separate video version of the EU icons; the Commission’s page refers the icons to deep fakes in image, audio and video (page as of 24 September 2026). As examples, the page names fully AI-generated deep fake videos for AI GENERATED and, for the basic icon, a deep fake video with the text label “voices generated with” followed by the basic icon. According to Annex 1 of the code, an audio-only icon is still to be developed. On accessibility, the Commission’s page recommends: “If the disclosure appears for a limited time, it should remain visible long enough to be read and understood by users with cognitive or processing difficulties.” Variants and the language question: [The EU icons for AI images](/en/guides/eu-icons-ai-images/). **Which variant for which video?** The Commission assigns AI GENERATED to content fully generated by AI with no human-created elements apart from prompting, and AI MODIFIED to pre-existing, human-made content partially modified with AI into a deep fake. For videos, our own reading is: Video Variant Text-to-video, AI avatar video without your own footage AI GENERATED Your own product photo or footage, animated or altered with AI AI MODIFIED; AI GENERATED if the origin of the source material is unclear Real footage, only the voice comes from AI Basic “AI” icon with a short text such as “AI-generated voice”, following the example on the Commission’s page ## How to put it into practice in your store and in ads This leads to a practical approach. It is this guide’s reading of Article 50(5), para. 143 and the code, not a requirement: 1. **Burn the label into the video, top right.** A label in the video itself travels with it when you use the same file in your store, in ads and on social media. Keep it clear of subtitles, logos and the player controls. 2. **Show it throughout short loops.** Product videos often loop silently in stores, and visitors join part-way through. A label only in the first seconds does not reach them; a label throughout is the simplest way to meet para. 143 here. For longer videos in which only one section contains AI, the code requires the label at the beginning, where possible at intervals and after interruptions; it encourages showing it during the AI section. 3. **Think of the preview image.** Before playback, visitors first see the preview image. If it already shows the deep fake, the label belongs there too (our own reading: first exposure). 4. **Label AI voices visibly.** In stores and feeds, videos often play without sound; a spoken notice alone then reaches no one. A visible text such as “AI-generated voice” with the basic icon therefore belongs in the picture; a short spoken notice at the beginning can add to it. 5. **Add a description and alt text.** Like images, Shopify videos have a field for alternative text. A sentence such as “Video created with AI” helps screen reader users, but it does not replace a visible label. 6. **Keep the original without a label.** Keep a version without the burned-in label and derive a labelled version for each channel. That way you can adjust position and variant when a platform adds its own label. 7. **Record your decision.** For each video, note the tool, the source material, where it is used and the decision for or against a label. Non-signatories of the code are expected to be able to explain their approach on request (para. 148). ## What about satire, art and fiction in advertising? Article 50(4) of the AI Act softens the obligation: “Where the content forms part of an evidently artistic, creative, satirical, fictional or analogous work or programme, the transparency obligations set out in this paragraph are limited to disclosure of the existence of such generated or manipulated content in an appropriate manner that does not hamper the display or enjoyment of the work.” The disclosure does not disappear; it just becomes more discreet (para. 123). For advertising, the guidelines draw narrow limits: - **It has to be evident.** The categories are to be interpreted strictly; content whose nature is “exclusively informative or commercial and is recognisable as such” is excluded (para. 122). - **Ads only case by case.** Ads containing deep fakes “might be regarded as evidently creative or fictional works in certain, specific situations, but not in others” (para. 122). Where content combines informative and creative elements, “the informative character” should always prevail. - **Product demonstrations are not art.