For a team of about 5 to 25 people learning ChatGPT, Copilot or Claude for daily work, on-site training fits best when most of the team shares an office and needs to practice on real files with their own logins. Online fits remote or scattered staff, one or two learners, and short refreshers. Many teams do best with both.
SafeLab AI Engineering, based in Irvine, runs on-site ChatGPT, Microsoft Copilot and Claude training at the client's office across Orange County, Los Angeles and San Diego. As of November 2026 the price is $1,250 per session plus $125 per attendee for the first 10, so a team of 10 pays $2,500.
The rest of this guide covers what each format costs, what the research says, the practical factors that usually decide it, and a table you can use to choose.
What does each format cost?
Prices below come from each provider's own page, checked on November 20, 2026. For a wider look at the market, see how much AI training costs.
| Format | Published example | Price as of November 2026 | 10 people |
|---|---|---|---|
| Self-paced course | Google AI Essentials | $49 per person per month after a 7-day free trial; under 5 hours | $490 for one month |
| Self-paced vendor training | Microsoft Learn Copilot learning paths | Free | $0 |
| Public live online class | Webucator Introduction to AI with ChatGPT | $289.10 per student, 3 hours | $2,891 |
| Public live online class | New Horizons MS-4004 Microsoft 365 Copilot use cases | $495 per seat, 1 day, virtual instructor-led | $4,950 |
| Private session, online or in person | MyAIEvolution AI Lab: Focused | $1,250 for up to 25 people; virtual, or in person in Central Florida, at the same price; length not stated | $1,250 |
| Private on-site session | SafeLab AI Engineering | $1,250 per session plus $125 per attendee (1 to 10), 4 hours at your office | $2,500 |
Two things stand out. Per-seat public classes get expensive fast for a team, so a private session usually costs less once about 10 people need training. And format does not always change the fee: MyAIEvolution charges the same for virtual and in-person sessions, and Webucator offers its private classes either online or at your site. The real difference between on-site and online is what happens during the session.
One cost every live format shares is staff time. Ten people in a 4-hour session is 40 staff hours, whether they sit in a conference room or on a video call.
What does the research say about hands-on and in-person training?
Two recent studies point the same way.
- More hours and in-person coaching. BCG's June 2025 AI at Work survey of more than 10,600 leaders, managers and frontline employees found that regular AI use is sharply higher for employees who receive at least five hours of training and have access to in-person training and coaching. A one-hour webinar falls short of that bar.
- Hands-on practice during work hours. The Conference Board's July 2026 study of nearly 1,300 workers found only 48% agree their organization provides enough time during work hours to build AI skills, and leaders in the study said hands-on experimentation and practical application are critical for adoption. A self-paced course relies on people finding that time on their own.
A year later, BCG's June 2026 survey of close to 12,000 employees found only 36% feel they have received adequate upskilling. BCG called training and support from leaders one of the biggest unmet needs it found.
None of this proves a video call can't work. It does suggest that a team learns faster when someone can see their screens, answer questions as they come up, and keep the session tied to the work on their desks.
Which practical factors usually decide it?
Real files, real logins, real permissions. Your team should practice in the accounts they will use every day. Even Microsoft's own one-day Copilot course requires each student to bring a Microsoft 365 subscription licensed for Microsoft 365 Copilot. On-site, a trainer can walk the room while ten people open their own client files, find out what their permissions allow, and fix sign-in problems as they happen. Online, the same problems turn into screen-sharing queues. For Copilot in particular, what each person can reach depends on file permissions, covered in Microsoft Copilot data safety and permissions.
Staff time and focus. A session in your own conference room removes travel, and a shared room makes it easier to keep attention on the work in front of everyone. Online removes the need for everyone to be in one place at once.
Multiple locations. Two offices close together can share one on-site session. Offices far apart are often cheaper to train online, or on-site one location at a time.
Remote staff. A team that is mostly remote should train online. A team that is mostly in the office with a few remote people can train on-site and add a short online session for the rest.
Follow-up. Daily habits form in the weeks after the session. Plan check-ins, short guides and a named owner for questions, whichever format you pick. The four-week plan in how to train employees on AI works with both.
What does a hybrid plan look like?
For many teams of 5 to 25, the strongest plan mixes formats:
- Before: short staff interviews so the session uses real tasks.
- Main session on-site: a half day of hands-on practice in your own tool, on your own work.
- Remote staff: a live online session on the same tasks, for people who could not attend.
- After: a 30 to 60 minute online check-in a few weeks later, plus free self-paced material from the vendor, such as Microsoft Learn for Copilot.
This keeps the hands-on, in-person part where it matters most and uses online time for reach and reinforcement.
Decision table: on-site, online or hybrid?
| Your situation | Best fit | Why |
|---|---|---|
| 5 to 25 people in one office, new to the tool | On-site | Hands-on help in your own accounts and files |
| Most staff in one office, a few remote | Hybrid | On-site session plus a short online session for remote staff |
| Several offices far apart | Online, or on-site by location | Avoids travel; on-site per office if budget allows |
| Fully remote team | Live online | Everyone joins from where they work |
| One or two people | Public live online class or self-paced | Per-seat pricing is cheaper for small numbers |
| Already trained, need a refresher | Online check-in or self-paced | Short and low cost |
| Regulated client data (health, legal, finance) | On-site | Safe data habits are easier to coach in the room |
How does SafeLab run on-site AI training?
SafeLab AI Engineering trains teams on-site at the client's office across Orange County (Irvine, Newport Beach, Costa Mesa, Santa Ana, Anaheim and the rest of the county), Los Angeles and San Diego. Teams train on the tool they already pay for:
- ChatGPT training. SafeLab is an OpenAI Select Partner.
- Microsoft Copilot training. SafeLab is a Microsoft AI Cloud Partner Program member.
- Claude training. The team includes Claude Certified Associates and Developers.
Gemini business plans are covered too when that is what the client uses.
Each engagement is one 4-hour hands-on session where staff work on their own real tasks. Staff interviews before the session are included, prompts and short guides are left behind, and the session covers safe data habits for client information.
Price, as of November 2026: $1,250 per session plus $125 per attendee for attendees 1 to 10, $100 each for attendees 11 to 25, and $75 each from attendee 26, with a 5-attendee minimum. That works out to $1,875 for 5 people, $2,500 for 10 ($250 each) and $3,500 for 20. Sessions held on or before December 31, 2026 are 20% off (10 people = $2,000). Larger teams, several locations and other formats get a custom quote. Full details are on the AI training page.
ETP funding. California employers may be able to get training costs reimbursed through the Employment Training Panel. ETP pays only for trainees who complete at least 8 hours of training, so a funded program spans two or more sessions. Its approved training methods include classroom training and instructor-led distance learning, which leaves room for a hybrid plan. SafeLab works with an ETP funding partner to check eligibility and handle the application; apply before training starts.
The bottom line
If most of your team works in one office and is learning ChatGPT, Copilot or Claude for daily work, train on-site in your own accounts and files, then reinforce online. If your team is remote or spread out, train live online and keep sessions hands-on. Either way, plan at least five hours of guided practice across the first month, because that is where daily habits form.

