AI Transparency Statement

Under the EU AI Act (Regulation (EU) 2024/1689) · FeedFocus App · As of: October 2026 · Version 5

1. Use of AI systems

FeedFocus uses artificial intelligence (AI) to generate content automatically. In accordance with Art. 50 of the EU Regulation on Artificial Intelligence (AI Act), we inform you transparently about the nature, scope and purpose of our use of AI.

2. AI-powered features

2.1 Video Summaries

Every AI Summary in FeedFocus is generated automatically from the transcript of a YouTube video. The user receives:

• A free-text summary of the video content
• Numbered key points
• A time-saved estimate
• A tone assessment (e.g. factual, opinion-driven)
• A source-character rating

Models used (as of October 2026): Anthropic Claude (claude-sonnet-4-6) for summaries you request when opening a video; Google Gemini (gemini-3.1-flash-lite) for summaries generated in the background for briefings and digests; Google Gemini (gemini-3.1-flash-lite) with direct video input as a technical fallback when no transcript is available. This statement will be updated if the models change. Processing takes place via server-side edge functions — the models have no access to personal user data.

2.2 AI Perspectives

The multi-perspective feature combines three AI providers:

• Perspective A (fact check): Anthropic Claude (claude-sonnet-4-6) — fact-based, balanced analysis
• Perspective B (devil's advocate): Google Gemini (gemini-3.1-flash-lite) — highlights overlooked aspects
• Perspective C (synthesis + blind spot): xAI Grok (grok-4-1-fast-reasoning) — contradictions and blind spots

Each perspective is marked as AI-generated and carries the name of the provider used. The final assessment of a video (e.g. “Exceptional Value”, “Solid Information”, “Entertainment Value”, “Handle with Care”) is likewise AI-generated and does not constitute a recommendation.

2.3 Audio Digest podcast (Briefings)

FeedFocus creates automated audio podcasts from video Summaries:

• Podcast script: created by AI (Anthropic Claude, claude-sonnet-4-6)
• Speech synthesis: xAI Grok TTS with five synthetic voices; no other speech-synthesis provider is used

All audio content is fully AI-generated. These are not human speakers; all voices are synthetic.

2.4 "Ask the Video"

Users can ask questions about the content of a video. The answers are generated by Anthropic Claude (claude-sonnet-4-6) solely on the basis of the video transcript and are marked as AI-generated in the app. The model does not access information outside the transcript; questions the video does not answer are responded to accordingly.

2a. FeedFocus's role under the AI Act

FeedFocus uses exclusively third-party AI models (Anthropic Claude, Google Gemini, xAI Grok) and does not develop or train its own AI models. Within the meaning of Art. 3 of the EU AI Act, FeedFocus therefore acts as a deployer of these AI systems, not as a provider. The obligations of the respective model providers (including technical documentation, risk management at the model level, and machine-readable marking under Art. 50(2) of the AI Act) remain with Anthropic, Google and xAI. FeedFocus fulfills the obligations incumbent upon it as a deployer, in particular the transparency obligations under Art. 50 of the AI Act toward the app's end users.

3. Risk classification

Under Article 6 of the AI Act, the AI systems used in FeedFocus do NOT fall into the high-risk AI category. They are limited-risk AI systems, subject to the transparency obligations under Art. 50. FeedFocus meets these by:

• Labeling all AI-generated content in the app
• Disclosing the providers and models used
• Noting the synthetic nature of the audio content

4. No automated decision-making

FeedFocus makes no automated decisions within the meaning of Art. 22 GDPR that produce legal effects concerning the user. All AI-generated content serves solely to inform and summarize. The decision whether and how to act on this content rests entirely with the user.

5. Human oversight

The user can open the original video on YouTube at any time and verify the AI-generated content independently. Every summary includes a direct link to the source video.

6. Limitations and notes

AI-generated content may:

• Contain errors, inaccuracies or distortions
• Fail to fully capture the video's contextual information
• Misunderstand irony, sarcasm or cultural nuance
• Be outdated if circumstances have changed since the video was created

The AI content does not constitute advice (in particular no legal, financial, investment, tax or health advice). See the separateDisclaimer for details.

7. Competent supervisory authority

The competent market surveillance authority for compliance with the transparency obligations under Art. 50 of the AI Act in Germany is, pursuant to the AI Market Surveillance and Innovation Promotion Act (KI-MIG), the Federal Network Agency (Bundesnetzagentur, BNetzA), which also acts as the central point of contact.

8. Contact

For questions about the use of AI in FeedFocus, contact: support@feedfocus.app

Change history

• Version 5 (October 2026): Model for summaries created in the background (briefings and digests) updated to gemini-3.1-flash-lite.
• Version 4 (September 2026): Specific model names per feature; speech synthesis now xAI Grok TTS only (ElevenLabs removed); new "Ask the Video" feature; AI-generated video assessment named for Perspectives.
• Version 3 (August 2026): Clarification of the deployer role under Art. 3 of the AI Act; addition of the competent supervisory authority (BNetzA, KI-MIG); editorial adjustments.
• Version 2 (July 2026): Complete description of the AI features and risk classification.
• Version 1: Initial version.

AI Transparency Statement | FeedFocus | feedfocus.app | Version 5 (October 2026)