** According to the examples after para. 124, the following are expressly not artistic works: an AI video in the style of a teleshopping channel in which simulated humans demonstrate a product in an AI-generated scene to persuade viewers to buy it, and an AI video in which a realistic synthetic influencer tests a sponsored real product. A spot with mice arguing about cheese needs no label at all under the guidelines’ example, because it is not a deep fake. A clearly fictional brand film with realistic AI actors may, in a specific case, fall under the softer regime (our own reading). The notice may then, under Commitment 3 of the code, sit in the description, in the opening or closing credits or next to the video frame, but it must be perceivable without a click. Third parties’ rights must still be respected (para. 124). ## What applies if a real person’s face or voice appears? A label does not replace consent. The guidelines state that deployers of deep fakes of real persons have to comply with the rules protecting “personality rights over personal images or voices” (para. 129) and, for identifiable living persons, with data protection law, for example by relying on a legal basis for the processing (para. 127). The code states that labelling does not exempt anyone from other laws, expressly including the “applicable legal requirements regarding obtaining consent of the depicted persons”. As a pointer, not advice on your specific case: - **Germany**: under Section 22 KUG, portraits may be distributed or publicly displayed only with the consent of the person depicted. Section 23 KUG contains exceptions, for example for portraits from the sphere of contemporary history, which do not apply where legitimate interests of the person depicted are infringed (Section 23(2)). The KUG does not cover the voice; for the voice, the general right of personality comes into consideration. The Wettbewerbszentrale writes, translated from German: “If the AI avatar is deceptively modelled on a real person, it counts as a deep fake.” - **Austria and Switzerland** have their own rules: in Austria, the protection of portraits under Section 78 of the Copyright Act, which prohibits distribution where legitimate interests of the person depicted would be infringed; in Switzerland, the protection of personality under Article 28 of the Civil Code ([Labelling AI images as a Swiss shop](/en/guides/ai-images-labelling-switzerland/)). - **Platforms go further.** In the EU, TikTok Shop prohibits content that shows celebrities, medical professionals or public figures, including their voices, for product promotion, translated from German, “even if the content is correctly labelled as AIGC”, as well as AI avatars, voices or personas that imitate medical professionals (as of 28 August 2026). TikTok’s advertising policy prohibits content that misuses the likeness or voice of public figures without their permission. This also applies to you and your team: according to RTR, a photorealistic digital twin is a deep fake that needs a label, and in Germany Section 22 KUG applies to an employee’s portrait as it does to anyone else’s. ## What applies regardless of the label? Article 50(6) of the AI Act leaves other transparency obligations unaffected. Two rules of German unfair competition law particularly affect AI videos: - **Product accuracy.** Under Section 5(2) no. 1 UWG, a commercial practice is misleading if it contains untrue or deceptive statements about essential characteristics of the goods, such as their type or design. A video showing texture, drape, size or effect differently from what is delivered does not become acceptable with a label; it needs replacing. Austria and Switzerland have their own laws: [Austria](/en/guides/ai-images-labelling-austria/), [Switzerland](/en/guides/ai-images-labelling-switzerland/). - **Invented customer voices.** Under no. 23c of the annex to the UWG, submitting or commissioning fake consumer reviews or recommendations, and misrepresenting consumer reviews or recommendations on social media to promote sales, is prohibited; under Section 3(3) UWG, practices listed in the annex are always prohibited towards consumers. No court has decided whether a UGC-style AI video in which an AI person appears as a satisfied customer is such a fake recommendation (our own reading: it is likely if the video comes across as a genuine customer experience). An AI label does not reliably remove that impression. ## What do Meta, YouTube, TikTok and Google require for videos? The platforms have their own rules, some narrower and some broader than the AI Act. Platform pages as of 1 October 2026; the rules change often, so check them there before every campaign. Channel Self-disclosure required AI voice What the platform does itself Meta (Facebook, Instagram, Threads), posts photorealistic video and realistic-sounding audio reel with a realistic AI voiceover and song with AI vocals given as examples that require a label “AI info” on industry signals or self-disclosure Meta, ads no requirement found for commercial ads; separate rule for ads about social issues, elections and politics no specific rule found label for Meta’s own AI tools; third-party tools via signals, label under “About this ad” in the