AI Act Checker

AI Labeling — Checklist

Compact overview: when to label, label examples, channel tips.

LabelNot requiredBorderline / special case

Rule of thumb: it's not about "AI yes/no", but about the impression of authenticity. Metadata or watermarks alone are not enough. The notice must be visible/audible to people.

Content type. Basic rules

Content / useRequired?Note
Fully synthetic AI images (Midjourney, DALL·E, Stable Diffusion, AI lifestyle mockups)YESAI-generated, often looks realistic
AI-manipulated images/videos/audio (generative fill, background swap, remove/add objects, ML skin smoothing)YESAI alters image content → deepfake risk
Classic image editing without AI (color, light, contrast, cutout, pixel-based retouching)NONo AI model used
Product photos, only technically optimized (noise reduction, exposure, sharpening, neutral background)NOSubstantive content unchanged
Product photos, content altered (objects/people/places/situations added, removed, recolored)YESAlters authenticity
Schematic graphics, infographics, icons (recognizably graphic, not a real photo)NONo deceptively realistic reference
Virtual models / usage scenes that look like real photosIN DOUBTCould be mistaken for a real photo
Motifs on T-shirts, mugs, etc.ONLY IF REALISTICPhotorealistic/authentic-looking → Yes · Abstract/stylized → No
AI voice / AI voice-over (product video, ad, podcast)YESAudio deepfake. Notice at the start + in the video
AI chatbot / AI assistant on a websiteYESUsers must recognize AI at first interaction (unless obvious)
Pure product/ad copy, instructions, FAQsNONot a matter of public interest. Usually no obligation
AI text on matters of public interest (politics, health, science)YESException: human-reviewed + editorial responsibility named

Online shop. Scenes

Scene in the shopRequired?Practical tip
Main image: AI-generated lifestyle mockupYESLabel directly on the image
Detail view: cutout only, color correctionNOClassic editing
Size/usage example (model on sofa, AI background)YESIf a realistic photo
Schematic diagrams (dimension sketch, icon set)NOObviously graphic
Before/after (wrinkles/smoothing, body shapes) with AIYESSignificantly alters authenticity
Review photos, AI-retouched (skin, lighting, background)YESAI manipulation + realistic impression
360° view with AI additions (missing angles generated)YESAI fills in content
AR try-on / room placement (AI-based)YESSynthetic overlay, looks real
Trust elements (logos, certificates). AI-fakedSPECIAL CASENo AI Act obligation, but risky under competition law

Campaign & social media

Campaign elementRequired?Note
Social post image (AI-generated, realistic)YESDeepfake logic
Social post image (illustration / comic / abstract)NOObviously artificial
Influencer content with AI model/person (photorealistic)YESCould be mistaken for a real person
Display banner with AI photo (person/place/usage scene)YESLabel visible in/near the banner
Video ads with AI narrator or virtual modelYESAudio and/or visual deepfake
Video ads with classic post-production (cut, color, no AI)NONo AI generation/manipulation
YouTube description / post caption as the only noticeNO. Not enoughNotice must be visible on the content itself
Chat ads / conversational ads (AI bot)YESRecognizable as AI at first interaction
Social post text on public-interest topics (news, politics, health)YESException: editorially responsible
Pure product advertising, promo, saleNONot a matter of public interest. Usually no obligation

Print & offline

Print elementRequired?Why / how
Flyer / poster with AI photo (person/place/scene, realistic)YESDeepfake logic also applies offline
Flyer / poster with illustration / 3D render (obviously artificial)NONo deceptively realistic impression
Product catalog: AI lifestyle scenes (living room, model)YESLabel in/near the image
Packaging: AI-generated motif (person/place/event)YESNotice on/near the packaging
Packaging: abstract pattern, icons, typographyNOObviously not real
Print ad (newspaper/magazine) with AI photoYESLabel in the image area or caption
Inserts with AI-voice QR videoYESNotice in the video / QR landing page
Synthetic testimonials / expert statements (AI person/AI voice)YESDeceptively real person/opinion
Clearly "fictional scenes" (cartoon, comic, obviously staged)NOObviously artificial/fictional

Phone & interaction

ScenarioRequired?Implementation
Outgoing calls with AI voice / AI assistantYESState clearly at the start of the call
Hotline with AI voice botYES"You are connected to an AI assistant."
Human agent with AI tools (transcript, suggestions, no AI voice)NOUser is speaking with a human
AI voice that sounds like a known personYES + RISKClear notice + check unfair-competition and personality-rights law

Internal & special cases

ScenarioRequired?Required?
Internal AI tools without publicationNO
Internal reports / presentationsNO
Legacy content from before 02 Aug 2026, unchanged onlineNO
Legacy content re-published / changedCHECK

Internal & special cases

ScenarioRequired?Required?
Product with an AI feature (chatbot, image generator)YES
Software/code generated by AINO
High-risk AI (medicine, scoring)OTHER OBLIGATIONS
Labeling replaces rights clearanceNO

How to label — web & print

🌐 Web (shop, landing page, social)

  • Label directly on the image/video or immediately below it
  • In the visible product area. Not in the T&Cs or at the end of the description
  • Videos: label at the start + repeat regularly
  • Audio: spoken notice at the beginning
  • EU icon set optional, but icon + text recommended
  • Caption/description alone is not enough

🖨 Print (flyer, catalog, packaging)

  • Notice in/near the image or as a caption
  • Packaging: label, tag, insert or back
  • If artistic/fictional: more discreet, but present
  • QR video: notice in the video itself or on the landing page
  • Same deepfake logic as online

📞 Phone & chat

  • AI call: state clearly at the start
  • Chatbot: recognizable at first interaction
  • Wording: "You are speaking with an AI-driven assistant"

✍ Text

  • Pure ad copy: usually no obligation
  • Unless the ad itself informs about public-interest topics (health, politics): then label it
  • Exception: human-reviewed + a responsible person named
AI-generated imageAI GENERATEDAI MODIFIEDAI-generated voiceThis video contains AI-generated content

Checklist before publishing

  • AI use identified (generation or content manipulation)?
  • Content is being published (not just internal)?
  • Checked for authenticity impression. Does it look like a real photo/person/place/voice?
  • If required: notice placed clearly, understandably and visibly/audibly?
  • For video/audio: notice at the start (not only at the end)?
  • For print: notice in/near the motif or on the packaging?
  • Rights checked (personality, copyright, trademark law). Labeling does not replace this?
  • Accessibility: sufficient contrast, alt text for screen readers?

Legacy content: content published before 02 Aug 2026 that remains unchanged does not need to be labeled retroactively. Re-published or substantially changed → check again.

Orientation based on the EU AI Act (Art. 50). Not legal advice. For borderline cases consult a qualified lawyer. Effective from 02.08.2026 · ai-act-checker.com