Anthropic Launches Reflect: Measures How You Use Claude and Invites You to Step Back if You Overdo It
In recent hours, Anthropic has added a feature called Reflect to Claude, a panel that collects and displays how the chatbot is used, inviting users to consider whether that time is spent in line with their goals. It is available in beta immediately for Free, Pro, and Max accounts, as long as memory is enabled, and can be accessed from the settings of the web version or desktop app. It will arrive on mobile later.
The overview opens with a summary of recent conversations, and users can set the timeframe to 1, 3, 6, or 12 months. Below this, information about the most active day, peak hour, and total chats appears, followed by a breakdown of topics discussed with a percentage for the most recurring ones.
In addition to providing a snapshot of usage, Reflect offers suggestions for working better with Claude, organized around the 4D AI Fluency Framework that Anthropic developed with a group of academics: delegation (deciding whether and how to involve AI), description (formulating requests well), discernment (evaluating the quality of responses), and diligence (taking responsibility for what one does with AI). The report summarizes the tendencies in collaborating with the model and provides practical guidelines: for example, if users notice they are reconstructing the same context every time, they are advised to open a Project to group requests and avoid repetition.
From the panel, users can also set quiet hours or schedule reminders to take a break after a certain amount of usage. These reminders serve as simple nudges of personal preferences and can be ignored when engaged in tasks; the same settings are accessible from the section dedicated to time and focus.
What is Excluded from the Report
Anthropic specifies that Reflect does not draw from incognito chats or underlying files from connected tools: if a user has asked Claude to summarize incoming mail, that summary may appear, but the original emails will not. Conversations related to a healthcare integration tool are completely excluded. Topics labeled as sensitive by the company may surface, but only at a general level, and collected information remains confined to the panel without being used for other purposes.
A Panel Designed to Disconnect
The stated goal is the most curious aspect. Anthropic has long promoted Claude as a collaborative tool that helps with thinking, and Reflect is presented in the same context, inviting users to build skills that support their original thinking.
The report periodically raises questions like "What is one thing you want to continue doing on your own, even if Claude could do it faster?", then offers the chance to discuss it, indeed, with Claude. However, one metric is still missing: the actual time spent chatting. Ryn Linthicum, head of well-being policies at Anthropic, links the absence of this metric to the fact that it was not tracked internally, as it was something the product team did not want to maximize; that indication will come later, along with mobile support and extension to Cowork.
The feature, Linthicum explained, stems from a study by the Societal Impacts team in which participants expressed a mix of optimism and anxiety towards Claude and similar products. "We built the panel with great intentionality, thinking about how to grow the skills of those using Claude, not to encourage spending more time with it, but to enable being more efficient in achieving their goals and, hopefully, to step back from Claude or to preserve the matters they want to continue thinking about on their own."