Role-Specific AI Guides Work Better Than AI Training

A nationally representative survey of over 5,000 Americans in August 2024 found that 28% of employed workers use generative AI at their job, with the most common uses being writing, information search, and obtaining detailed instructions. Those are not general AI tasks. They are the daily work of a marketer, a support agent, an operations coordinator. The training most founders give their teams does not match that reality.

Why general training fails before lunch

The 70-20-10 model from workplace learning research holds that roughly 70% of durable skill change comes from on-the-job practice, 20% from feedback with peers, and 10% from formal instruction. A two-hour AI awareness session sits entirely in that 10%. Workers leave with a general sense of what AI is, no clear answer on whether to use it for the specific email they need to write at 9am Monday, and no rule about whether pasting a customer complaint into ChatGPT crosses a line. The behavior the founder wanted does not appear, because the training never touched the behavior.

The EY 2024 Work Reimagined Survey, which covered 17,350 employees across 23 countries, found that 58% of workers who use AI at their organization rated their training programs as above average or excellent. That correlation does not prove role-specific guides caused the improvement. Organizations with high AI adoption likely invest in multiple forms of support at once. But the direction is consistent: structured guidance and sustained AI use move together.

What a one-page guide actually does differently

A role-specific guide answers three questions for one role: when to use AI, what data to keep out of AI tools, and how to check outputs before they leave the building. A marketer's guide names the specific tasks where AI is appropriate (first-draft copy, subject line variants, research summaries) and the ones where it is not (anything touching unreleased pricing, client contracts, campaign performance data). A support agent's guide tells them not to paste ticket content containing account numbers or personally identifiable information into any external AI tool, and to verify any factual claim the AI produces before sending it to a customer.

The guide does not explain how large language models work. It does not need to. The worker who spends 15 to 60 minutes per day using AI for writing tasks, per the same August 2024 survey, is not navigating the philosophy of machine cognition. She is deciding whether to use AI for one specific output. A one-page document answers that question faster than a training deck.

The objection worth taking seriously

A worker who follows a guide without understanding the underlying logic will reach situations the guide never covered. A new AI feature appears inside her existing CRM. The guide named ChatGPT and Gemini. It said nothing about this. She has no basis for deciding whether the same data rules apply.

This is a real failure mode, not a hypothetical one. The research report this piece draws on names it directly: some experts argue workers need a general understanding of AI limits, not only role-specific rules, precisely because novel situations appear faster than guides get updated. The one-page format does not resolve this on its own.

What it does do is reduce the cognitive load of the 90% of situations the guide does cover, which is the set of tasks the worker faces every day. Small business generative AI adoption nearly doubled in one year, from 23% to 40% per Federal Reserve survey synthesis. Most of those workers are not getting any structured guidance at all. A one-page guide that covers the common cases is not a complete solution. It is a better starting point than the nothing most teams currently have.

Build the guide before you schedule the training

The NIST AI Risk Management Framework and OECD due diligence guidance both address data handling and output review at the role level, not at the organization level. Governance frameworks assume someone has already decided which tasks a given role will use AI for. Most founders have not made that decision explicitly. The guide forces the decision.

Start with your highest-volume AI user. Write down what they use AI for, what data they touch in that work, and what happens if an output is wrong. One page. Ship it. Then do the next role.