Does understanding AI mechanics matter more than using it often?

Tanner Kohler's Nielsen Norman Group (NN/g) study guide argues that how well a user understands the basic mechanics of AI — not how often they use it — governs whether their use of it is effective. The guide states the claim directly: "the better you understand some of the basics, the more accurate your mental model is. Accurate mental models make you more efficient and help you avoid problems." It then connects model accuracy to adoption: "When users have inaccurate mental models of how the system functions or lack a larger vision of its potential, their usage of it can be underwhelming," immediately followed by a cited figure — "Out of ~5,000 working Americans, 81% reported they did not use AI at work." The guide closes the point by separating literacy from mere exposure: "AI literacy requires both prompt fluency and the ability to critically assess outputs; frequent use does not imply skilled use."

The reasoning Kohler gives is a vocabulary problem compounding a comprehension problem. He notes that terms like "neural networks, unsupervised learning, hallucinations, and sycophancy" circulate "like we all know what they are and why they matter," and the guide exists to translate this vocabulary "into normal language" so users build working models of actual mechanics — training phases, why hallucination is structurally hard to eliminate, why models are built to prioritize pleasing users over accuracy — rather than picking up surface-level usage habits. The guide is explicitly a curated index to other explainer resources, not an original study; the source, sample, and methodology behind the 81% figure are not given within this excerpt.

This sits alongside Why do people trust AI outputs they shouldn't?, which also traces interaction failure to a gap between a user's model of AI and what the system is actually doing, but specifies three compounding cognitive traps rather than a single accurate/inaccurate axis. It extends Why do users fail with AI interfaces designed like conversations?: both locate the source of underwhelming AI use in a mismatch between what users assume about the system and how it actually works, with Kohler adding a concrete (if unsourced) adoption statistic to that claim. It also stands in tension with Does chatbot language style actually shape how much we trust it?, which shows trust forming from expertise-sounding language regardless of genuine understanding — the opposite route to confidence from the one Kohler's guide prescribes.

The excerpt does not establish a causal link between mental-model inaccuracy and the 81% non-use figure; the two claims are presented adjacently but the figure's source, sampling method, and definition of "use AI at work" are absent here. Nor does the guide measure what counts as an "accurate" mental model or how literacy should be assessed. What it supports, at the strength the excerpt allows, is a design and communication implication: products and explainers that leave users' mechanical understanding of AI vague are leaving a known adoption lever unpulled, whether or not the particular 81% figure generalizes.