Do I need to disclose that an image was made with AI? Only when it looks real, or when text informs on a matter of public interest: what the AI Act actually asks of marketers.
Since 2 August 2026 you will read that everything made with AI has to be labelled, and the text of the AI Act says something else. Here is what Article 50 requires, the marketing cases one by one and a six-question checklist, in Word and PDF.
There is no general duty to write “made with AI” on every piece of marketing. Article 50 of the AI Act, applicable from 2 August 2026, requires anyone using AI professionally to disclose it in two cases: deepfakes, and text informing the public on matters of public interest without human review.
The European Commission’s guidelines on Article 50 go into marketing cases. Ad copy and product descriptions stay outside, unless they make claims about health, safety or sustainability; an AI image that makes the product look better than it is falls inside.
Above the AI Act, the usual rules still apply: misleading advertising, national deepfake offences such as Italy’s Law 132/2025, and platform labels.
The six-question checklist, with the cases and a log, is below, in Word and PDF, with no email required.
Written by: Mattia Esposito, AI and marketing consultant, founder of Itria AI. From Puglia, based in Turin since 2023, he works remotely in Italy and abroad. LinkedIn profile.
Disclosing AI-generated ads: the six-question checklist to download
Six questions to ask before publishing an ad, a post, a product page or a video made with AI, each with the article that answers it. The file also has the most common cases, sample labels and a content log to keep up to date. You don’t need to know the regulation: you need to know how the content was made.

| File | Contents | Link |
|---|---|---|
| Wordto fill in, in Italian | The six questions with the rule behind each, marketing cases, sample labels, the log. | |
| PDFto print, in Italian | The same checklist, to keep next to the people who prepare the content. |
| Question | How to answer | The rule |
|---|---|---|
| 1. Who used the AI? | The party that decides on and controls the use of AI: the company, or the agency if the company only commissioned the work. | Art. 3(4) |
| 2. What is it? | Image, video or audio; text; chatbot. Each has its own rule. | Art. 50(1) and (4) |
| 3. Does it look real? | If it resembles real people, products, places or events and could appear authentic, it is a deepfake. | Art. 3(60) |
| 4. Public interest? | Text on health, safety or finance, published without a human review someone answers for, must be disclosed. | Art. 50(4) |
| 5. Creative work? | For an evidently creative or satirical work, a disclosure that does not spoil it is enough. | Art. 50(4) |
| 6. Always | Consent from anyone portrayed, the product shown as it is, the platform’s AI setting, files kept. | Consumer law; criminal law |
In October I counted how many times Spotify writes “AI” in its five press releases about its DJ, the feature that builds playlists and talks over them with a voice: 8, 9, 2, 1, 0. For the Italian voice they held a casting. In the post where I told the story I put it like this: “Spotify doesn’t even call its AI AI.”
How much you say about AI, in marketing, is first of all a positioning choice. Since 2 August 2026 part of that choice is made by the law, and the part is narrower than most articles suggest. In August, preparing Itria’s first Meta campaign, I read around a hundred other people’s ads: how to do it well is in the guide to the Meta Ad Library.
AI Act Article 50 in plain words: what it requires
Article 50 of Regulation (EU) 2024/1689 splits the transparency duties between providers of AI systems and the businesses using them. Providers, such as a company making an image generator, must tell people they are talking to an AI (paragraph 1) and mark synthetic output in a machine-readable format (paragraph 2).
For anyone using AI professionally, the deployer, paragraph 4 sets two duties. For deepfakes, deployers “shall disclose that the content has been artificially generated or manipulated”. The same applies to text published with the purpose of informing the public on matters of public interest, unless it has gone through human review or editorial control by a person who holds editorial responsibility.
Paragraph 5 says how: the information must be given “in a clear and distinguishable manner at the latest at the time of the first interaction or exposure”. The regulation does not prescribe any wording. Under Article 3(4) a deployer is anyone using an AI system under its authority, “except where the AI system is used in the course of a personal non-professional activity”: a small business generating an image for an ad is one.
When it applies: 2 August 2026, with no delay for deepfakes
Article 113 of the AI Act sets general application at 2 August 2026, and Article 50 is not among the exceptions: it applies from that day. Regulation (EU) 2026/1744, the Digital Omnibus on AI published in the Official Journal of the EU on 24 July 2026, delays the high-risk rules but not Article 50.