three-dot menu YouTube realistic content generated or meaningfully altered with AI cloning your own voice for voiceovers or dubs expressly needs no disclosure; AI music does label below the player, on Shorts on the video; automatically for C2PA, own AI tools and internal detection TikTok, posts (from 24 September 2026) realistic-looking people or scenes required if AI audio mimics the voice of a real person; generic text-to-speech needs none AIGC label, also automatically on technical metadata TikTok, ads completely AI-generated images, video or audio, significantly altered content required for AI voice cloning toggle “This ad contains AI-generated content”; otherwise rejection or restriction Google Ads, including YouTube ads no specific requirement found; label setting per asset no specific rule found overlay on the ad for EU campaigns with labelled assets, worldwide notice in the menu ### Meta - **Posts**: “Meta requires you to label content you share that has photorealistic video or realistic-sounding audio that has been digitally generated or altered, including with AI” (Meta help on Threads). As examples, the page names “A reel narrated with a realistic AI-generated voiceover” and “A song created using AI-generated vocals”; a label is not needed, for instance, for a landscape video in a cartoon style. For Facebook and Instagram, Meta announced the same rule for organic content and added: “we may apply penalties if they fail to do so” (Newsroom, 6 February 2024). - **Ads**: no self-disclosure requirement for ordinary commercial ads could be found. Meta labels images and videos created or significantly edited with its own generative tools; if the ad then includes a photorealistic AI human, the label appears next to the “Sponsored” label. Since an update of 1 June 2026, Meta also aims to detect ads made with third-party AI tools through industry signals and to show the label under “About this ad” in the three-dot menu. For ads about social issues, elections or politics, advertisers must disclose digitally created or altered images, video and audio. ### YouTube - **Requirement**: creators must disclose AI content that makes a real person appear to say or do something they didn’t do, alters footage of a real event or place, or generates a realistic scene that didn’t actually occur; this is done in YouTube Studio under “AI use”. YouTube expressly requires no disclosure for “Cloning one’s own voice to create voice overs or dubs”, nor for beauty filters, colour adjustment or caption creation. You do have to disclose “AI generated music”. - **What YouTube does itself**: since May 2026, the label for photorealistic content appears directly below the player, and on Shorts as an overlay on the video (blog, 27 May 2026). YouTube applies it automatically to content from its own AI tools, to content with C2PA metadata and on internal detection. - **Note**: the fact that YouTube requires no disclosure for cloning your own voice says nothing about the AI Act. RTR sees a labelling obligation where your own voice is reproduced with AI (see above). ### TikTok - **Posts**: under the Community Guidelines effective 24 September 2026, you must label content that uses AI or significant edits to show realistic-looking people or scenes, with the AIGC label or your own clear caption, sticker or watermark. It is expressly required when “AI-generated audio mimics the voice of a real person”, and not needed for “generic text-to-speech (TTS) narration, when the TTS isn’t a recognizable voice of a known individual”. Among other things, AI content showing a public figure supporting products they have never actually addressed is not allowed. - **Ads**: under the advertising policy (as of April 2026), you must label “Content that contains images, video, or audio that are completely AI-generated” and content substantially altered with AI, for example making someone say something they didn’t say using AI voice cloning. You do this with the toggle “This ad contains AI-generated content” in Ads Manager or your own clear disclaimer, caption, watermark or sticker. “If we identify AI-generated content that has not been disclosed, your ad will be rejected or restricted.” Spark Ads follow the Community Guidelines. - **TikTok Shop**: the EU version of the Shop policy of 28 August 2026 requires disclosure in the content or via a toggle and prohibits the AI depictions of celebrities and professionals mentioned above. ### Google Ads and YouTube ads “Starting in July of 2026, Google permits adding text or visual labels directly within image and video ad creatives that were generated or modified using AI”, without breaching its rules on text overlays and watermarks. Assets marked with the label setting get a worldwide notice under “How this ad was made” in the ad’s menu; for campaigns targeting the European Union, India and New