All insights · Article
ChatGPT at work: five rules every small business should set
Your team is almost certainly using AI already. In a May 2026 survey of 1,100 private-sector employees, 73% said they use AI at work, up ten points in a year. Fewer than one in two said their company had actually rolled it out (Le Monde Informatique, Greenworking x Ipsos BVA). The gap between those two numbers is where the risk sits. Five simple rules close most of it.
1. Say which tools are allowed
If you don't choose, your staff will, often with free personal accounts. Pick one or two tools, pay for business accounts where it makes sense, and tell everyone which ones are approved. A short list people actually use beats a ban nobody follows.
2. Keep sensitive data out of prompts
Anything typed into a public AI tool may be reused to train it. The government's France Num guide for small businesses says to keep confidential, personal and strategic information out, or to anonymise it first (France Num, fiche 4). In practice: no customer files, no contracts, no payslips, no figures you wouldn't email to a stranger.
3. A person checks before anything goes out
AI tools write with confidence, including when they are wrong. The same guide reminds firms that they are responsible for the AI they use and for the decisions taken with it. Once personal data is involved, you are the data controller under GDPR (France Num, fiche 4). A quote, a customer email or a contract clause drafted with AI gets read by a person first.
4. Be open with customers when it matters
If customers talk to a chatbot, or you publish AI-made images or video that look real, tell them. From August 2026 this is a legal requirement under the EU AI Act (Article 50). It is also simply good manners.
5. Write it down and train once
Put the four rules above on one page, with a name to ask when in doubt. Then spend an hour with the team going through it, with real examples from your business. Keep a note of who attended. That covers your AI literacy duty under the AI Act, and people trained on AI tend to worry less about it (Le Monde Informatique).
Sources
- Le Monde Informatique on the Greenworking x Ipsos BVA observatory, September 2026 (in French)
- France Num, fiche 4: precautions for responsible and secure use (in French)
- AI Act Article 50
Related reading
Want to talk this through for your business?
Let's talk