The delay people talk about concerns something else. The new Article 111(4) gives providers of generative systems placed on the market before 2 August 2026 until 2 December 2026 to comply with the marking duty in paragraph 2. The deployer’s duty on deepfakes has no transition period.
In its questions and answers on Article 50, updated on 24 July 2026, the Commission adds: “Content generated prior to 2 August 2026 does not need to be labelled retroactively.”
Ads and product descriptions: when no label is needed
The Commission’s Article 50 guidelines leave out of the text duty any AI-manipulated text forming part of a company’s advertisement or product descriptions, as long as it makes no claims about health, consumer safety or sustainability. The guidelines are not binding, but they are the Commission’s reading.
Their examples of text within scope include articles on a lifestyle site about the effects of diets and company reports with information for investors. The human review that removes the duty has to be substantive: spelling or grammar correction alone, the Commission writes, is not enough. How to write ad copy that holds up is covered in the guide to hooks and headlines.
Deepfakes: when an advertising image becomes one
Under Article 3(60), 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 or truthful”. The Commission reads three criteria together: resemblance, existence (even if only plausible) and a false appearance of authenticity, judged against the context and the expected audience.
| Marketing case | Does the AI Act require a label? | Source |
|---|---|---|
| Real product on an AI background | No, if the ad does not mislead about the product. | Guidelines |
| Product better-looking than in reality | Yes, and it risks being misleading advertising. | Guidelines; consumer law |
| Talking animals in an ad | No. | Guidelines |
| Famous influencer generated with AI | Yes; without consent and causing harm, also a crime in Italy. | Guidelines; Italian Criminal Code, art. 612-quater |
| Realistic synthetic influencer | Yes, and the creative exception does not apply. | Guidelines |
| Realistic avatar of the CEO | Yes. | Guidelines |
| Colour, extended background, resizing | No, a minor edit. | Guidelines |
| Product description written with AI | No, unless it makes health, safety or sustainability claims. | Guidelines |
| Health article without review | Yes. | Art. 50(4) |
| Post published before 2 August 2026 | No, no retroactive labelling. | Commission QA |
On the creative exception the Commission is strict: an ad can be an evidently creative work in some situations, but purely informative or commercial content recognisable as such stays outside. Among its examples of non-creative content it lists a teleshopping-style video with simulated consumers and a realistic synthetic influencer trying out a sponsored product.
If the agency makes it, who is the deployer?
The guidelines say that a company which merely commissions an ad from an agency, without deciding or controlling whether and how the agency uses AI, is not a deployer: the agency is. Employees are not separate deployers, and the company remains one even when it works with freelancers under its control.
For a small business the practical question is who chooses the tool and who approves the images. If the company does, the duty is its own. How to divide it in a contract with an agency or a freelancer is a decision to take with a lawyer.
How to label AI-generated content: visible, at first exposure
The invisible watermark the tool puts in the file is not enough: the Commission writes that deployers cannot rely on the machine-readable marking embedded in the content by the provider. You need a visible label on images and video and an audible notice on audio, at the first contact with the content.
The Code of Practice on marking and labelling AI-generated content, published by the Commission on 10 June 2026, is voluntary and has a section for deployers. The Commission stresses that signing up is voluntary while the Article 50 requirements are legal obligations, and that the EU has created a set of icons deployers may use to label their content.
A label, the guidelines add, does not make a misleading or unlawful deepfake lawful, nor does it override privacy, trade marks, copyright and rights over a person’s image and voice. Those are the subject of the guide on AI images, advertising and copyright.
Italy’s Law 132/2025 and the deepfake offence
Italian Law no. 132 of 23 September 2025, in force since 10 October 2025, adds no labelling duty for marketing content: article 1 says it must be applied in line with the AI Act. It does add a crime. Article 26 introduces article 612-quater into the Criminal Code, on the unlawful spreading of content generated or altered with AI.
In short: anyone who causes unjust harm to a person by handing over, publishing or otherwise spreading, without that person’s consent, images, video or voices faked with AI and capable of misleading faces one to five years in prison.
Legislative Decree no. 160 of 9 September 2026 added this offence to the list for which the company itself is liable, with a fine of 200 to 700 quotas. A real person’s face in a campaign needs their consent, label or no label.
