Does AI-generated content need to be labelled from 2 August 2026? Rules for businesses and advertising

From 2 August 2026, the transparency requirements laid down in Article 50 of the European Union's Artificial Intelligence Regulation, or AI Act, will apply. These concern certain artificial intelligence systems and content created or modified using AI.

However, this does not mean that the company will have to add the label “created using artificial intelligence” to every photo, advertising text, or social media post from now on.

The labelling requirement depends on the type of content, the extent of the changes made, the purpose of publication and whether a person may consider material created using artificial intelligence to be authentic.

Four areas are particularly important for companies:

chatbots and other AI systems that communicate directly with customers;
AI images, videos and audio recordings that may be considered deepfakes;
texts addressing matters of public interest created using AI;
emotion recognition and biometric categorisation systems.

Article 50 applies from 2 August 2026. The European Commission has also drawn up a voluntary code of conduct for labelling content created using AI. Compliance with the code of conduct is voluntary, but the obligations arising from the Artificial Intelligence Regulation are mandatory.

Does all content created using AI need to be labelled?

No. A visible label does not automatically need to be added to all content created using artificial intelligence.

For companies, the main disclosure requirements concern the following cases:

images, audio and videos that may be regarded as deepfakes;
text created or substantially modified using AI, published for the purpose of informing the public about a matter of public interest, and which has not undergone human review or editing;
AI systems that interact directly with people;
emotion recognition or biometric categorisation systems.

It should also be taken into account that the rules of advertising platforms and social networks may be stricter than the requirements of the AI Act.

Seega ei tähenda märgistamiskohustuse puudumine tehisintellektimääruse alusel automaatselt seda, et märget ei nõua näiteks Meta, Google, TikTok või mõni teine platvorm.

Who is responsible: the developer of the AI service or the company using it?

The Artificial Intelligence Act distinguishes between the provider and the user, or deployer, of an AI system.

a provider of the AI system is a company or organisation that develops the AI system or places it on the market under its own name or trademark. This may be, for example, a developer of an image, text or video generator.

a user or implementer of the AI system is a company, entrepreneur or organisation that uses the AI system in its professional activities.

A small business that uses a ready-made AI service to create advertisements, photos or texts is usually the user of the system, not its provider.

The parties have different obligations.

Providers of generative AI systems must, to the extent possible, ensure that generated images, videos, audio files and texts are labelled in a machine-readable format and are technically identifiable as content generated or altered using artificial intelligence.

However, a company using an AI tool must above all ensure disclosure that is understandable to people when publishing a deepfake or certain AI text concerning matters of public interest.

In other words: technical labelling embedded in the file and a notice visible to the customer are not the same thing.

Do product photos created using AI need to be labelled?

There is no single answer. It all depends on what is shown in the image and how extensively the original material has been altered.

When is labelling generally not required?

A visible label is generally not necessary if artificial intelligence was used, for example, to:

to remove the background;
to resize an image;
to adjust light, contrast or sharpness;
to remove a minor flaw from a photo;
to create a clearly decorative or fantasy-based background;
a modification that does not affect the product's properties or alter its meaning.

Tehisintellektimääruses on arvestatud tavapäraste redigeerimisvõtetega ja olukordadega, kus AI-süsteem ei muuda oluliselt sisendandmeid ega nende tähendust.

However, this exception must not be used to justify misleading product images.

When should an image be labelled?

Labelling is necessary, or at least legally safer, when artificial intelligence:

adds features or details to the product that it does not actually have;
places a real product in an environment presented as a real interior, shop or location;
creates a non-existent result that cannot actually be achieved by using the product;
shows a AI model who may appear to be a real customer or employee;
creates a non-existent hotel room, apartment, restaurant, dish or place where a service is provided;
modifies a real photograph so that the consumer may consider the result authentic.

The concept of a deepfake under the AI Act may not be limited to people's faces and voices.

In certain circumstances, content created or modified using AI that resembles an existing object, place, organisation or event and gives the false impression that it is authentic material may also be considered a deepfake.

The European Commission has cited the example of actual photos of an empty apartment to which furniture is added using artificial intelligence. Such an image may require an explanation that the furniture was added using AI and is not part of the actual offer.

Labelling does not make misleading advertising permissible

The label “image created using artificial intelligence” does not give a company the right to advertise non-existent product characteristics.

If an AI image shows a product, space or service result as better, larger or more functional than it actually is, the advertisement may violate consumer protection and advertising requirements.

In the European Union, advertising that contains false information or, due to its overall presentation, may mislead the average consumer about the characteristics of a product or service is prohibited. The same principles also apply in Estonia.

