What 146 AI adoption job ads ask for, and what they skip

Companies in France are hiring people to make AI stick. I read 146 of their job ads to see what they expect from that person. They ask for training far more often than for any way to know whether the training worked.

What they ask for

• Adoption appears in 72 of the 146 ads, about half.
• Training users, workshops or upskilling: 63 ads.
• Generative AI is named in 75 ads, and Copilot in 36. Microsoft’s assistant is the tool most of these teams are rolling out.
• Stakeholders and communication each appear in over 70 ads. The job is described as change management, even when the title says "AI".

What they skip

• Measuring usage or adoption: 17 ads. Fewer than one in eight says how anyone will know that people use the tool.
• A champions or ambassador network: 15 ads. Yet colleagues who help colleagues are what carries a tool past launch week.
• Data protection or GDPR: 15 ads, although data rules are usually the first question users ask.

Why the gap matters

Training is also a legal duty. Article 4 of the EU AI Act, in force since 2 February 2025 and rewritten in July 2026, requires companies that deploy AI systems to take measures that support their staff’s AI literacy. A training plan is the minimum. I explain what Article 4 asks of employers since the change.

The law does not ask you to prove that people use the tool. Your budget does. Licences are paid per seat, every month. Without a usage measure, the only signal left is the renewal invoice.

None of this is new to AI. At Crédit Agricole CIB I drove adoption of a mobile-fleet and telework rollout for 7,000 users. Training is the part everyone can plan. Usage is the part that tells you whether it worked.

Three lines to add to your job description

If you are writing one of these ads, or briefing a consultant, add:

1. How usage will be measured, for example weekly active users and the tasks people use the tool for. I wrote about four numbers worth tracking.
2. Who supports users after launch, in weeks two to six, when real cases appear.
3. The data rules users must follow, written so a non-specialist can apply them.

Those three lines change the profile you attract: someone who owns the result, not only the training calendar.

Method and limits

On 3 November 2026 I ran ten LinkedIn job searches (France, "remote" filter, posted in the last 30 days), on terms such as conduite du changement IA, adoption IA, product owner IA and Copilot adoption. That gave 156 postings; 146 had readable text. Each count is the number of ads where a French or English keyword pattern appears at least once. A mention is not a priority, and an ad can be silent on something the team still cares about. This sample is not the whole French market.

Source for the legal point: Regulation (EU) 2024/1689 (AI Act), Article 4, "AI literacy", as amended by Regulation (EU) 2026/1744.

Planning a rollout? Start with the free AI rollout readiness checklist, or describe your project.