Why do people trained on AI tools report bigger productivity wins than those left to figure it out alone?

This asks why people trained to use AI say they get more out of it than untrained workers, and whether that gap is real skill, better-designed jobs, or a quirk of self-reporting. The collection has no head-to-head study of trained versus untrained users. It does have pieces that point to a different answer than 'training teaches you the tool.' The best clue is a Workday survey of 3,200 AI users Where does AI's time savings actually go in practice?. Most saved hours each week, but nearly 40% of those savings went into fixing errors and checking outputs. Only 14% came out consistently ahead, and those people worked at organizations that retrained staff and redesigned roles. Training seems to work less as a lesson in prompting and more as a way of reorganizing work so that AI's mistakes don't eat the time it saves.

That fits an argument about where AI actually speeds things up. Narayanan and Kapoor split knowledge work into deciding, executing, and delivering, and say AI mainly shrinks the middle step Does AI really compress all layers of knowledge work equally?. Untrained workers tend to push AI output straight through, so the deciding and checking still fall on them, unplanned. Trained workers have usually been shown where to put AI and where to check it. A BCG survey shows what happens without that structure Does using more AI tools always boost worker productivity?. Self-reported productivity rose with up to three AI tools, then fell sharply. The cause was the burden of overseeing everything, not the tool count itself.

There's a twist. Training may help most when it builds on skill people already have. One finding is that AI gains appear when workers apply skills they already know, and disappear when they use AI to learn something new When does AI actually boost worker productivity?. So 'trained users' may simply be people using AI inside their own expertise. Yet in a study of 5,172 customer-support agents, the least experienced gained the most (about 15% on average, concentrated among novices) Does AI assistance help less experienced workers most?. The most experienced saw small speed gains and slightly lower quality. These two findings don't conflict. AI helps novices do a job they've been placed in, but it doesn't teach them a new one.

The word 'report' in the question matters too. Self-reported gains are unreliable. Executives perceive bigger AI gains than the numbers show Do AI productivity gains feel larger than they actually measure?. Smooth, polished AI output can also make users feel more capable than they are, because people judge their own skill by how easy the work felt Does processing ease mislead users about their own competence?. Training may also make people more willing to talk about their gains. An Anthropic interview study found that about 70% of workers hide or downplay their AI use because of workplace stigma Why do workers hide productivity gains from AI use?. Untrained workers in workplaces that don't openly endorse AI may be underreporting, not underperforming.

The takeaway: some of the 'training gap' is probably real, but it comes from redesigned roles and less rework, not from better prompting. Some of it may be a reporting effect, because sanctioned, trained users have both permission and reason to report gains. The collection can't say how much of the gap comes from each.