For example:

a restaurant should not publish a food photo generated using AI that differs significantly from the dish actually served;
a hotel should not show a non-existent view from a hotel room window;
a beauty salon should not present a completely artificial procedure result as a photo of a real customer;
a construction company should not show a completed project that does not actually exist.

These articles may also be of interest

EL simplifies artificial intelligence rules: what this means for businesses

Which requirements of the AI Act may affect companies, which rules will be postponed, and what is a company using AI already responsible for?

Digital Fairness Act: combating dark patterns and unfair marketing

See how the European Union aims to restrict misleading web design, covert advertising, manipulative sales practices and unfair influencing of consumers.

Optimising a website for AI search: AEO, GEO and LLMO

A practical guide for businesses that want their website, articles and services to be more discoverable in ChatGPT, Google AI answers and other AI searches.

How to recognize telephone scammers in 2026: AI voices and spoofed bank numbers

How is artificial intelligence used to clone voices, create convincing conversations and defraud companies, and which verification rules help reduce the risks?

Do advertising images created using AI need to be labelled?

An advertising image does not automatically require labelling merely because it was created using an AI tool.

The main question is whether the material could give the impression that it is an authentic depiction of a real person, product, place, company or event.

Examples where labelling is generally not required

an abstract advertising background;
a clearly fantasy-based illustration;
an illustrated or unrealistic character;
decorative graphics;
an image that does not give the impression of a documentary photograph;
a design that uses a genuine and unaltered product photo.

Examples where labelling is necessary or strongly recommended

the AI person is presented as a real employee of the company;
a fictional “satisfied customer” is used in advertising;
a real person is made to appear to say words that they did not actually say;
AI is used to create a non-existent office, restaurant, hotel, or service location;
the impression is created that a well-known person recommends the product;
an existing object is altered in a way that may give the customer a false impression of its actual condition.

In borderline situations, it is advisable to use a brief and clear explanation. This reduces the risk of both violating the AI Act and engaging in misleading advertising.

Do AI videos and AI voices need to be labeled?

Videos, cloned voices and realistic digital characters are among the higher-risk types of content.

The use of artificial intelligence must be disclosed if the content resembles a real person, place, object or event and could be mistakenly perceived as authentic.

This may concern, for example:

a cloned voice;
face replacement;
a digital replica of a real person;
a fabricated appearance by a company executive;
a fake customer review;
a video of an event that did not actually take place;
a product demonstration created using AI that could be considered a real recording.

In such cases, information about the use of AI must be made available to the person no later than when they first view the content.

If the work is clearly artistic, satirical, creative or fictional, the method of disclosure may be less intrusive. However, this does not automatically mean that disclosure can be omitted entirely.

Does the customer need to be informed that they are interacting with a chatbot?

Yes. If a customer interacts with an AI system on a website, in an online store, in an app or through a messaging app, they must understand that there is no real person on the other end.

An exception may apply if the use of AI is obvious to a reasonably attentive user based on the context and interface. However, relying on this exception is risky for a company.

Suitable disclosures may include, for example:

You are interacting with a AI assistant.

This is an automated assistant based on artificial intelligence.

The response was prepared by a AI assistant. If necessary, you can contact a member of our staff.

The notice must be provided before the conversation begins or, at the latest, during the first interaction.

The robot's name alone, such as “EasyBot”, may not be sufficient if the user interface and communication style give the impression that a real employee is responding.

Does text created with the help of ChatGPT or another AI need to be labelled?

Not all texts created using AI need to be labelled.

For texts, the requirement in Article 50 is narrower. Disclosure may be necessary if the following conditions are met simultaneously:

the text was created or significantly modified with the help of AI;
it is published for the purpose of informing the public;
the text addresses a matter of public interest;
the text has not undergone human review or editing;
no person or organisation has assumed editorial responsibility for the published content.

If a person has reviewed and edited the text and the company assumes responsibility for its publication, the obligation to separately disclose the text's AI origin generally does not apply under Article 50.

Usually, labeling is not required for

product descriptions;
advertising slogans;
price quotes;
customer letters;
ordinary social media posts;
SEO texts introducing the company’s services;
drafts created using AI that an employee has reviewed and edited;
translations that have been reviewed by a person;
texts for which AI was used only to correct grammar.

When is greater attention required?

Topics of public interest may include, for example:

laws and taxes;
public services;
economy and finance;
healthcare;
public safety;
environment;
consumer rights;
significant events affecting society.

If a company publishes articles on such topics, the workflow should not be limited to automatically generating text and pressing the publish button.

Fact-checking, editing and a specific person responsible for the final content are required.

A superficial review of the text without checking dates, figures, sources and conclusions may not be sufficient to demonstrate editorial oversight.

Must a company’s expert articles be labelled?

