In 2025, major AI answer engines including Claude, ChatGPT, and Gemini all rolled out a strikingly similar feature: they began showing users their reasoning process in real time.
The shift toward visible AI reasoning marks a significant departure from the black-box approach that dominated earlier iterations. Users now watch as engines identify which websites they search, which documents they open, and which assumptions they reconsider. Anthropic, Claude’s parent company, offers three public explanations: it helps people verify answers, it exposes gaps between reasoning and final responses, and it simply makes for compelling viewing.
Yet a fourth explanation lurks beneath the surface, one that AI providers have chosen not to highlight.
## The Labor Illusion
Back in 2011, researchers at Harvard Business School published findings in Management Science challenging the assumption that faster always equals better. Their experiment involved 266 participants using a flight search tool similar to Skyscanner or Kayak.
One group watched a plain loading spinner against a white background. Another group saw that same spinner alongside a live feed showing airline names scrolling past as fares accumulated.
The wait times varied randomly between 10 and 60 seconds.
The outcome surprised many. Users who witnessed the transparent loading process rated the website 8.1% higher in value compared to those who saw only a blank spinner. Even more striking, people maintained that preference even when the transparent version took significantly longer to deliver identical results.
A follow-up study with 118 participants confirmed the pattern. Given a choice between a faster site with no visible search process and a slower site that displayed its work, most chose the slower option.
## Effort Shapes Perception
Subsequent research reinforced these findings. In 2022, a team of researchers published work in Information and Management examining how signaling effort affects user evaluations of recommendation systems.
Their studies covered two contexts: a car search engine with 306 participants and a dating app with 294 participants. Four groups experienced different combinations of user effort and system effort. Some saw a spinner labeled “calculating results” for seven seconds. Others received instant answers.
The recommendations remained identical across all conditions.
Those who watched the loader rated the recommendation quality significantly higher than those who received immediate results. Seven seconds of visible effort changed everything.
Anthropic and OpenAI frame the transparency as entertainment. The real psychology runs deeper. People value what they perceive as effortful work. A slower answer engine that reveals its process will beat a faster one that hides it.
