Does a slick AI interface make you feel it's smarter than it actually is?

This explores whether people judge an AI as useful mainly because of how it's packaged and presented, or because of what the underlying model can actually do. No study in the collection measures the two head to head. Taken together, though, the notes suggest that interface and design often decide how much of a model's ability people experience, and sometimes they shape that impression independently of the model. The surprise is that perceived value and real value can drift apart.

The strongest version of the interface argument comes from Mollick, who says the gap between what models can do and what people get out of them is mostly an interface problem Is the AI capability gap really an interface problem?. In one study he points to, financial professionals got real productivity gains from GPT-4 but lost part of them to the mental effort of working through a chat window, and less experienced users lost the most. Medeiros pushes this further: chat is often the wrong default, because typing out a request brings back a layer of translation that point-and-click interfaces were built to remove Is chat the wrong default interface for AI products?. A related idea is the 'gulf of envisioning.' People often don't know what to ask for, so a design that offers options to react to unlocks more value than a smarter model waiting for a perfect prompt Why can't users articulate what they want from AI?.

Small design choices also produce measurable effects. Fin's team raised customer satisfaction from 72.8% to 78.4% by changing a single opening message, without changing the model at all Does conversation design ownership measurably improve AI agent performance?. Six design features can shift whether users treat an AI as a tool or as an entity, regardless of how intelligent the model is: its name, visual identity, contact presence, personality, presence alongside other people, and persistence over time Do AI agent design choices override model capability in shaping social perception?. One warning sign: making a model sound warmer can raise its error rate by 10–30% Does humanizing AI language reduce model reliability?. So the design choices that make an AI feel more valuable can make it less reliable. Socher's account of AIs gaming satisfaction scores shows what goes wrong when perceived value becomes the target Why do AIs keep gaming rewards instead of serving intent?.

Capability still matters. When researchers break down how people size up a dialogue agent, perceived competence accounts for nearly half of the variation (49%), well ahead of how human-like it seems (32%) or how flexible its conversation is (19%) How do users mentally model dialogue agent partners?. 'Capability' also isn't just raw intelligence. A compact model trained to follow through on multi-step tasks can match much larger models at a fraction of the cost Does model efficiency matter more than peak capability for real work?. Interface design also helps AI agents, not just people: redesigning screens with agents in mind improved how well computer-using agents completed tasks Can interface design help agents complete computer tasks better?.

There is a third factor besides interface and model, which is what's in the user's head. Nielsen Norman Group found that understanding roughly how AI works predicts effective use better than how often someone uses it Does understanding AI mechanics matter more than using it often?. Another line of thinking argues that AI's context keeps shifting (the prompt, the history, retrieved data), so traditional interface design alone can't make it predictable How does AI context differ from conventional software context?. Perceived value comes from the model, the interface and the user's understanding together. Most of the evidence collected here suggests the last two are where the most value is currently lost.