Anthropic on Thursday launched Reflect, a usage analytics dashboard built into Claude that lets users see what they have been asking the AI to do, how often, and during which hours of the day — and that recommends ways to work with Claude more effectively. The feature, now in beta for Free, Pro, and Max subscribers with Memory enabled, went live the day before Anthropic’s identity-verification policy took effect, and arrives while the company is managing federal litigation, a confidential IPO filing, and a mounting wave of AI-backlash coverage. For users, Reflect is a tool for deliberate AI use. For Anthropic, it is also the company’s first product feature aimed explicitly at deepening engagement rather than expanding capability.
Ryn Linthicum, Anthropic’s head of wellbeing policy and a former global mental health policy lead at TikTok, described the design intent to Engadget: “We were really intentional about building it with an eye toward how we can upskill people’s usage of Claude, not in a way that encourages them to spend more time with it.” Notably, the one metric that would most directly measure engagement — total time spent with Claude — is absent from the current dashboard. Linthicum confirmed that number is coming but said it was one the product team “didn’t want to maximise.”
How Reflect Pulls Usage Patterns From Claude’s Memory
Reflect depends on Claude’s Memory feature, which became available to all users — including the free tier — in March 2026. Memory works through extractive summarization: Claude processes your conversations roughly every 24 hours, distilling facts, preferences, and working patterns into a compressed profile stored on Anthropic’s servers. Reflect reads from that compressed store to generate its usage report.
What Reflect can show is therefore a layer of abstraction above your actual conversations. It surfaces the topics you spent time on, your most active day and peak hour, and how your tasks break down by category — coding, drafting, research, scheduling, and similar. It does not surface raw conversation content, and it cannot access the underlying files from any connected tool; if you asked Claude to summarize your inbox, the summary may appear in your reflection but the source emails would not. Conversations in incognito mode are excluded at the architecture level — Memory never captures them in the first place, so Reflect has no record of them to surface. Health-integration conversations are excluded entirely.
A time-spent view — showing cumulative hours with Claude rather than conversation counts by topic — is listed as coming soon. Reflect for Cowork conversations is also not yet available.
“Better Use Beats More Use” — and the Design Tensions That Follow
The clearest signal of Reflect’s intent is what it periodically asks users: “What’s one thing you want to keep doing yourself, even if Claude could do it faster?” Users can explore answers directly in conversation with Claude. They can also set quiet hours or schedule a nudge to take a break after a set amount of use. Both settings are dismissible at will and described by Anthropic as “reminders of your own preferences,” not enforcement.
TechCrunch’s Sarah Perez drew a direct parallel to Google’s Gmail Meter, a 2012 analytics feature that showed users how thoroughly Gmail had become embedded in their lives. That framing was not incidental: the more comprehensively a dashboard shows you that a tool is central to your work, the higher the psychological cost of switching to something else. Perez noted that Reflect’s recommendations steer users toward Claude’s own features — specifically Projects — rather than toward competitors’ tools. Anthropic does not dispute the retention logic. The company’s argument is that users who understand their own AI habits are more satisfied, more deliberate, and more likely to stay. That may be true. It is also a business proposition.
Linthicum’s timing statement, made in the same Engadget interview, was direct: “AI backlash is loud right now.” Studies warning about cognitive offloading, data center protests, and a string of lawsuits over AI-related psychological harms have all made headlines in recent months. A feature that says “use me thoughtfully” is a response to that environment as much as it is a product innovation.
What the 4D AI Fluency Framework Is — and Who Built It
Reflect’s skills layer maps users’ habits against what Anthropic calls the 4D AI Fluency Framework: Delegation (deciding when and whether to involve AI), Description (crafting effective prompts), Discernment (evaluating outputs critically), and Diligence (taking ownership of AI-assisted results). The framework traces to academic work by Prof. Rick Dakan of Ringling College of Art and Design and Prof. Joseph Feller of University College Cork — but Anthropic co-produced the framework and hosts its foundational course on Anthropic Academy.
That co-authorship matters for how users read Reflect’s feedback. The company selling the AI product also defined what good AI use looks like, and now delivers a dashboard that tells you how your habits measure up against that standard. The framework may be well-designed; the academic collaborators are credible; the dimensions themselves — delegation, description, discernment, diligence — map coherently onto what AI research recommends. But users should understand that the measurement standard was created by Anthropic, not by an independent body.
