OpenAI announced its ChatGPT for small business program on July 21, betting that most small business owners are still using ChatGPT like a search engine instead of the multistep agent it's become. The program wraps free training around ChatGPT Work, the GPT-5.6-powered agent that can pull data from connected apps and files, draft documents, and keep working on a task for hours without a person babysitting every step.
What's actually in the program
The program has a few concrete parts, not just a marketing page. There's hands-on virtual training with webinars built around specific small business use cases in accounting, marketing, and ecommerce. There are in-person "small business AI academies" that bring local business owners together across the US for guided exercises. And there's a set of prebuilt partner integrations meant to get a business connected to ChatGPT Work without hiring anyone to do the setup.
OpenAI is leaning on its earlier Small Business AI Jam program for credibility here: the company says 78% of participants built a functional workflow in a single day, and 42% reported saving more than five hours a week afterward. One cited example, Kevin English of Keg Built, says he saved at least 10 hours on a single contractor quote by cutting down manual data entry.
What it costs
ChatGPT Business runs $25 per user a month billed monthly, or $20 per user a month billed annually, in most countries. A shared business workspace needs at least two seats, so the real floor is $50 a month or $480 a year even for a two-person shop. On top of the subscription, ChatGPT Work usage is metered similarly to Codex: it scales with how complex the task is, so a long autonomous job can add cost beyond the seat price. Admin spend controls exist to cap that, but you have to set them.
| Cost | |
|---|---|
| Minimum seats (shared workspace) | 2 |
| Per seat, billed monthly | $25/month |
| Per seat, billed annually | $20/month |
| Realistic monthly floor | $50/month |
| ChatGPT Work task usage | Metered on top of subscription, scales with task complexity |
The part OpenAI isn't publishing
Here's the gap worth flagging: OpenAI hasn't published an error or hallucination rate for ChatGPT Work. Launch coverage from outlets like Reuters and BNN Bloomberg didn't include an independent risk assessment either. An agent that works autonomously for hours on real business tasks, invoices, customer emails, vendor communication, needs a human checking its output, and right now the honest starting point is that nobody outside OpenAI has measured how often it gets things wrong.
That doesn't make the program useless. A lot of small businesses genuinely don't know where to start with agentic tools, and free structured training closes a real gap that a lot of AI companies skip entirely. It just means the "10 hours saved on one quote" testimonial is doing a lot of work that a published error rate should be doing instead.
Who should try it
If you're running a small business and already pay for [Claude Opus 5](https://questloops.com/blog/claude-opus-5-explained-pricing-benchmarks-and-where-it-actually-wins) or [Grok 4.5](https://questloops.com/blog/grok-4-5-is-out-pricing-benchmarks-and-how-to-try-it-free) for occasional chat use, the free training alone is worth attending even if you don't switch platforms; it's a rare case of a vendor teaching workflow design instead of just pushing a subscription. If you're evaluating agent tools more broadly for a small team, it's also worth comparing against workflow-focused options like [Sim](https://questloops.com/tools/sim), which takes a more open, build-it-yourself approach to agent automation.
FAQ
**Do I need ChatGPT Business to join the program?** The training and academies are open access, but ChatGPT Work itself runs on ChatGPT Business or Enterprise plans.
**Is ChatGPT Work included in the regular ChatGPT Plus subscription?** No, ChatGPT Work is part of the Business tier and above, with usage metered on top of the seat price.
**Has OpenAI published accuracy numbers for ChatGPT Work?** Not as of launch. No error or hallucination rate has been published, so plan on reviewing its output rather than trusting it unsupervised.