Anthropic built a dashboard that tracks how much you use its own AI, then gently asks whether you should be using it less.
Anthropic quietly shipped one of its stranger features on July 9, 2026. It is called Reflect, and it lives inside Settings on Claude for web and desktop. Turn it on, and Claude hands you a report on your own habits: which topics you keep coming back to, your busiest day, your peak hour, and a breakdown of the kinds of tasks you actually bring to it.
Here’s the thing. Most AI companies want you glued to the screen. Anthropic built a tool that tells you exactly how glued you already are, then offers to help you use less of it. That’s either a genuinely useful bit of self-awareness tooling or a clever way to make Claude feel indispensable, depending on who you ask. It is probably both.
This post breaks down what Reflect actually shows you, where the privacy lines sit, and the specific steps you can take this week to turn your own report into a sharper, faster way of working with Claude.
What Reflect actually shows you
Open Settings, find the Reflect tab, and pick a window: past 1, 3, 6, or 12 months. Claude generates a short written summary of what that stretch looked like for you, followed by a few concrete numbers.
- Your most active day of the week
- Your peak hour of usage
- Total number of conversations
- A percentage breakdown of the topics you discuss most
Below that sits the part worth paying attention to: a breakdown of your habits against Anthropic’s 4D AI Fluency Framework, covering delegation, description, discernment, and diligence. If that framework is new to you, we already broke it down in detail here, and it’s worth a read before you dig into your own report, because Reflect assumes you already know what each of the four terms means.
Reflect doesn’t just label your behavior. It reacts to it. If it notices you keep re-explaining the same background across separate chats, it suggests starting a Project so you stop repeating yourself. According to one early reviewer’s report, it even proposed building a custom fact-checking Skill for them, then wrote a working template that made Claude cite its sources and flag its own confidence level. If Skills are new territory for you, we covered how to build them step by step here, and Reflect is arguably the first Anthropic feature that actively pushes people toward using them.
What Reflect can see, and what it can’t
Before you generate a report, it helps to know what data actually feeds it.
- Incognito conversations are excluded entirely.
- Reflect can summarize what you did with a connected tool, but not the underlying files. Ask Claude to summarize your inbox, and that summary might show up in your report. The emails themselves won’t.
- Any conversation linked to a health integration is left out completely.
- Sensitive topics can still surface, but only at a high level, never word for word.
- Anthropic says the data stays inside the Reflect experience and isn’t reused anywhere else, including for model training.
Anthropic built this 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, according to Anthropic’s own announcement. That’s a fairly serious lineup for what is, on paper, a usage dashboard.
Why Anthropic actually built this
The framing Anthropic uses internally is efficiency, not restriction. Ryn Linthicum, who leads wellbeing policy at the company, told Dataconomy that the goal was building the dashboard around helping people get sharper with Claude, not around getting them to spend more time inside it.
Not everyone is buying the framing at face value. TechCrunch’s take is that laying out everything Claude has done for you, in one tidy report, makes the tool harder to give up, not easier. Nudging people toward Projects deepens the exact habit loop the feature claims to interrupt. Both readings can be true at once. A wellbeing feature and a retention feature aren’t mutually exclusive, and Reflect is probably a bit of both.
Whatever the motive, the mechanics are genuinely useful if you use them right. That’s the part most of the coverage skipped past.
Step-by-step: generate your first report
- Turn on Memory. Reflect needs it to work. If you keep Memory off, open Settings, find Memory, and switch it on before continuing.
- Open Settings and select the Reflect tab. It sits under a section Anthropic calls Time and Focus on both the web and desktop apps.
- Pick your time window. Start with the last month if this is your first report. It’s the fastest way to check whether the summary actually matches how you remember using Claude.
- Read the written summary before the numbers. It’s the part most people skip straight past, and it usually surfaces patterns the raw stats don’t.
- Scroll to the 4D Fluency breakdown. Note which of the four categories, delegation, description, discernment, or diligence, has the thinnest detail. That’s usually the skill worth working on next.
- Act on at least one suggestion the same day. If Reflect flags repeated context, start a Project immediately. If it flags a repeated task, ask Claude to help turn it into a Skill.
- Set quiet hours or a break nudge if that’s useful to you. Both live under Time and Focus, and both are optional and dismissible mid-conversation.
- Re-run the report in a month. Compare the new 4D breakdown against your first one. That comparison is the real signal your AI habits are improving, not your raw chat count.
Turning your report into real efficiency gains
The dashboard does the diagnosis. It doesn’t do the fix. That part is on you, and it comes down to a short list of moves mapped to each of the four dimensions.
| Dimension | What it measures | What to do if Reflect flags it |
|---|---|---|
| Delegation | Whether you’re deciding what to hand off to Claude and what to keep doing yourself | Before every task, ask whether it needs your judgment or just your time |
| Description | How clearly you explain goals and context | Start a Project so context persists instead of retyping it every session |
| Discernment | Whether you actually check Claude’s output before trusting it | Build a fact-checking Skill that forces citations and a confidence flag |
| Diligence | Whether you take responsibility for what you ship with AI help | Review, don’t just accept, anything going out under your name |
None of these fixes require a paid plan or a new tool. They just require reading your report instead of closing the tab the moment it loads.
Honest limitations
- No time-spent metric yet. Anthropic says it never tracked this internally on purpose, and the feature to show it is still being built.
- Web and desktop only, for now. Mobile support and visibility into Cowork sessions are both listed as coming soon, not live yet.
- It’s beta, and it shows. Early users on MacRumors Forums reported identical summaries showing up across different time windows, and activity counts that didn’t line up with how often they remembered using Claude.
- Memory has to be on. If you keep Memory off for privacy reasons, and plenty of people do, Reflect simply won’t generate a report.
The most useful feature Anthropic shipped this month might be the one that quietly asks you to use it less.
Sources
- Anthropic, “A new way to reflect on how you use Claude” (official announcement)
- TechCrunch, “Anthropic’s new Claude feature is quietly selling you on AI”
- Engadget, “Claude’s new Reflect dashboard wants to help you log off of Claude”
- TheNextWeb, “Claude Reflect: Anthropic’s ‘Claude Wrapped’ wants you to log off”
- Dataconomy, “Claude Gets Reflect Dashboard To Track AI Usage”
- Claude Help Center, Release Notes
Just Being Resourceful! 
Leave a comment