Funding: Own budget (19), Public broadcaster / state media (3), Tech company (other) (2)
AI audio news summary in app — AI voice reads the day's three main stories selected by editors in approximately 1.5 minutes, enabling users to catch up on top news during commutes or brief moments; launched April 10, 2026; available free to all users (not registration-gated); paid subscribers can access archived editions while free users get only the current day's audio; integrates with a new 'playlist' feature…
Generative AI for editorial workflow: automated headline generation (40–70% reduction in per-article workload), article summarisation and content restructuring into multiple formats (child-friendly versions, Q&A, quizzes); developed with Dentsu Digital from April 2024; AI Governance Committee established; restructured content commercialised externally
Jiji Trend Quiz (時事トレンドクイズ) — Gemini 2.5 Pro automatically generates current affairs quiz content from veteran reporters' commentary articles, supplemented via vector search with related straight news articles from Jiji archive; converts archival journalism into interactive social learning content; published Tuesdays and Fridays on Jiji.com; won Excellence Award (優秀賞) at 4th Google Cloud Generative AI Innovation…
In a Nikkei explainer ('日経の文系記者がAI作ってみた' — a non-technical Nikkei reporter builds an AI in ~15 minutes with only clicks), a reporter used the no-code AutoML platform AMATERAS RAY (with LightGBM under the hood) to build a model predicting the Nikkei 225 average and analysing economic indices — showing how accessible machine learning has become for journalists without a coding background.
AI割付" (AI Layout) system — automatically places newspaper advertisements on pages using AI; encodes complex client conditions (colour specs, competing advertiser exclusions, seasonal requirements) into algorithms; developed jointly with Data Artist (Dentsu Group); after ~2 years of development achieved near-human placement quality
Proprietary LLM for broadcast production support — fine-tuned on approximately 40 years of NHK broadcast materials (~20 million sentences of news scripts, articles and subtitles); reduces factual errors in AI responses about broadcast content by ~10%; planned applications include information gathering support, translation and manuscript editing; target deployment 2026