You have invested in licenses (ChatGPT, Claude, Copilot). You have announced it internally. A bit of enthusiasm at first… then not much. Most of your employees open ChatGPT or Copilot now and then, type a vague question, get an answer they can’t use, and conclude that “AI isn’t much use in our line of work”.
This scenario plays out in hundreds of companies, and it isn’t a tooling problem. It is a method problem.
According to an Ipsos BVA study, 76% of French people have never been trained to use AI. They use it by intuition, with no framework, no method and no sharing of good practices. The result: AI remains an individual gadget, when it could become a lever of performance for the whole organization. For that, you need a simple method that can be shared and measured.
Here is one, in 5 steps.
Step 1: Make a precise request, the number 1 skill to develop in your teams
Why precision is the foundation of everything
AI doesn’t read minds: it knows neither your industry, nor your context, nor what you mean by “a good result”. It takes your request literally, and if that request is vague, it produces a vague answer.
This is the first barrier to adoption we see in organizations: employees who write requests that are too general, get disappointing answers, and don’t understand why.
The method: 3 questions before you write
Before sending any request to an AI tool, your teams should ask themselves three questions:
- What exactly do I want? (a draft, an analysis, a list, a decision?)
- Who is it for? (my manager, a customer, a new hire, myself?)
- In what form? (an email, a table, a 5-point summary, a presentation?)
Before: “Write an email.”
After: “Write a 5-line follow-up email, professional but warm, to a key account customer who hasn’t replied to our sales proposal for 2 weeks.”
The result isn’t just better, it can be used as is, without corrections. That is where the time saved becomes real.
Key takeaway: precision in a request is a professional skill in its own right. Train your teams to structure their thinking before they talk to AI, not after.
Step 2: Give context in every exchange: what AI will never know without you
The zero-memory problem
Every conversation with an AI tool starts from scratch. The AI knows nothing about your company, your industry, your customers, yesterday morning’s meeting, the tensions in your team or the decisions made last week. It works with what you give it, nothing more.
This is one of the most common causes of disappointment: you expect an answer that fits your situation, and you get a generic one.
The 30-second context rule
Before each request, add one sentence of context, just one, and it changes everything!
Without context: “Summarize this document.”
With context: “Summarize this document in 5 key points for a CEO who doesn’t have time to read it in full and has to make a decision by Friday.”
The first result is a summary; the second is a decision-making tool.
Key takeaway: context is also what lets you build on what works. If you document the typical contexts of your organization (types of requests, recurring audiences, specific constraints), you can create prompt templates that all your teams can reuse, and make the quality of AI results consistent across the organization.
Step 3: Break complex tasks down into small steps
The single-request trap
The more complex a request, the worse AI answers it in one go. It isn’t a limitation of the tool, it is a fundamental feature of the way these models work.
Asking in a single request to “create a complete presentation of our HR strategy for the next 3 years” almost always produces a result too generic to be reused.
Work in sequence: analyze → structure → write
The effective method is to sequence the work:
- First analyze: “List the 5 key points of this HR strategy document.”
- Then structure: “Organize these 5 points into 3 coherent parts for a presentation to the leadership team.”
- Finally write: “Write a punchy introduction for this presentation, one that grabs a leadership team in 30 seconds.”
Three short exchanges always beat one long, fuzzy request. A practical rule to share with your teams: if your request is more than five lines long, it can probably be broken down, and should be.
Key takeaway: this way of sequencing translates directly into documented processes. Identifying the complex, recurring tasks of each job, breaking them down into AI steps and writing a usage guide for each department is one of the fastest levers to multiply productivity without any extra technical investment.
Step 4: Give the AI a role, the most underused lever at work
Why the role changes everything
An AI model is trained on billions of texts covering thousands of jobs, levels of expertise and contexts. When you give it a precise role, you steer it toward the part of its “knowledge” most relevant to your need.
Without a role, the AI answers in a neutral, general way. With a role, it adopts the tone, the vocabulary, the level of detail and the examples that fit that context.
The two-part formula
Start with “You are…” to set the profile: “You are a senior HR manager with 15 years of experience in manufacturing.”
Add “You are speaking to…” to set the audience: “You are speaking to a strategic customer who expects concrete recommendations, not generalities.”
Combining the two produces an answer that fits far better than a request with no framing.
Key takeaway: a training manager can build a library of “standard roles” suited to the recurring needs of the organization (business expert, coach, facilitator, writer) and make it available to every team. It is a simple way to raise quality across the board without making the tools more complex.
Step 5: Always check, because AI can be wrong with total conviction
The AI confidence paradox
AI produces fluent, well-structured, self-assured answers. It never says “I’m not sure” on its own. And yet it can be wrong (on figures, dates, sources, interpretations) with exactly the same apparent confidence as when it is right.
This is probably the most underestimated risk in rolling out AI at work. Not the risk of people refusing to use it, but the risk of people using it without judgment.
These reflexes make each person more effective. What remains is to go from individual gains to collective gains. The method: rolling out AI starting from the front line.
Move from passive consumer to active user
Checking every time isn’t a constraint: it is what makes the difference between an employee who uses AI and an employee who really gets value from it.
Three reflexes to anchor in your teams:
- Ask the AI to flag what it isn’t sure about: “Tell me clearly when you’re not certain of a piece of information.” Most recent models do this well.
- Always check figures, dates and sources before sharing them or using them in a decision.
- Treat AI output as a first draft, not a final deliverable: the human added value lies in checking, adjusting and approving it in context.
Key takeaway: building this habit of checking into the organization’s AI practices is also a matter of governance and accountability. AI errors that go unnoticed and end up in reports, customer communications or strategic decisions can cost far more than the productivity gain you expected.
Conclusion
These 5 steps make each of your employees more effective with AI. That is a first victory.
But AI only changes scale in an organization when it becomes collective. As long as each employee experiments alone in their corner, you have a sum of individual gains that nobody consolidates. You lose the feedback, the consistency, and what the organization as a whole could learn.
The organizations that really get value from AI do one more thing: they document what works, share it, and gradually build a common way of working, with tested prompts, approved processes and proven business use cases.
That is the difference between AI as an individual tool and AI as a lever of performance for the whole organization.
Spreading these good practices across the organization takes one specific lever: training. We explain why it changes everything: training your people on AI.
FAQ: your questions about using AI at work
How do you use AI well at work?
Five steps make the difference: make a precise request (what, for whom, in what form), give context, break complex tasks down into small steps, give the AI a role, and always check its answers. These reflexes make each employee more effective. AI becomes a lever of collective performance when the organization documents and shares what works: tested prompts, approved processes, proven use cases.
How do you write a good prompt?
Before writing, ask yourself three questions: what exactly do I want, for whom, and in what form? Add a sentence of context, give the AI a role (“You are…”, “You are speaking to…”), and break a long request into several short exchanges: if your request is more than five lines long, it can probably be broken down. For example, “Write an email” becomes “Write a 5-line follow-up email, professional but warm, to a key account customer who hasn’t replied to our sales proposal for 2 weeks”.
Why does AI give disappointing answers?
Most often, the request is too vague and has no context: the AI takes it literally, and every conversation starts from scratch, knowing nothing about your company, your customers or your decisions. According to Ipsos BVA, 76% of French people have never been trained to use AI: they use it by intuition, with no method. It isn’t a tooling problem, it is a method problem.
Should you check AI’s answers?
Yes, always. AI can be wrong about figures, dates or sources with the same confidence as when it is right. Ask it to flag what it isn’t sure about, check figures, dates and sources before sharing them, and treat every result as a first draft, not a final deliverable.