AI Act fines: what a small business risks
Article 99 of the AI Act sets fines of up to EUR 15 million or 3% of worldwide annual turnover for breaching Article 50, and paragraph 6 says that for SMEs the lower of the two applies. Member States set penalties taking the interests of SMEs into account.
In Italy, Legislative Decree no. 179 of 7 October 2026, in force since 23 October, names the National Cybersecurity Agency as market surveillance authority for AI and adopts the same ceilings for Article 50, unless the conduct is a crime, with the lower-amount rule for SMEs and innovative start-ups.
For minor breaches the authority can order the infringement to be removed, or a public statement, instead of a fine. Misleading advertising stays with the Italian competition authority under the Consumer Code.
How to label AI-generated content on Google, YouTube, TikTok, Facebook and Instagram
Platforms have their own rules, which add to the AI Act and do not replace it. Google says so about its own: “Use of the AI label setting in Google’s advertising products doesn’t guarantee compliance with specific regulations.”
| Platform | What it requires or allows | Source and date |
|---|---|---|
| Google Ads | AI labels inside image and video ads, an “AI label” setting, automatic labels on assets made with Google’s own AI tools. | Google, 9 July 2026 |
| Google, election ads | Mandatory disclosure of altered or synthetic content showing real people or events. | Google, July 2024 |
| YouTube | Disclose realistic content with the “AI use” setting: real people saying things they never said, events that never happened. Not minor edits. | YouTube Help |
| TikTok | Label AI-generated or heavily edited content showing realistic scenes or people; otherwise removal, restriction or a label added by TikTok. | TikTok, in force from 24 September 2026 |
| Facebook and Instagram | An “AI info” label when Meta detects industry-standard signals or the user discloses it; for content only edited with AI, in the post menu. | Meta, 2024 |
In Italy, influencers above AGCOM’s thresholds also follow its code of conduct (decision 197/25/CONS, annex B): anyone editing content with filters or software in a way that suggests product effects that don’t match reality has to flag it with wording such as “edited photo” or “edited video”. The text read for this page does not mention artificial intelligence. The Italian Advertising Self-Regulation Code has no article on AI, but requires testimonials to be genuine and responsible.
UK and US: no blanket rule on disclosing AI in ads
In the UK, the Advertising Standards Authority writes: “There is no blanket legal requirement in the UK to disclose the use of AI in ads”. Existing rules apply, and a disclosure does not rescue a misleading message. In June 2026 the ASA added that an AI-generated celebrity endorsement realistic enough to mislead is likely to breach the Code.
| Where | Duty to disclose AI in ads | Rule and date |
|---|---|---|
| European Union | For deepfakes and public-interest text without review; no general duty. | AI Act, Art. 50, from 2 August 2026 |
| Italy | The same as the AI Act; on top, a criminal offence for spreading deepfakes and liability for the company. | Law 132/2025, from 10 October 2025; Legislative Decree 179/2026 |
| United Kingdom | No general duty; misleading ads are banned, disclosed or not. | ASA, 29 May 2025 and 11 June 2026 |
| United States | No federal duty; testimonials that misrepresent the person or their experience are banned, including generated ones. | FTC, 16 CFR 465.2 |
Google, on its July 2026 page, lists the EU, India and New York among the rules it follows. If you sell in the US, state rules need reading one by one. How ads work inside assistants is in the guide to advertising on ChatGPT, and what Meta does with chats is in Meta AI and ads.
A line next to the image. On my Instagram art page, @artartism, the same artwork with a sentence next to it got 2,300 likes against 1,000, 2.3 times as many. A description taken from Wikipedia did 1.5 times, a sentence of my own 2.5. A label is also a line next to an image: the post.
Five press releases. Spotify writes “AI” 8, 9, 2, 1 and 0 times across the five press releases on its DJ, from the first to the latest.
No advertising. The same page earned 632,288 organic likes in 7 months and 10 million views in one month, with zero spent on ads.
In practice: keep a log
The log is the part people forget. For each piece of content made with AI, note the channel, the tool, whether it is a deepfake and why, the label and where it sits, anyone portrayed and their consent, who reviewed it and the date. If an authority or a platform asks, the answer is there.
The work on a company’s copy and materials, ads included, is described on the page about sales copy and materials. What a campaign costs is in how much Meta Ads cost, and the other guides are in the index of marketing for small businesses.
