AI Content Disclosure Policy
Effective Date: May 15, 2026 · Last Updated: May 15, 2026
BytePulse is an AI-native platform. Transparency about how AI is used is central to our commitment to media literacy.
1. AI Systems Deployed
| Function | AI Technology | Human Oversight |
|---|---|---|
| Surface summaries | LLM with source grounding | Editorial QA on Breaking/High-priority |
| Context explainers | LLM with structured prompting | Editorial QA on all outputs |
| Deep Dive chat | LLM with RAG (Retrieval-Augmented Generation) | Automated guardrails + confidence fallback |
| Beyond Bias — Perspectives | LLM comparative framing analysis | Automated quality scoring |
| Beyond Bias — Stakeholders | LLM impact extraction | Automated quality scoring |
| Beyond Bias — Blind Spots | Coverage frequency analysis engine | Automated; editorial thresholds |
| Beyond Bias — Narrative | Temporal framing shift detection | Automated + editorial review triggers |
| Quiz generation | LLM with answer verification | Random sampling QA |
| Content categorization | ML classification | CMS override capability |
| Feed personalization | ML recommendation | Diversity safeguards (no echo chambers) |
| Poll question generation | LLM with tone/complexity rules | Editorial review for sensitive topics |
| Content moderation | ML classifier + LLM filter | Human review for flagged content |
2. Disclosure and Labeling
All AI-generated content is clearly labeled:
- AI-generated summaries show an "AI" badge on story cards
- Beyond Bias lens outputs labeled "AI-powered analysis"
- Deep Dive chat identifies responses as AI-generated
- AI-generated quiz/poll questions display the "AI" tag
We do not present AI content as human-written journalism — the distinction is maintained at all times.
3. Limitations and Known Risks
- Accuracy — AI may produce inaccurate or incomplete content despite controls
- Hallucination — LLMs may generate plausible-sounding statements that are factually incorrect (RAG mitigates but does not eliminate)
- Bias — AI models may reflect biases in training data; we actively test and mitigate
- Timeliness — analyses reflect data at generation time and may not capture subsequent developments
- Interpretation — Beyond Bias outputs are computational interpretations, not definitive editorial judgments
4. Source Grounding and Anti-Hallucination
- Deep Dive responses are generated via RAG from verified source-article chunks in our vector database
- Below confidence thresholds, AI falls back to: "I'm not confident enough to answer accurately — here's the original article."
- AI is explicitly prohibited from generating information not present in source material
- All outputs include citation references
5. Age-Appropriate Safeguards
For users under 16: simplified language complexity, enhanced sensitivity filters, and restricted access to AI chat topics related to violence, substance use, or self-harm.
6. Right to Human Review
Any user may request human review of an AI output via the in-app flag icon or by emailing admin@bytepulse.club. Corrections are prioritized and processed within 24 hours.