つながりAIチャットふじさわ 結果報告書 ～孤独・孤立対策における「デジタル窓口」の導入成果と今後について～

Key Findings

• — Over four months 1,389 people registered, 2,582 consultations were held and more than 22,000 messages were exchanged; about 65% of messages were sent at night or on holidays, and about 70% of users had never used a city consultation service before.
• — Crisis and safety topics (domestic violence and suicidal ideation) together with loneliness and isolation made up more than half of consultation content.
• — The AI flagged 4,189 messages as risk-detected; cases judged to need a caseworker were switched to human consultation by the vendor's psychologists and social workers, and 18 people were connected to relevant agencies; 346 referrals to social resources were made.
• — A monthly count of positive words (24 terms such as thank you, happy, relieved) and negative words (30 terms such as anxious, lonely, painful) showed negative words rising in November and December and the positive rate reaching its highest in January.
• — In a web survey of users (22 December 2025 to 5 January 2026; 265 of 1,180 responded, 22.4%), 76% wanted the service to continue and 60% were satisfied; consultation counts at the existing municipal windows did not fall, which the city reads as the AI reaching a population the windows did not.

Methodology Notes

Eleven-page municipal results report dated May 2026 (month precision; no day stated). Administrative counts from the vendor platform plus a low-response user survey; no control group, no outcome follow-up on the 18 connected cases, and the risk-detection criteria and precision are not described. Word-frequency sentiment analysis is descriptive. The PDF is hosted on the city's document server without a landing page; the city names the vendor's professional staff (psychologists, social workers) as the human takeover tier. The report also lists operational problems (response accuracy, internal coordination, user concerns about copyright and data handling).

Sources