What do the new AI requirements change in corporate communications?

When do you need to disclose the use of AI?

If a customer is interacting with an AI chatbot, it must be clearly disclosed that they are not communicating with a human employee. Users should also be informed how they can contact a company representative if needed.

Similarly, realistic AI-generated or significantly AI-manipulated images, videos, and audio content must be clearly labelled if audiences could reasonably perceive them as authentic.

For example:

• A company's CEO's voice has been generated using AI.
• An advertisement depicts an event that never actually took place.
• An AI-generated customer testimonial is presented as genuine.
• An AI-generated image of a person is presented as a real photograph.
• A real product is placed in an AI-generated environment that is presented as the result of an actual photoshoot.

It is important that AI-generated visuals do not create a false impression of the product itself for example, its size, colour, material, packaging, contents, or performance. If an AI-generated visual depicts a cosmetic effect, product feature, or situation that does not actually exist, a label alone is not sufficient. Advertising must not be misleading.

By contrast, standard image editing such as colour correction, background cleanup, or resizing does not generally constitute the creation of AI-generated content.

Can AI be used to prepare written content?

Yes. AI can be used to help prepare drafts, structure information, shorten text, or suggest headlines. However, the final version should always be reviewed by a human. This includes verifying facts, figures, sources, quotations, and claims, as well as ensuring the text aligns with the company's positioning and tone of voice.

For example, if AI states in a press release that a company is a "market leader" when this is not actually the case, simply editing the wording is not enough. Such a claim must either be substantiated with evidence or removed.

The same applies to visuals. An image may appear convincing while inaccurately representing a product, a person, an environment, or the results of using a product. Before publication, organisations should carefully assess both the legal and reputational risks.

AI-generated texts intended to inform the public about matters of public interest do not have to be labelled, provided they have been reviewed by a human, are subject to editorial oversight, and a specific individual or organisation assumes responsibility for the published content.

How should AI-generated content be labelled?

It is not sufficient to disclose the use of AI only in a file's metadata, technical description, or website terms of use. If a company publishes deepfake content, the disclosure must be clear, prominent, and easily understandable from the moment a person first sees or hears the content.

Depending on the type of content, organisations may use disclosures such as:

• "Image generated using artificial intelligence."
• "This video contains AI-generated elements."
• "Voice synthetically generated using AI."
• "Image significantly modified using AI."

For video content, the disclosure should preferably appear within the video itself, rather than only in the accompanying description. For audio content, the disclosure should be audible or otherwise made clearly and unambiguously accessible. A machine-readable label embedded by the AI tool provider alone does not relieve organisations of their obligation to disclose the use of deepfake content in a way that is visible or audible to people.

An AI-generated environment, effect, or stylisation does not automatically qualify as a deepfake. The key consideration is whether the content resembles a real or plausibly real person, place, object, or event, and whether the audience could reasonably mistake it for authentic. For this reason, placing a real product in an AI-generated environment may trigger a disclosure requirement, but this should be assessed on a case-by-case basis, taking into account the specific execution and the context in which the content is published.

Before publishing content created with the assistance of AI, organisations should ask themselves the following questions:

1. Could the audience reasonably mistake this content for authentic?
2. Have all facts, figures, quotations, and claims been verified using reliable sources?
3. Does the visual content accurately represent the product, its characteristics, and the expected results of its use?
4. Have all necessary permissions been obtained for the use of the likeness and voice of any depicted individuals, as well as any trademarks or other protected elements?
5. Which individual has reviewed and approved the material and is accountable for the final version? 

If any of the relevant questions cannot be answered with confidence, the material should not be published solely because it is technically sound or was produced quickly.

What happens if the requirements are not met?

Failure to comply with the AI transparency requirements may result in a warning, other supervisory measures, or a fine of up to €15 million or 3% of the company's total worldwide annual turnover, whichever is applicable under the rules. The specific consequences will depend on the severity of the infringement and the circumstances of the case.

However, reputational damage may arise long before any regulatory decision is made. For example, if audiences discover that what appeared to be a genuine customer testimonial, photoshoot, or CEO video was in fact AI-generated, the issue will not only be the lack of appropriate disclosure, but also the company's credibility and integrity.

AI can quickly produce a draft of a text, generate an image, or develop an initial concept. However, it does not understand the full context of a business, cannot assess all reputational implications, and does not take responsibility for published content. That is why it is essential to have a person who verifies the facts, evaluates potential risks, aligns the content with the company's positioning, and approves the final version.

Particular care should be taken with press releases, executive quotes, advertisements, campaign visuals, and social media content. Errors in such materials can cause significant damage to a company's reputation and credibility.

"I believe, the new requirements encourage organisations to focus not on how much they can save by using AI tools, but on who should be entrusted with such a critical function as communications, given its direct impact on both reputation and financial performance. External communications are a management responsibility and a long-term investment. Even the content itself and the decision whether or not to disclose the use of AI sends a message about an organisation's values, transparency, and accountability," emphasises Olga Kazaka, Partner at communications and impact agency Olsen+Partners.