The same is worth noting about the privacy claim. Anthropic says Reflect was designed with privacy constraints built in from the start — a statement that is consistent with the feature’s architecture (no raw content surfaced; health data excluded; incognito excluded). “Privacy by design” as a framework was formalized in the 1990s by Ontario’s Information and Privacy Commissioner and incorporated into GDPR. It has no universal certification standard; Anthropic’s claim is self-assessed. The privacy architecture appears sound based on published documentation, but no independent audit of the Reflect feature has been disclosed.
Expert Context: Research on AI Dependency Supports the Design Rationale
The academic backdrop to Reflect is real. A 2025 review of negative effects of AI use on cognition found that “users who actively monitor, question, and restrict their AI use demonstrate stronger critical thinking and problem-solving outcomes than passive users.” A longitudinal controlled study by researchers from MIT and OpenAI, analyzing nearly 40 million ChatGPT interactions, found that short-term LLM use reduces loneliness but that extended heavy use correlates with emotional reliance, and that users with stronger attachment tendencies could not predict their own negative outcomes.
Research from the OpenAI/MIT team estimated that approximately 490,000 users — about 0.15 percent of the user base — show measurable signs of increasing emotional dependency. A Psychology Today columnist summarized the current consensus: the solution to AI dependency “isn’t quitting AI — it’s restoring intentional use and healthy friction.” That framing is precisely what Reflect’s design encodes.
Anthropic developed the feature with input from the MIT Media Lab’s Advancing Humans with AI program, the Digital Wellness Lab at Boston Children’s Hospital, and the Family Online Safety Institute. Linthicum led the collaboration; her background — doctoral research on suicide prevention, former global mental health policy head at TikTok — is specific expertise for a product that may surface sensitive conversation patterns.
Privacy Controls, Exclusions, and What Memory Actually Stores
Because Reflect reads from Claude’s Memory, understanding what Memory captures is essential before enabling the feature. Memory operates through extractive summarization — Claude does not store transcripts. It stores a compressed profile: your stated preferences, recurring topics, professional context, and working style, updated roughly every 24 hours. Routine Q&A exchanges are not retained; Memory captures facts and patterns the model judges worth preserving across sessions.
That compression means Reflect shows categories and patterns, not content. Users who want to see exactly what Claude has stored can ask Claude directly, or navigate to Settings > Memory to view and edit stored items. Consumer Free, Pro, and Max accounts allow memory data to be used for training by default; opting out is available at Settings > Privacy > Help improve Claude, with data purged within 30 days.
Incognito conversations are excluded from Memory architecturally — not as a policy choice but because the feature works by writing to the Memory store, and incognito mode never does. Any conversation connected to a health integration tool is excluded from Reflect’s insights entirely, at the application level.
The dashboard is accessible at [LINK: Claude Reflect Settings page] under Settings on Claude for web or the desktop app.
How to Access Reflect Right Now
Reflect requires Memory to be enabled. If Memory is off, Claude’s Settings page will indicate this when you try to generate a report. To turn Memory on: Settings > Memory (on the web or desktop app), toggle enabled. Once Memory is active and has accumulated some history, the Reflect tab in Settings will offer a report covering the past one, three, six, or twelve months.
Free, Pro, and Max subscribers are all eligible. Mobile app support and Cowork conversation analysis are not yet available.
Frequently Asked Questions
Does Claude Reflect show my actual conversation content?
No. Reflect draws from Claude’s Memory feature, which stores a compressed summary of your patterns and preferences — not transcripts of what you said. Incognito conversations are excluded because Memory never captures them. Health-integration conversations are excluded entirely. If you asked Claude to summarize your inbox, the summary could appear in your reflection, but the source emails would not. Raw file content from connected tools does not surface.
Can Reflect tell me how many hours I have spent on Claude?
Not yet. The current beta shows conversation counts, topic breakdowns, most active day, and peak hour, but a total time-spent view is still in development and listed as coming soon. Anthropic’s head of wellbeing policy said the team deliberately chose not to launch with that metric because it was not a number they wanted to optimize.
Who built the 4D AI Fluency Framework Reflect uses to evaluate my habits?
The framework was developed by academic researchers Prof. Rick Dakan (Ringling College of Art and Design) and Prof. Joseph Feller (University College Cork) — in direct collaboration with Anthropic, which co-produced and hosts the foundational course on Anthropic Academy. The dimensions are Delegation, Description, Discernment, and Diligence. Users should be aware that the standard against which Reflect evaluates their habits was co-created by the company whose product is being evaluated.
Is Reflect available on the Claude mobile app?
Not at launch. Reflect is currently available only on Claude for web and the desktop app. Anthropic has not announced a timeline for mobile availability. Cowork conversation analysis is also not yet supported and is listed as coming soon.
Enjoyed this article? Sign up for our newsletter to receive regular insights and stay connected.