York, the ads also show a visible overlay. If you upload a video to YouTube from within Google Ads, Google asks whether AI was used. Google itself points out that the setting does not guarantee compliance with specific regulations. ### What this means for you If a very large online platform provides a tool that lets you disclose a deep fake clearly and distinguishably, you can rely on it within that platform; your responsibility remains (guidelines, para. 126). A notice in menu layers is not considered clear, however (para. 142). Where the platform label sits only behind a menu, as with Meta’s ads, your own label belongs in the video. More on ads and marketplaces, including images: [AI images on marketplaces and in ads](/en/guides/ai-images-marketplaces-ads/). ## What does Shopify do with product videos? - **Formats and limits**: Shopify accepts video files as .mp4, .mov or .webm, up to 10 minutes, up to 1 GB and up to 4K. “Shopify converts and serves all videos as .mp4 or HLS format”; in browsers that support streaming, videos are shown in 480p, 720p or 1080p depending on the connection and the resolution of the upload (Shopify Help Center). Alternatively, you can embed YouTube or Vimeo links. - **Stored versions**: according to the Admin API, a Shopify video has several sources: “Types of sources include the original video, lower resolution versions of the original video, and an m3u8 playlist file.” The video object also has a field for alternative text. - **Metadata**: whether C2PA data or other provenance markers in an AI video survive in the versions delivered has not been measured. The help page says nothing about it. For images, it has been measured that the delivered copies mostly lose their markers: on 30 September 2026, only 6 of 38 marked images still carried their marker on the delivered copy. This cannot be transferred to videos. - **Consequence**: a label burned into the video is part of the picture; a marking in the metadata is not visible to visitors anyway and is not enough under para. 117. ## What about older videos? Under the guidelines, deep fakes generated or manipulated before 2 August 2026 do not need to be labelled retroactively; deployers and others who possess or disseminate such content are encouraged to do so without disproportionate effort (para. 154). RTR reads this more strictly. Details, which apply to videos in the same way: [Do AI images from before 2 August 2026 need a label?](/en/guides/images-uploaded-before-august-2026/) If you alter or re-voice an older video with AI after 2 August 2026, there are good reasons to treat this new manipulation as covered by the obligation (our own reading). ## What is still open? - **Image-to-video**: the guidelines do not say whether a video animated from your own product photo is a deep fake. The classification above is our own reading, as is the choice between AI MODIFIED and AI GENERATED. - **Neutral AI voices**: there is no official statement on a synthetic voice without a real model. RTR’s example is the case where AI generates a well-known human voice; the platforms differ (Meta requires a label, TikTok does not). - **Your own voice and dubbing**: RTR sees an obligation where your own voice is reproduced with AI, YouTube requires no disclosure; the guidelines do not mention AI dubbing or lip-syncing. - **AI music**: whether generic AI background music in a product video can be a deep fake is unclear. - **How often to repeat**: the code does not define “at regular intervals” in seconds. - **Shopify**: whether provenance markers in videos survive the conversion to MP4 and HLS. - **Binding force**: the guidelines are non-binding (para. 5), the code binds only signatories, and RTR is not a supervisory authority. No enforcement case on labelling AI images in online retail was known as of 30 September 2026 ([tracker October 2026](/en/guides/ai-labelling-enforcement-october-2026/)); cases involving videos were not searched for separately. ## In short The same rule applies to AI videos as to images: if the video works as a deep fake, you need a visible label. Under the Commission guidelines, realistic AI avatars, AI models, cloned voices of real persons and photorealistic generated scenes are included; according to the guidelines’ examples, ad videos demonstrating a product are not art within the meaning of the exception. The label belongs at the start at the latest and has to be repeated if viewers can foreseeably join part-way through; for short product videos, a label in the top right corner throughout is the simplest solution. A label replaces neither a real person’s consent nor a video that shows your product differently from what you deliver. General information, not legal advice on your specific case. ## Frequently asked questions ### Do I have to label an AI product video in my store? If it is a deep fake, yes: Article 50(4) of the AI Act expressly covers image, audio and video content. Under the Commission guidelines, a video is a deep fake if it appreciably resembles existing persons, objects, places or events and would falsely appear authentic (para. 113); this expressly includes consumer goods and how they actually look or are used. You therefore label a photorealistic generated product video or a video with an AI model. Real footage that AI has only denoised, colour-corrected or adjusted in its audio usually needs no label (para. 116). ### Is a label at the start of the video enough? Not always. Under the guidelines, a disclosure only at the beginning is not enough if it is foreseeable that viewers will not see the video from the start; the disclosure should then be repeated later where possible (para. 143). The code of practice requires its signatories to show the label at the beginning, where possible at regular intervals and at a minimum after interruptions, and encourages showing it throughout the deep fake (sub-measure 1.2.2(b)). For product videos that loop in your store, a label in the top right corner for the whole video is the simplest solution (our own reading). ### Do I need a label if only the voice in the video comes from AI? For a cloned voice of a real person, yes: the guidelines expressly list the voice among the characteristics of persons (para. 113) and give AI audio with cloned voices as an example of a deep fake. The AI Service Desk at the Austrian regulator RTR also sees a labelling obligation where an avatar reproduces your own voice with AI although you did not speak the text. A neutral synthetic voice without a real model remains open; the guidelines say nothing about it. Meta names a reel with a realistic AI voiceover as an example that requires a label, while TikTok requires no disclosure for generic text-to-speech voices. ### Are AI avatars allowed in UGC-style ad videos? Article 50 of the AI Act does not prohibit them, but it requires a label: realistic AI avatars are persons within the meaning of the deep fake definition (para. 113), and the guidelines expressly name a video of a realistic synthetic influencer testing a sponsored product as a deep fake that is not an artistic work. If AI people appear as satisfied customers, German unfair competition law also comes in: fake consumer recommendations are always prohibited under no. 23c of the annex to the UWG; whether an AI testimonial falls under it has not been decided (our own reading). On TikTok Shop, AI depictions of celebrities and medical professionals in product promotion are also prohibited, even with a label. ### Does the exception for artistic or satirical works apply to ad spots? Rarely. Article 50(4) of the AI Act only softens the obligation for evidently artistic, creative, satirical, fictional or analogous works, and even then a disclosure is still needed, just in a form that does not hamper the enjoyment of the work. Under the guidelines, content that is exclusively commercial and recognisable as such is excluded; ads can be creative or fictional works only in certain, specific situations (para. 122). An AI video in the style of a teleshopping channel and a synthetic influencer testing a product are expressly not such works. ### Is the AI label from Meta, TikTok or YouTube enough? Within the platform it can be: if a very large online platform provides a tool that lets you disclose a deep fake clearly and distinguishably, you can rely on it there; your responsibility remains (guidelines, para. 126). A notice that sits only in menu layers is not considered clear (para. 142); this affects, for example, Meta's “About this ad” in the three-dot menu. For your own store and for channels without a visible platform label, you need your own label in the video. ## Sources 1. Regulation (EU) 2024/1689 (AI Act), CELEX 32024R1689, Articles 3(60), 50(2), (4), (5) and (6), recitals 133 and 134, Cellar resolver. [publications.europa.eu](http://publications.europa.eu/resource/celex/32024R1689) (accessed 1 October 2026) 2. Regulation (EU) 2026/1744 of 8 July 2026 (Digital Omnibus on AI), CELEX 32026R1744, Article 1 points 4, 20 and 39(b), Cellar resolver. [publications.europa.eu](http://publications.europa.eu/resource/celex/32026R1744) (accessed 1 October 2026) 3. European Commission, Guidelines on the implementation of the transparency obligations under Article 50, annex to C(2026) 5054 final of 20 July 2026, paras. 