Questions and answers
Do I need to disclose that an image was made with AI?
Only in some cases. Article 50(4) of the AI Act requires anyone using AI professionally to disclose deepfakes: images, audio or video that resemble real people, objects, places or events and would appear authentic. An obvious fantasy image, or a real product on a generated background that does not mislead, carries no such duty.
Misleading advertising rules and platform rules still apply.
When does the AI Act labelling duty apply?
From 2 August 2026, under Article 113 of the regulation. Regulation (EU) 2026/1744 did not postpone it: it gives providers of generative systems already on the market until 2 December 2026 for machine-readable marking, and nothing more. The Commission writes that content generated before 2 August 2026 does not need to be labelled retroactively.
Do I need to disclose ad copy written with ChatGPT?
As a rule, no. The duty on text covers text published to inform the public on matters of public interest, and it falls away if a person reviews it and takes editorial responsibility. The Commission’s guidelines leave out advertising copy and product descriptions that make no claims about health, consumer safety or sustainability.
What does a small business risk if it fails to label a deepfake?
Article 99 of the AI Act sets fines of up to EUR 15 million or 3% of worldwide annual turnover for breaching Article 50, and for SMEs the lower of the two applies. In Italy, Legislative Decree 179/2026 adopts the same ceilings and, for minor breaches, lets the authority order the infringement to be removed instead of a fine. Spreading a person’s faked face or voice without consent, causing harm, is also a criminal offence under article 612-quater of the Italian Criminal Code.
If an agency makes the ad, who has to add the label?
The company or agency that decides on and controls the use of AI. The Commission’s guidelines say that a company which merely commissions an advertising agency, without deciding or controlling whether and how the agency uses AI, is not a deployer: the agency is. If the company chooses the tool and approves the generated images, the duty is back with it. The contract with the agency should reflect this, drafted with a lawyer.
Notes on sources
- Regulation (EU) 2024/1689, AI Act, Articles 3(4) and (60), 50, 99 and 113; Regulation (EU) 2026/1744, Digital Omnibus on AI, OJ of 24 July 2026, new Article 111(4). Reuse permitted under Decision 2011/833/EU. Read on 10 November 2026.
- European Commission, guidelines on the Article 50 transparency obligations, not binding: PDF version downloaded from the official page on 10 November 2026. The marketing examples and the agency’s role come from there.
- European Commission, Transparency obligations under Article 50, updated on 24 July 2026, and Code of Practice on AI-generated content, page updated on 19 October 2026.
- Italian Law no. 132 of 23 September 2025, Official Gazette no. 223 of 25 September 2025, articles 1 and 26 (art. 612-quater of the Criminal Code). Legislative Decree no. 179 of 7 October 2026, Official Gazette no. 234 of 8 October 2026, articles 5, 22 and 23. Legislative Decree no. 160 of 9 September 2026, Official Gazette no. 214 of 15 September 2026, article 15. Italian texts are paraphrased in English.
- Italian Consumer Code, article 21; IAP, Advertising Self-Regulation Code, 71st edition, articles 2, 4 and 13; AGCOM, influencer code of conduct, decision 197/25/CONS, annex B, § 3.3.
- Google, Updates to AI labeling requirements, 9 July 2026, and altered or synthetic content in election ads; YouTube, Disclosing use of GenAI content; TikTok, Community Guidelines, in force from 24 September 2026; Meta, approach to labelling, updated on 12 September 2024. Read on 10 November 2026.
- ASA, Disclosure of AI in advertising, 29 May 2025, and AI and deepfakes, 11 June 2026. FTC, 16 CFR Part 465, § 465.2.
- The @artartism figures and the count of Spotify’s press releases come from Mattia Esposito’s posts of 25 September and 13 October 2026.
A label is one line. The copy that sells is everything else. Positioning starts from the market.
Mattia Esposito, who wrote this guide, also works with Itria on positioning: in-depth market analysis to find where competitors are exposed and take that ground. He did it on his own site first. In August, asked who had founded itria.io, ChatGPT couldn’t find him. From 11 October to 7 November 2026 itria.io pages appeared 13,803 times in Google and Bing results and 3,818 times in AI answers: Google’s AI Overviews and AI Mode, and Microsoft Copilot.