From August 2, 2026, the transparency requirements of the European Union Artificial Intelligence Act (Regulation 2024/1689) have become important for Latvian companies as well. This means that if you use artificial intelligence in communication, marketing or customer service, you must also be able to show where and how AI is involved.
When to label AI content
- Article 10 sets out four situations in which transparency is required.
Communicating with a chatbot or AI assistant
If a customer is communicating with a chatbot or virtual assistant, it should be clear to them from the first interaction that they are not talking to a human. The information should be visible and understandable – not hidden in terms of use or a footer.
Exception: if it is obvious from the context that the machine is involved, a repeated warning is not necessary.
Photo: www.tet.lv indicates when communicating with a virtual assistant
Realistic AI-generated or modified image, video, or audio
If a realistic, AI-generated or significantly modified image, video or audio recording is used, that can be mistaken for a real person or event (deepfake), it should be labeled in a way that is immediately understandable to the viewer or listener – without the need to use additional tools.
In the case of video, this usually means a permanent visual cue or a clear statement at the beginning of the recording. In the case of audio, an audible warning.
The exception applies to obviously artistic, creative, satirical or fictitious works, if the context already makes their conditional nature clear.
AI-generated text on issues important to society
If AI-generated or substantially modified text is published on topics that concern politics, public administration, the judiciary, fundamental rights, health, the environment, or economic, financial, scientific, and cultural matters, it must be marked as AI-generated – unless it falls under the human review exception (see below).
Photo: Facebook.com Back and forth account photo about using AI in image creation
Emotion recognition and biometric categorization
If a company uses systems that recognize emotions or categorize people based on biometric data, the individuals affected by these systems must be informed about their operation.
When labeling is usually not necessary
- AI is used as an aid to ideas, structure, or draft, but the final text is reviewed and approved by a human with editorial responsibility.
- AI is used to correct grammar, translate, or improve standard text – this is not considered content manipulation.
- The content is obviously fantastical or impossible (e.g., a creative visual with elements that do not physically exist) and does not create a misleading impression of authenticity.
- AI outputs are not displayed to humans – for example, system-to-system communication in background processes.
What should the company do now?
- Find out where AI is already being used in the company. Marketing, customer service, internal communication, content creation – the requirements may differ in each area.
- Distinguish between content that requires labeling and content that does not. Use the criteria above, not assumptions.
- Implement internal procedures for approving AI-generated content. Determine who does it, according to what criteria, and how it is documented – this is also proof if questions arise from the supervisory authority.
- Test chatbots and AI assistants. Is it clear to the user from the first second that he is talking to a machine?
- Follows the clarifications of the Code of Practice. Technical standards for labeling AI-generated content (e.g., watermarks) are still being refined – they will influence how labeling should be implemented in practice.
Frequently asked questions
Does every text written with the help of AI have to be tagged? No. Labeling is only required if the text touches on a topic of public importance and has not undergone substantive human review with editorial responsibility.
Should AI-generated images for advertising also be labeled? Yes, if the image is realistic and could be confused with a real photograph or event. The label must be machine-readable and identifiable.
What is the penalty for not complying with the requirements? Up to 15 million euros or 31% of the company's worldwide turnover, whichever is greater.
Do the requirements also apply to small businesses? Yes. The obligation is independent of the size of the company, but supervisory authorities take into account proportionality with regard to small and medium-sized enterprises when assessing violations.
What happens to AI systems that were already operational before August 2? The transition period for them lasts until December 2, 2026 - by that time they must be brought into line with the new requirements.
AI can and should be used if it helps you work more efficiently. There is one condition – communication with the customer must remain honest and verifiable.
The article was prepared by: Mg.sc.comm. Katrīna Bartkeviča, strategic communication specialist and head of SIA SISTERING.
