01

AI with 'significant journalistic impact' requires transparency

As a guiding principle, we suggest that media consumers only need to be informed about AI when the implementation of such tooling has had a significant journalistic impact. The definition of what "significant journalistic impact" means (and thus, the decision about the need for transparency) needs to be adapted to the specific needs and circumstances of each publication.

We do not find it suitable or desirable to define a fixed limit or type of AI-impact, but instead argue that the responsibility for evaluation should be placed in the editorial process.

02

Other internal AI-tooling does not require transparency

Provided that the above principle is adhered to, we argue that public information is not needed for internal applications where AI is used to support editorial or commercial work. For such use cases, internal AI policies apply.

03

AI transparency must be approached as an iterative theme

Many of the AI technologies now used in the media industry are at an early stage, during which it is necessary to be open and clear about their usage. This openness will need to be reassessed over time as technology and user habits and expectations evolve.

We are positive towards continued industry dialogue, and also see a great need for media companies to have an ongoing internal dialogue on the topic. We intend to follow up on our recommendations in six months, to evaluate their continued relevance.

04

Be specific about the type of AI tool applied

To counter audience perceptions of AI being an autonomous force operating without editorial control, we encourage media companies to be specific when telling our audiences what form of AI we used. This can serve an educational function for both the media companies and the media audience, where we demystify the concept of AI and instead communicate about the technology as a tool. 

05

Share information in connection with consumed content

Where AI transparency is needed (see “Principle recommendations”), information should be shared in connection with the AI implementation. By sharing information in connection with consumed content, we give Swedish media consumers a chance to understand for themselves how much (or little) AI is used with a significant impact on journalism.

06

Harmonise the industry's language around generative AI

We recommend a harmonised approach for describing AI within and between Swedish media companies. This is mainly motivated by the fact that media consumers can find it easier to understand and relate to AI in Swedish media if we communicate similarly about use. As a first step, we recommend a harmonized use of language regarding generative AI.

​

As a basis, we recommend the wording "with support of". This is to signal the actual influence of AI on the content, and remind the media consumer that there is an editorial process (and employees) that produced the content.

07

Avoid visual labels (icons) for AI in editorial media

We advise against standardized visual labeling (e.g. an icon) for AI in editorial media. An explicit visual differentiation can give the impression that the credibility of AI-generated/influenced content is different from content created by human journalists or editors, and it can be difficult to implement consistently across different media platforms and content types.