If AI helped draft an expert article, but a specialist in the field or an editor:

fact-checked;
corrected errors;
substantively revised the content;
added professional explanations;
verified the sources used;
assumed responsibility for the final text,

then, under Article 50, a separate label stating “the article was created using AI” is generally not required.

At the same time, AI must not be used to imitate authorship or professional expertise.

The following is prohibited or, at the very least, highly risky:

use the name of a fictitious expert;
attribute the text to a real specialist without their consent;
give the impression that the text has been reviewed by a doctor, lawyer, accountant, or other specialist when it has not;
publish automatically generated advice under the guise of a verified expert opinion.

Human oversight is particularly important for legal, health and financial materials because incorrect information can cause real harm.

Do social media posts need to be labelled?

The rules do not depend on whether the material is published on Instagram, Facebook, TikTok or the company's website. What matters is the nature of the content and the purpose of its publication.

A standard advertising post created using AI and reviewed by a person does not automatically require labelling under Article 50.

However, a label may be necessary if the post contains:

a realistic AI video;
a cloned voice;
the face or digital replica of a real person;
a fictional customer review;
a non-existent event;
an altered image of a real object;
unverified AI text on a matter of public interest.

The social media platform's own requirements must also be taken into account. The platform may require the “created using artificial intelligence” option to be enabled even if the AI Act does not directly require it in that particular situation.

How should AI content be labelled correctly?

The information specified in Article 50 must be:

clear;
distinguishable from other content;
understandable to a person;
presented no later than upon first contact or when the content is first displayed;
accessible also to users with disabilities.

Information should not be hidden:

into the website's general terms and conditions;
at the end of a long description;
among a large number of hashtags;
in very small or poorly visible text;
only in the file metadata;
on a separate page that the user is unlikely to open.

Suitable wording may include, for example:

The image was created using artificial intelligence.

The image has been partially modified using AI.

The video contains an image and voice created using AI.

The voice used in the video was created using artificial intelligence.

This is an illustrative visualization, not a photograph of the actual object.

The text was created with the help of AI and has not undergone editorial review.

When advertising a specific product or service, a general statement may be too vague. It is more useful to specify which part of the material is artificial.

For example:

The image background was created using AI. The product photo has not been modified.

The furniture visible in the room was added using AI and is not included in the offer.

The depicted result is a AI visualisation showing a possible interior design.

This is not a photograph of a completed project, but an example solution created using artificial intelligence.

Is the official European Union AI label required?

No. The European Union has developed separate labels for content created entirely using AI and content partially modified using AI, but their use is voluntary.

Even the use of an official label does not in itself guarantee compliance with the requirements.

The company is responsible for ensuring that the notice is:

understandable;
sufficiently visible;
specific to the material in question;
published at the appropriate time.

For a small business, a simple text explanation may be easier for the customer to understand than an unfamiliar symbol.

Do old AI posts need to be labelled retroactively?

According to the European Commission's clarification, AI content and deepfakes created and already published before 2 August 2026 generally do not need to be labelled retroactively.

However, if old material is republished, added to a new advertising campaign or significantly altered, it is worth reassessing whether disclosure requirements may apply to it.

An old publication date should not be treated as an indefinite exemption if the same material is used in a new commercial context after 2 August 2026.

Practical table for companies

How AI is used	Is visible labelling required?
Correcting the grammar of advertising copy	Usually not
AI prepared the draft, and an employee reviewed and edited it	Usually not
The company reviewed and approved the product description	Usually not
Decorative fantasy background	Usually not
Removing the background without changing the product	Usually not
AI added furniture to a photo of an empty apartment	Required or strongly recommended
A non-existent hotel room or service area was created	Necessary if the image could be considered real
AI client gives a positive review of the product	Very high risk; labelling alone may not make the practice permissible
A real person's voice was cloned	Required
A real person is made to appear to say fabricated words	Required
A AI chatbot responds on the website	The user must be informed
AI wrote a tax-related news article without human review	Disclosure may be mandatory
The editor reviewed the tax-related article, and the company is responsible for the text	A separate notice is usually not required
AI was used for the translation, and a human reviewed it	Usually not

Important: the table provides a general overview. The need for labelling always depends on how the AI content is used, whether it can be considered authentic, and whether it may influence the decisions of customers or the public.

What should a company do before 2 August 2026?

1. Compile a list of the AI tools used

Check whether the company uses:

chatbots;
image generators;
video editing AI tools;
voice generators;
text robots;
automated content publications;
systems for analysing faces, emotions, or behaviour.

2. Distinguish between technical and visible labelling

Ask the AI service provider:

whether the service adds a machine-readable label to the file;
whether the metadata is retained after the file is exported;
whether the service has a built-in AI content disclosure feature;
what guidance the service provider gives to commercial users.

