"Get everyone trained on AI" sounds like one request. It is usually four, and only one of them is training.
Read this if a sponsor has asked you to "get everyone trained on AI." Skip it if you already know the tool, the rules and the task.
The short version. People use AI at work only when four doors are open: access, permission, a real task, and skill. Training opens only the last one. Check the first three on the call.
4-minute read
The request sounds simple: "Get our people using AI." Before you scope a course, walk it through four doors, in order.
Only door 4 is a training problem. If an earlier door is shut, a course walks people up to it and leaves them there.
Gallup asked US employees who don't use AI in their role why not. 44% said the main reason is they don't believe AI can help with their work. [1] 16% said they lack access. Only 10% named not knowing how to use the tools, and 8% said they feel unsafe using them. [1]
So a skills course aims at the smallest group. The biggest group is stuck at door 3: nobody has shown them a task in their own job.
Rules matter too. In Gallup's 2026 survey, 68% of employees who strongly agree their organisation has clear AI policies are frequent users. For everyone else, it's 47%. [2]
Fred Davis measured two things that might predict whether people use a new system at work: how useful it seems, and how easy it seems. [3] In both of his studies, "usefulness had a significantly greater correlation with usage behavior than did ease of use." [3]
He went further. His analysis suggests ease of use may matter mainly because it makes a tool look useful. [3]
Most AI courses teach ease: prompts, features, shortcuts. Davis says lead with usefulness: one task, in their week, done better. That's door 3.
BCG surveyed 11,749 people for its June 2026 AI at Work report. [4] Among regular AI users with a clear AI strategy but limited tool access, 80% reported measurable impact. With strong tool access but little clarity, it was 60%. [4]
Training still matters. Only 36% of respondents feel they've had enough of it. [4] The point is order, not either-or. Open doors 1 to 3, and the training lands.
"Use AI to be more productive" isn't a task. It's an adjective with a tool attached.
Run the Did Swap. Ask about the last time someone used AI well on real work. What did they actually do? If nobody can name one, door 3 is shut, and that's the finding.
An illustrative composite, not a client.
Before: "Train all 300 sales staff on AI by Q1." Plan: a two-hour prompting workshop. Measure: attendance.
After the Four Doors:
| Door | What the call found | Who owns it |
|---|---|---|
| 1 · Access | Only 120 of 300 have a licence | IT, before any training |
| 2 · Permission | Nobody knows if client names can go in | Legal: a one-page rule |
| 3 · Use | One task: the follow-up email after a sales call | Sponsor confirms it |
| 4 · Skill | A 45-minute practice on that task, using cleaned real emails | L&D |
Measure: hours from call to follow-up email, before and after. The first plan trains 300 people to stand at a locked door.
When door 1 or 2 is shut, say it plainly: "Training can't fix this. Who decides it, and by when?" That isn't refusing the work. It names what must be true for the work to pay off.
Managers matter at every door. In Gallup's 2026 data, 78% of employees whose manager actively supports AI use are frequent users. For the rest, it's 44%. [2] Put the managers' part in the brief, not just the learners'.
If most of the request sits behind closed doors, here is how to say training may not be the answer.
Half of BCG's regular AI users who know about agents say their company hasn't set clear guidance on managing teams of people and AI. [4] So "train them to work with agents" is coming, and door 2 will be the one that's shut.
Ask it first: what may people hand to an agent, and who checks its work?
Want to rehearse the whole conversation before a real call? The free Diagnostic Conversation Coach walks you through it. No signup, no email.
The Four Doors · copy this
When the request is "train them on AI", ask in this order. Stop at the first closed door. 1 · ACCESS: "Which tool are they allowed to use? Can every one of them open it today?" No → IT decision. Not training. 2 · PERMISSION: "Can client or company data go into it? Who decided, and where is it written?" No → Policy decision. Not training. 3 · USE: "Name one task in their week you want done differently. Who did it well last time, and what did they do?" No → Sponsor decision. Find the task before you design. 4 · SKILL: "When someone tries that task with the tool, where do they get stuck?" Yes → Now it's training. Design practice on that one task. WRITE THE BRIEF LINE: When ________ [task], ________ [who] will use ________ [tool] to ________, so that ________.
[1] Kemp, A. (8 November 2025). Manager support drives employee AI adoption. Gallup. gallup.com. Gallup Panel, 19,043 employed US adults, May 2025; reasons asked of those who do not use AI in their role. Read 29 October 2026.
[2] Kemp, A. (12 April 2026). Gallup workplace research on AI adopters and holdouts. gallup.com. 23,717 employed US adults, February 2026. Read 29 October 2026.
[3] Davis, F. D. (1989). Perceived usefulness, perceived ease of use, and user acceptance of information technology. MIS Quarterly, 13(3), 319–340. doi.org/10.2307/249008. Abstract read 29 October 2026.
[4] Boston Consulting Group (June 2026). AI at Work: Strategy Matters More Than Tools, fourth edition. bcg.com (PDF). Survey of 11,749 people; the impact comparison is among 9,923 regular AI users. Read 29 October 2026.
The before/after is an illustrative composite, not a client conversation. The Four Doors is our own sorting test; none of the sources above names it.