5, 12, 14, 60, 92, 113 to 117, 119 to 129, 141 to 143, 148, 154. [ec.europa.eu](https://ec.europa.eu/newsroom/dae/redirection/document/131215) (accessed 1 October 2026) 4. European Commission, Code of Practice on Transparency of AI-generated Content, Section 2, Commitment 1 (Measures 1.1, 1.2), Commitment 3, Annex 1. [ec.europa.eu](https://ec.europa.eu/newsroom/dae/redirection/document/129555) (accessed 1 October 2026) 5. European Commission, EU icons for labelling AI-generated content (page as of 24 September 2026). [digital-strategy.ec.europa.eu](https://digital-strategy.ec.europa.eu/en/policies/eu-icons-labelling-ai-generated-content) (accessed 1 October 2026) 6. European Commission, Transparency obligations under Article 50 of the AI Act (FAQ, page as of 24 July 2026). [digital-strategy.ec.europa.eu](https://digital-strategy.ec.europa.eu/en/faqs/transparency-obligations-under-article-50-ai-act) (accessed 1 October 2026) 7. RTR, AI Service Desk (KI-Servicestelle), FAQ, questions on AI-generated videos, on videos with AI avatars or digital twins and on voicebots, in German. [rtr.at](https://www.rtr.at/rtr/service/ki-servicestelle/faq/FAQ.de.html) (accessed 1 October 2026) 8. Wettbewerbszentrale, Kennzeichnung KI-generierter Inhalte, guide version 2.0, as of 29 July 2026, in German. [wettbewerbszentrale.de](https://www.wettbewerbszentrale.de/wp-content/uploads/2026/07/2026_07_Leitfaden_KI_generierte_inhalte_version2.pdf) (accessed 1 October 2026) 9. German Art Copyright Act (KUG), Sections 22 and 23, in German. [gesetze-im-internet.de](https://www.gesetze-im-internet.de/kunsturhg/__22.html) (accessed 1 October 2026) 10. German Act against Unfair Competition (UWG), Section 3(3), Section 5 and annex no. 23c, in German. [gesetze-im-internet.de](https://www.gesetze-im-internet.de/uwg_2004/anhang.html) (accessed 1 October 2026) 11. Austrian Copyright Act, Section 78 (protection of portraits), RIS, in German. [ris.bka.gv.at](https://www.ris.bka.gv.at/NormDokument.wxe?Abfrage=Bundesnormen&Gesetzesnummer=10001848&Paragraf=78) (accessed 1 October 2026) 12. Meta Help Centre (Threads), labels for AI content, section “When the AI content label is required”. [help.instagram.com](https://help.instagram.com/407718162047721/) (accessed 1 October 2026) 13. Meta Newsroom, Labeling AI-Generated Images on Facebook, Instagram and Threads (6 February 2024). [about.fb.com](https://about.fb.com/news/2024/02/labeling-ai-generated-images-on-facebook-instagram-and-threads/) (accessed 1 October 2026) 14. Meta Newsroom, Expanding GenAI Transparency for Meta’s Ads Products (3 February 2025, updated 1 June 2026). [about.fb.com](https://about.fb.com/news/2025/02/gen-ai-transparency-metas-ads-products/) (accessed 1 October 2026) 15. YouTube Help, Disclosing use of GenAI content. [support.google.com](https://support.google.com/youtube/answer/14328491) (accessed 1 October 2026) 16. YouTube Blog, Improving AI labels for viewers and creators (27 May 2026). [blog.youtube](https://blog.youtube/news-and-events/improving-ai-labels-viewers-creators/) (accessed 1 October 2026) 17. TikTok Community Guidelines, Integrity and Authenticity, section Edited Media and AI-Generated Content (effective 24 September 2026). [tiktok.com](https://www.tiktok.com/community-guidelines/en/integrity-authenticity) (accessed 1 October 2026) 18. TikTok Ads Help, TikTok Ads Policy: Misleading and false content (as of April 2026). [ads.tiktok.com](https://ads.tiktok.com/help/article/tiktok-ads-policy-misleading-and-false-content) (accessed 1 October 2026) 19. TikTok Ads Help, Add disclaimers to ads (as of September 2025). [ads.tiktok.com](https://ads.tiktok.com/help/article/add-disclaimers-to-ads) (accessed 1 October 2026) 20. TikTok Shop EU Academy, AI-generated content (AIGC), 28 August 2026, in German. [seller-de.tiktok.com](https://seller-de.tiktok.com/university/essay?knowledge_id=6752261875877653&default_language=de-DE) (accessed 1 October 2026) 21. Google Advertising Policies, Updates to AI labeling requirements (July 2026). [support.google.com](https://support.google.com/adspolicy/answer/17257106) (accessed 1 October 2026) 22. Google Ads Help, Use AI content label settings and disclosures. [support.google.com](https://support.google.com/google-ads/answer/17140115) (accessed 1 October 2026) 23. Shopify Help Center, Product media types, section Videos. [help.shopify.com](https://help.shopify.com/en/manual/products/product-media/product-media-types) (accessed 1 October 2026) 24. Shopify, GraphQL Admin API 2026-07, objects Video and VideoSource. [shopify.dev](https://shopify.dev/docs/api/admin-graphql/latest/objects/VideoSource) (accessed 1 October 2026) This article is general information, not legal advice. Whether an image needs a label is your decision; for your specific case, please ask a lawyer. Read next [02 · Obligation EU AI Act Article 50: transparency obligations, paragraph by paragraph](/en/guides/eu-ai-act-article-50/) [27 · Practice Do AI images from before 2 August 2026 need a label?](/en/guides/images-uploaded-before-august-2026/) I find the AI evidence. You decide with one click. [View on the Shopify App Store](https://apps.shopify.com/easy-ai-act-image-labels?utm_source=easyaiactlabels&utm_medium=website&utm_content=artikel)