Key findings
- BBVA reports that 83% of licensed users use ChatGPT Enterprise at least once a day after a year. The bank provided training and a practice community, plus an internal library of about 1,000 shared GPTs.
- Pennsylvania's pilot participants reported saving 95 minutes a day, from exit surveys of 175 employees, and named inaccuracy and learning time as barriers.
- At California State University, 97% of faculty surveyed say AI output must be verified, and communication was the hardest part of a 22-campus rollout.
- Without enablement, use stalls. One consultancy found about 20% of a client's licensed employees had activated ChatGPT Enterprise months after rollout.
- Of 37 public accounts from teams and organizations, 11 also use or pay for another assistant.
ChatGPT licenses turn into daily use when the organization trains people and builds shared assistants, and every account keeps a person reviewing the output. A bank, a state government, a university system, and a consultancy report this in public, with time savings their own staff reported. Controlled studies of the same kind of tool measure real gains on some tasks and worse answers on others.
Adoption figures, as reported
Each figure comes from a different organization and measures something different, so read each on its own. The organization or its adviser reported every one.
| Figure | Organization | Share reported |
|---|---|---|
| Faculty saying AI output must be verified | California State University, fall 2025 survey | |
| Pilot participants reporting a positive experience | Pennsylvania, 175 employees | |
| Licensed users using ChatGPT at least once a day | BBVA, 11,000 licenses | |
| Licensed employees who had activated ChatGPT | SVA Consulting client, months after rollout |
BBVA (May 12, 2025). Commonwealth of Pennsylvania (March 21, 2025). California State University via the Changing Higher Ed podcast (June 9, 2026), survey n = 94,060. SVA Consulting (March 23, 2026), one client engagement.
Enablement decides use
The organizations with the highest reported use included training in the rollout. BBVA trained employees on data security and confidentiality. It also ran a practice community and maintained a shared GPT library. CSU trained more than 4,000 people at system level and funded 63 faculty projects. Pennsylvania ran a year-long pilot with 175 employees before any wider purchase.
The cautionary account is about licenses without that work. SVA Consulting describes a client that licensed ChatGPT Enterprise across the organization and found about 20% of employees had activated it several months later. SVA's response was hands-on training built around role-specific use cases.
Review stays part of the job
BBVA's nine-lawyer advisory team uses a custom GPT to answer more than 40,000 legal queries a year. A person reviews every answer before it goes out. Pennsylvania stressed human oversight, and its report lists inaccuracy concerns among the barriers. A ChatGPT Team user reported fabricated citations to files in their own project.
What controlled studies measure
Staff reported the time savings in these accounts themselves. Pennsylvania's exit survey reports 95 minutes a day, and BBVA employees who automate tasks report 2.8 hours a week. Ethan Mollick, checking GPT-5 Pro's work by hand, notes that checking sometimes takes far less time than doing the work and sometimes a lot more.
Controlled studies supply the baseline. A Harvard Business School field experiment followed 758 Boston Consulting Group consultants. On tasks inside the model's strengths, those given GPT-4 completed 12.2% more tasks, finished them 25.1% more quickly, and did better work. On a task chosen to fall outside them, consultants using AI were 19% less likely to produce a correct solution.
Across Denmark's labor market, Anders Humlum and Emilie Vestergaard linked chatbot adoption surveys to payroll records in a National Bureau of Economic Research study. They found no effect on earnings or recorded hours, ruling out effects larger than 2% two years after ChatGPT launched, while workers still reported productivity benefits.
Methodology
GILD read 86 public accounts of ChatGPT at work, published between January 7, 2025 and October 27, 2026. Of these, 41 describe team or department use, IT and business leadership decisions, or organization-wide rollouts. Each figure keeps the measure its organization reported. The controlled-study figures come from the Harvard Business School and National Bureau of Economic Research papers linked under Sources.
Products in this piece
Sources
- Public source
- 10
- Official documentation
- 4
- BBVA: BBVA expands its agreement with OpenAI to 11,000 ChatGPT licences (May 12, 2025)
- BBVA: BBVA is now using ChatGPT to streamline legal queries and marketing processes (Jan 7, 2025)
- Commonwealth of Pennsylvania: First-ever generative AI pilot for state workers (Mar 21, 2025)
- Digital Government Hub: Lessons from Pennsylvania's Generative AI Pilot with ChatGPT (Jun 4, 2025)
- Changing Higher Ed podcast 315: CSU's ChatGPT Edu rollout, with Leslie Kennedy (Jun 9, 2026)
- SVA Consulting: Driving ChatGPT Enterprise Adoption (Mar 23, 2026)
- OpenAI Community: Fabricated citations from Project documents (May 14, 2025)
- Ethan Mollick: On Working with Wizards (Sep 11, 2025)
- Harvard Business School: Navigating the Jagged Technological Frontier, Working Paper 24-013 (Dell'Acqua et al., Sep 2023)
- NBER: Still Waters, Rapid Currents, Early Labor Market Transformation under Generative AI, Working Paper 33777 (Humlum and Vestergaard, revised Mar 2026)
- ChatGPT admin docs: ChatGPT Work usage and cost
- ChatGPT admin docs: ChatGPT Work admin FAQ
- OpenAI Help Center: What is ChatGPT Enterprise?
- OpenAI Help Center: ChatGPT Business overview