Teach employees one real task per role first, not the whole tool. Then add three things everyone shares: house rules on which data never goes into ChatGPT or Claude, a habit of checking every output before it leaves the team, and a small library of shared prompt templates. Measure use before and after so you know what changed.
#Key takeaways
- The first lesson should be a task from the person's own job, not a tour of features.
- House rules on data come before any prompt tips, and they should fit on one page.
- Checking outputs is a skill people practise, not a line in a policy.
- Shared prompt templates turn one person's good prompt into a team habit.
- Measure who uses the tool, on which tasks and how often outputs are kept or corrected.
#What does a team's week with ChatGPT or Claude look like today?
In most companies, some people already use it daily and most do not. A sales coordinator pastes call notes into ChatGPT to draft a follow-up email for HubSpot. A finance analyst asks Claude to explain a variance in an Excel export. An HR partner rewrites a job advert. Nobody agreed on which account to use, what data is allowed, or who checks the result.
That is the normal starting point, not a failure. The Microsoft and LinkedIn Work Trend Index, may 2024 found that 75% of knowledge workers use AI at work, 78% of AI users bring their own tools, and only 39% of AI users had received AI training from their company. Training starts from what people already do.
#Which tasks should each role learn first?
Pick one task per role that is frequent, written and easy to check. The first win should come from the person's own inbox or spreadsheet, so the habit has somewhere to live.
- sales: turn call notes into a follow-up email and a CRM note in HubSpot or Salesforce.
- customer support: draft a reply from the knowledge base article, then edit it before sending in Zendesk.
- finance: write first-draft commentary on a variance from an Excel export, with the numbers checked by hand.
- HR and L&D: draft job adverts, interview question sets and policy summaries in plain language.
- operations: summarise a long supplier or WhatsApp thread into decisions, owners and open questions.
- managers: turn meeting notes into a short update and a list of actions.
#What house rules should everyone learn before the first prompt?
Keep the rules short enough to remember. Stennir's baseline for a team starting with ChatGPT or Claude covers five points:
- Use only the approved accounts, such as a company ChatGPT or Claude workspace, never a personal login.
- Never paste personal data, salaries, health information, client contracts or anything under a confidentiality clause into an unapproved tool.
- A named person checks any output that reaches a customer, a regulator or a payslip.
- Do not let the tool act on instructions found inside pasted documents or emails; the OWASP top 10 for LLM applications lists prompt injection as its first risk.
- Report mistakes early, without blame, so the rules improve.
#How do you teach people to check AI outputs?
Give people a short checklist and practise it on real drafts: are the facts and figures traceable to a source document, are names and dates right, does the tone fit the customer, and is anything missing that a colleague would expect. The habit forms when a trainer shows a plausible but wrong answer and the group finds the error together.
In Arabic and English teams, also check the language itself: a reply in stiff formal Arabic, or in English, can read badly to a customer who wrote in Gulf dialect.
#How do shared prompt templates help?
A template is a good prompt someone already tested, saved where the team can find it: the task, the context to paste, the format wanted and the check to run afterwards. Templates cut the gap between confident users and hesitant ones, and they give the trainer something concrete to improve as the tools change.
#Are the vendors' own free courses enough?
For the basics, they are a fine starting point. OpenAI Academy offers free, self-paced courses on using AI at work, according to the OpenAI Academy, checked november 2026, and Anthropic publishes a free AI fluency course built on its AI fluency framework, november 2026. Use them as the shared baseline.
What they cannot do is teach your house rules, your templates or your workflows. That part needs training on your own documents, by role.
#Where does Stennir fit?
Stennir runs this as the AI Operator Program in AI training: role-specific cohorts that learn on the real workflow, with house rules, output checks and templates built from your own documents. Trainers are practitioners who do this work, and cohorts run in Arabic and English. Usage is measured before and after, and the program ends with an adoption report.
We are
Stennir trains each role on its own first tasks with ChatGPT, Claude or Copilot, using your data rules and your documents, and measures the change in use.
We aren't
Stennir is not a tool reseller, and not a generic prompt course that ends when the slides do.
#What should an HR or L&D lead do next?
Ask each department head for the one task their team would hand to ChatGPT or Claude tomorrow, and write the house rules on one page. If staff already use ChatGPT with company data, read how to train staff who already use ChatGPT with company data; if you bought licences nobody opens, read Copilot licences nobody uses. See AI for operations, then take the free readiness assessment or book a 30-minute call.
#Common questions
Should we train on ChatGPT, Claude or Copilot?
Train on the tool your company has approved, with the account people will actually use. The first tasks, house rules, output checks and templates carry over between tools, so the choice matters less than people expect. If several tools are allowed, teach the rules once and show each role which tool fits its task.
What do you need from us to start?
Stennir needs the list of approved tools and accounts, your current acceptable-use policy if you have one, and one person per role who can describe a real task and share sample documents. Where those documents hold personal data, we redact them or build realistic samples before anyone practises on them.
Does the training work in Arabic?
Yes. Stennir runs cohorts in Arabic and English, and other languages on request. People practise on documents and messages in the language they work in, including mixed Arabic and English threads, and the output checks cover tone and dialect as well as facts, so a reply reads right to the customer.
How do we measure whether people use it?
Record a baseline before training: who uses the approved tool, on which tasks, and how often outputs are kept or heavily corrected. Measure the same things afterwards. Admin usage reports help, but a short role survey and a sample of real outputs tell you more about quality than log-in counts alone.
