In July 2025, San Francisco began rolling out Microsoft 365 Copilot Chat, built on OpenAI's GPT-4o, to city employees following a six-month pilot involving more than 2,000 staff. The pilot, run under then-new Mayor Daniel Lurie's technology agenda, tested the tool across departments including public health and social services; participants self-reported productivity gains of up to five hours per week on tasks such as email drafting, project planning, research and document summarisation.
By mid-2026, access had scaled roughly fifteen-fold: nearly 30,000 of the city's employees could use Copilot, spanning more than 40 departments, with over one million Copilot interactions logged between July 2025 and April 2026. The city paired the rollout with a five-week training programme delivered with InnovateUS and updated its Generative AI Guidelines (8 July 2026) under the pre-existing 22J AI Transparency Ordinance, which requires staff to take responsibility for any AI-assisted material they use or share. The deployment used San Francisco's existing Microsoft enterprise licensing, so officials describe it as carrying no additional software cost.
The five-hours-per-week productivity figure comes from self-reported pilot surveys rather than an independent time-and-motion study, and no city-wide before/after service-delivery metrics (e.g., case-processing times, error rates) have been published alongside the adoption numbers. The scale of the rollout - one of the largest of any US city government - and its transparency-ordinance safeguards are well documented, but the practice remains, in the city's own framing, an early, actively-training phase rather than a rigorously evaluated productivity gain.
Read the full analysis: https://www.govtech.com/artificial-intelligence/san-francisco-gives-thousands-of-city-staffers-ai-access
Where this practice's information was retrieved from, and when.