Technical labels and metadata should not be deliberately removed without a compelling reason.

3. Identify potential deepfakes

For each realistic piece of material, ask:

whether a person might consider it a real photo or video;
whether it depicts an existing person, object, place or event;
whether the material may influence the customer's purchasing decision;
whether a change made using AI could create a false impression of the product or service.

If the answer is yes, use a visible explanation.

4. Establish editorial review of texts

For articles, news items, and expert materials, it is advisable to document:

who reviewed the text;
when the review was carried out;
which sources were used;
who is responsible for the final content;
what significant improvements were made.

5. Prepare standard labels

Prepare several ready-made phrasings for use within the company:

images created entirely using AI;
partially modified photos;
for AI videos;
to the cloned voice;
for chatbots;
for unverified automatically generated texts.

6. Verify the truthfulness of the advertisement

Even correctly labelled AI material must not:

exaggerate the product's features;
show a result that the company cannot actually provide;
present fabricated reviews as the opinions of real customers;
create the impression that a well-known person endorses the product;
presenting a visualisation as a real photograph;
conceal important information about the actual terms of the offer.

What fines may be imposed?

In certain cases, violations of the transparency requirements of the Artificial Intelligence Act may result in administrative fines of up to 15 million euros or up to 3% of the company's worldwide annual turnover.

For small and medium-sized enterprises, the AI Act provides for a more lenient principle: as a rule, the lower rather than the higher amount of the applicable maximum limits is applied.

When imposing a fine, the following must also be taken into account:

- the nature of the infringement;
- its severity and duration;
- the size of the company;

damage incurred;
intent or negligence;
the company's actions to remedy the violation.

This means that the maximum fine is not automatically imposed for every missing label or disputed advertising image.

In addition, an infringement may have consequences under the rules on advertising, consumer protection, personal data protection, copyright or the use of a person's image.

Frequently asked questions

Does every AI post need to include the hashtag #AI?

No. The AI Act does not impose a general obligation to add the hashtag #AI to every piece of material.

When disclosure is mandatory, the information must be clear and visible. A single vague hashtag may not be sufficient.

Does an image created with Midjourney or another generator always need to be labeled?

No. The need for visible labelling does not depend solely on the tool used.

What matters is whether the image can be considered an authentic depiction of a real person, object, place or event and whether it creates a false impression of reality.

Does text written using ChatGPT need to be labeled?

Usually not, if the company has reviewed and edited the text and takes responsibility for its publication.

Disclosure may be mandatory for unreviewed, automatically published text concerning a matter of public interest.

Is it sufficient to mention the use of AI in the privacy policy?

No, if the user is informed when the specific material is first displayed or at the beginning of their interaction with the AI system.

A general statement in the privacy policy does not replace a visible notice in the chat window, next to the video, or on the advertising image.

Does a AI avatar need to be labelled?

If the avatar is clearly a fictional, drawn or unrealistic character, it may not be a deepfake.

If a realistic avatar is presented as a real employee, expert, or customer, the risk is significantly greater and its artificial nature should be disclosed.

Do the requirements also apply to small businesses and individual professionals?

Yes. There is no general exemption for the professional use of AI simply because a company is small.

The size of the company may nevertheless be taken into account when determining supervisory measures and potential fines.

Summary

From 2 August 2026, companies do not have to label absolutely all texts, banners, photos, and posts created using AI.

The most attention should be paid to:

chatbots that interact with customers;
realistic AI photos, videos, and voices;
deepfakes of people, objects, places and events;
unreviewed AI texts on topics of public interest;
advertisements that may create a false impression of a product or service.

It is sensible for a company to follow a simple principle:

If content created or modified using AI may be perceived as authentic and may influence the customer's decision, the use of AI should be disclosed clearly and immediately.

However, labeling is no substitute for truthful advertising. Artificial intelligence must not be used to show non-existent product features, fabricated reviews, or results that the company cannot actually deliver.

The article was prepared based on information available in July 2026. The European Commission's final guidelines may further clarify some borderline situations. This material is for informational purposes and does not replace legal advice regarding a specific case.

Showcase your services and company on Easyfind

Easyfind helps companies and independent professionals become more visible online and easier for potential clients to find.

Each listing serves as a small website for the company. A listing can include:

a detailed description of the service or product;
photos and video;
a PDF file, such as a price list or documents;
company details;
contact options;
customer reviews and ratings.

Special offers, discounts and seasonal services can also be published on Easyfind. Separate expert articles help demonstrate the company's expertise, increase trust and create useful content that may also appear in search engines.

Explore the options and choose a suitable package: easyfind.ee/hinnakiri/