By Niels Geominy, founder of RecruitmentAds.com
This is part 3 of a three-part series from RecruitmentAds.com, sponsor of EB EXP Fall 2026. Part 1 goes live Mon Nov 16, part 2 Wed Nov 18, part 3 Fri Nov 20.
Europe now has a general rule for labeling AI-generated people in ads. The US does not yet: outside the EU there is no general labeling statute, only platform policy, advertising-standards law, and a growing set of narrow national and US state rules. But if your ad is seen in the EU, the EU rule reaches you regardless of where you sit.
So should you label AI-generated recruitment ads? If your ad shows a photorealistic AI-generated person and reaches candidates in the EU, yes. The label is your job, not the platform's.
Does it reach US and UK employers?
Yes, on the Commission's reading. Article 2(1)(c) covers providers and deployers outside the EU where the output is used in the Union. For deployers, the guidelines apply the duty where you foresee your content being used in the EU, including by posting it on the globally accessible internet. The limit is EU viewers reached through channels that are unforeseeable and outside your control, such as a VPN viewer of a geo-fenced US campaign.
A US employer deliberately advertising to candidates in the EU is squarely in scope. So is a UK agency producing an ad for a Dutch client. Whether a single incidental EU viewer of a public post is enough is contested, and no court has ruled on it. Geo-fencing is a meaningful mitigation.
What Article 50 requires
Article 50, the EU AI Act's transparency regime, has applied since August 2, 2026. It was not delayed: the Digital Omnibus on AI deferred the high-risk regime, including the employment entry in Annex III, to December 2, 2027, and Article 50 sits outside the risk tiers. The one grace period, to December 2, 2026, is for providers of tools already on the market, covers machine-readable marking only, and does not touch your duty.
Two duties matter for recruitment ads, and they fall on different parties. Under Article 50(2), the provider of the generation tool must mark its output in a machine-readable format. Under Article 50(4), the deployer must disclose that a deep fake is artificially generated or manipulated. Article 50(5) says how: clearly and distinguishably, at the latest at the time of first exposure.
A recruitment marketer is a deployer, and the Commission's guidelines say deployers cannot rely on the provider's machine-readable marking, because it is not immediately clear and distinguishable.
An invented person still counts
The Commission's guidelines of July 20, 2026 read the deep fake definition broadly. It is enough that a simulated person resembles someone who can plausibly exist, and "persons" includes realistic AI-generated avatars. No intention to deceive is required.
Photorealism is the lever. A deliberately stylized or illustrated treatment can take an asset outside Article 50(4) altogether. The carve-out for artistic or fictional work will not rescue a job ad, because a job posting ad is commercial and informative.
Will the platform label it for you?
No, for two reasons. Meta's AI info tag, LinkedIn's Content Credentials icon, TikTok's AIGC label and YouTube's altered-content disclosure are platform policy, not compliance with Article 50(4). And the automatic labels will not appear on an AI-avatar job ad anyway. They fire when the platform reads a C2PA or IPTC manifest, and no mainstream synthetic-presenter tool writes one.
LinkedIn, the platform that matters most for recruitment, has no self-declaration mechanism at all. The whole disclosure has to live in the creative.
What to do
Label by default. It is cheaper than the argument. Then work through six decisions, in order:
Decide whether the asset is a deep fake at all. A non-photoreal treatment removes the obligation instead of satisfying it.
Work out whether you are the deployer: whoever decided to use the tool and controlled how it was used. An agency that runs the tool is one. An employer that only commissions the agency, and takes no decisions about the AI use, is not. Both can be deployers on the same campaign.
Put the label on the asset, not around it. The Code of Practice publishes a free EU icon, with the capitalized acronym AI as the main visual element, placed where no overlay intrudes, such as the top-right corner. Ad copy, a caption or a platform toggle is not the label.
Make it survive a scrolling candidate. For video, show it at the start, at regular intervals and after any interruption. A start-only label fails for anyone who joins a loop midway.
Check the ad is honest without the label. In September 2025 the UK's Advertising Standards Authority upheld a complaint against a Midnite ad even though it carried an "ai-generated parody" disclaimer. The most exposed format is the AI-generated employee testimonial.
Keep the record: which tool generated what, who approved it, which label was applied, and the AI-literacy briefing given to whoever operated the tool.
If a real employee's face or voice drives the generation, the label is the easiest of your problems. Data protection and likeness rights are stricter, and a standard photo release does not cover synthetic generation.
Breaches of Article 50 can cost up to €15 million or 3% of total worldwide annual turnover, whichever is higher. For SMEs and start-ups it is whichever is lower. Following an adequate code of practice is an express mitigating factor.
This is not legal advice. Have your own counsel check anything you intend to rely on.
Sources
https://recruitmentads.com/resources/guides/ai-generated-rec...
https://recruitmentads.com/resources/employee-photo-release-...
This article is part of the RecruitmentAds.com series on the EB EXP blog. RecruitmentAds.com sponsored EB EXP Fall 2026 — watch their session, "Recruitment Videos at Scale, Without Picking Up a Camera," with Niels Geominy and Andrew Hall.