San Francisco city workers sent 1M AI messages in 9 months

San Francisco's municipal workforce has generated over 1 million AI chatbot interactions in the nine months following the city's deployment of Microsoft Copilot to 30,000 employees, according to data the city shared with the San Francisco Chronicle.

The OpenAI-powered tool became available to more than 40 city departments in July 2025, intended to accelerate administrative tasks including report drafting, data analysis, and document summarization. Yet adoption patterns reveal a government workforce split between AI power users and those who never engage with the technology at all.

Adoption follows job type, not just availability

White-collar departments showed the highest usage rates. The Department of Police Accountability led major departments in Copilot adoption since December 2025, using the tool for large-scale data analysis, body-camera footage review, and generating summary transcriptions.

Marshall Khine, the department's chief of staff, emphasized that AI serves as a focusing tool rather than a replacement for human judgment. "AI is not a replacement for their investigations," he told the Chronicle. "They still need to review all the materials."

The Department of Public Health recorded the highest raw number of interactions, deploying AI for communications drafting, document summarization, presentation generation, and analysis tasks.

Meanwhile, physically intensive roles showed consistently low adoption. Firefighters and police officers rarely used the technology, and overall adoption across most departments remained below 50% of eligible employees.

Power users drive department totals

Some departments show concentrated usage among a small group of employees. A single collections officer in the city's Port department logged over 1,200 interactions between December 2025 and April 2026—approximately 10 per workday.

Overall usage intensity has grown substantially even as the share of active users plateaus around half of those with access. AI interactions increased more than 200% between July 2025 and April 2026.

Guardrails and training accompany rollout

The city updated its AI guidelines in July 2025, requiring employees to review and fact-check all AI-generated content, particularly for public-facing or sensitive work. Prohibited uses include creating official documents without expert review, generating deepfakes or impersonations, fabricating survey respondents, and reviewing legal or regulatory matters.

Jane Gong, the city's director of emerging technologies who led the implementation, stressed individual accountability. "No matter what you put out as a city employee, you are responsible," she said in a December interview. "Whether it's AI or not, you always have to check your work."

The Copilot deployment followed a six-month pilot program where over 2,000 staff members tested ChatGPT. More than 3,000 employees attended OpenAI training sessions. In pilot surveys, 70% of respondents reported saving up to five hours weekly.

Why it matters

San Francisco's experience offers a real-world benchmark for government AI adoption in the technology's birthplace. The data shows that simply providing access doesn't guarantee usage—organizational culture, job function, and individual workflow compatibility drive actual adoption. For enterprise leaders evaluating AI deployments, the pattern suggests that targeted rollouts aligned with specific job functions may prove more effective than blanket access policies.

The San Francisco Chronicle first reported these details based on city data obtained through July 2026.