What actually changes from 2 August 2026?

Article 50 of the AI Act requires that AI-generated or substantially altered image, audio, and video content must be machine-readably identifiable as such – specifically when it appears realistic, meaning it shows or fakes real people, places, or events. The Austrian Federal Economic Chamber clarifies this for domestic businesses: Labelling requirements apply primarily to “deceptively authentic” content and to texts on public-interest topics that are published automatically without human review. Mere production assistance – editing, colour grading, text suggestions, hashtag generation – does not fall under this.

In practical terms, this means that a social media post reviewed and accounted for by a human often doesn’t fall under the legal obligation at all. However, the platforms themselves have long been stricter than the law. Meta, TikTok, and Pinterest now automatically detect AI content via metadata – regardless of whether a legal obligation exists or not. Anyone who only aligns with the legal minimum will still get caught by the platform.

Who really needs to take note – solo creators or corporate compliance teams?

Both, just for different reasons. Anyone posting daily as a sole proprietor or small team has long been using AI for image ideas, captions, or initial drafts – and often doesn’t even realise that Meta now actively cites undeclared AI content as a reason for ad rejection. The good news according to the WKO: Most everyday, human-reviewed social media content does not trigger a legal labelling obligation. The bad news: This does not automatically translate to platform rules, which are often stricter.

For larger companies and brands, the risk scales. Multiple teams, multiple channels, multiple approval processes – without standardised phrasing and placement of the label, uncoordinated sprawl quickly emerges across departments. This is precisely where a central guideline pays off: one team, one wording, one approval process across all channels – which is exactly what we at oha! are currently rolling out with several clients before 2 August turns into a stressful rush.

How do you properly label AI content on Meta, YouTube, LinkedIn, TikTok, and Pinterest?

Every channel works differently. On Meta, the system automatically applies an “AI Info” label to Facebook and Instagram ads as soon as tools like Background Generation or Image Generation are used in the Ads Manager, or C2PA metadata from third-party tools is detected – for ads, declaration is mandatory. On YouTube, you activate the “Altered or synthetic content” toggle when uploading in Studio; YouTube then automatically adds a label to the video description. LinkedIn has no technical labelling tool, relying purely on self-declaration in the text or as a hashtag. On TikTok, there is the “AI-generated content” toggle – once set automatically via C2PA, the label can no longer be removed retrospectively. Pinterest automatically scans images for AI signals and adds a “Gen AI” label directly on the Pin upon detection.

How do you integrate AI labelling into your editorial calendar?

1. Ask for every asset: Does it contain AI-generated or substantially altered material – image, video, voice?

2. Check the realism: Could it be mistaken for a real person, a real place, or a real event?

3. Set the channel toggle where available: Meta AI Info, YouTube Altered/Synthetic Content, TikTok AI-generated Content.

4. On LinkedIn and Pinterest, also add a text notice in the caption, Pin title, or description.

5. For ads and branded content: check the second toggle and keep an eye on ad approval.

6. Standardise phrasing across the team: same wording, same placement, consistent across all channels.

Is labelling a trust booster or a showstopper?

According to a Marketagent study, 54 per cent of Austrians now trust AI-heavy content less, and 47 per cent scrutinise it more critically. At the same time, 85 per cent demand clear labelling. That is not a contradiction; that is the equation: Those who label transparently don’t lose trust – those who don’t and get caught, do. According to the same study, 71 per cent of respondents already perceive a noticeable increase in meaningless AI content spam, and that is precisely what brands want to set themselves apart from through proper labelling.

Our recommendation at oha! digital therefore aligns with what many clients already want to implement: Always actively label AI use yourself, regardless of what the individual platform technically requires or what the law prescribes in a specific case. Labelling twice is never a problem; not labelling quickly becomes one.