Product launch · XiaoHu explainer

Anthropic adds Reflect to Claude: like Screen Time, but for your AI habits

One click in Settings for a usage report. What you talked about, when you used it most, what you asked it to do. Memory must be on.
At a glance
  • On 2026-07-09 Anthropic shipped Reflect in beta: a Settings report on how you use Claude - like "Screen Time" for AI habits.
  • Look back 1 / 3 / 6 / 12 months: key topics, when you use Claude most, task types; time spent is coming later.
  • More than charts: it asks boundary questions; you can set quiet hours or break nudges (both dismissible).
  • Uses the 4D framework to summarize how you collaborate with Claude, with concrete tips (e.g. use a Project instead of re-explaining context).
  • Free / Pro / Max with Memory on; incognito and health-related chats excluded; insights stay in Reflect. Cowork reflection is coming soon.
⚑ Based on Anthropic's blog (2026-07-09). TechCrunch / Axios coverage exists; facts follow the official post.
1The problem

You already had this problem - nowhere to look

A real scene

You've used Claude for months. A friend asks: "Are you too dependent on AI?"

You open your mouth - and can't answer. You know you chat a lot, but not what for, when it spikes, or whether you handed off thinking you should still own.

In user interviews, Anthropic kept hearing the same questions: How should AI fit into daily life? How much is enough? When is AI right for a task - and when should a human keep it?

Before · blind use
  • Only chat logs, no overview
  • "I use it a lot" - can't verify
  • Faster work, blurrier boundaries
  • Want to change habits - no starting point
Now · Reflect
  • A readable usage report
  • Topics, hours, task types laid out
  • Questions that force boundary talk
  • Concrete collaboration tips by dimension

Phones already have Screen Time: see first, then decide whether to cut back. Reflect does the same for how you live with Claude. It does not judge "you're addicted" - it hands you a mirror.

2Mapping

Three hard questions - how the report answers

Click each question to see which part of Reflect answers it. All three titles stay visible - nothing important is hidden; the click only lines up problem → capability.

Click a question to see the matching capability ↓
How the report answers

It breaks down when you use Claude most. Later it will add total time. In the dashboard you can set quiet hours or a break nudge after heavy use (both dismissible).

Seeing "still chatting hard at 2 a.m." often works better than a lecture on moderation.

How the report answers

The top of the report covers key topics + task types. Switch 1 / 3 / 6 / 12 months to compare a recent sprint with longer-term habits.

If 80% of time is rewriting email and almost none is strategy - that's data, not moral judgment.

How the report answers

It periodically surfaces questions. The official example:

What's one thing you want to keep doing yourself, even if Claude could do it faster?

You can talk that through with Claude. The 4D summary also shows patterns like reworking drafts in your own voice, or only delegating after you set the strategy - those are the parts you still keep.

3What it looks like

Open a report and you'll roughly see this

Path: Claude web or desktop → Settings → reflect on usage → generate. Screenshots below are official.

Usage overview
Official · overview: topics, patterns, common tasks · Anthropic
When you use it
Official · when you use Claude most · Anthropic
What you work on
Official · what that time is spent on · Anthropic
4Next step

From seeing to working better

Reflect also scores collaboration with Anthropic's 4D AI Fluency Framework - in plain language:

Delegation
Should you hand it off?

Official example: do you settle strategy first, then give Claude the work - or open with "just handle it"?

Description
Did you say it clearly?

Fuzzy goals → fuzzy output. The report can show if you keep re-explaining and re-prompting.

Discernment
Can you judge the result?

Reworking email drafts into your voice means you're still filtering - not rubber-stamping.

Diligence
Who owns the outcome?

After AI helps, whose is the final send? The framework keeps responsibility with the human.

4D summary
Official · collaboration summary by 4D · Anthropic
One concrete tip

Official example: if you re-explain the same project every chat, the report may suggest a Project so long-term context lives in one place. More useful than "get better at AI."

5Boundaries

What enters the report - and what does not

  • Incognito chats stay out
  • If Claude summarized your inbox, the report may note that - not the raw emails
  • Chats tied to a health integration are fully excluded
  • Sensitive topics may appear only at a high level; built with MIT Media Lab AHA, Boston Children's Digital Wellness Lab, Family Online Safety Institute
  • Insights stay in Reflect; Anthropic says they aren't used for other purposes
6How to open it

Three steps

1

Turn Memory on

Without Memory, Reflect has nothing to compute - no report.

2

Generate from Settings

Web or desktop → Settings → reflect on your usage → generate.

3

Pick a range and act (or not)

1 / 3 / 6 / 12 months. Quiet hours and break nudges are optional.

Available on Free, Pro, Max (consumer). Cowork reflection is "coming soon." Team/Enterprise: check your Settings page.

7Who cares
Heavy daily users

Is this good tool use - or dependence? Start with hours and task mix.

People setting team norms

4D examples help debate what must stay human vs what AI can take.

Privacy-conscious users

Read incognito / health / raw-file boundaries before turning Memory on.

Always re-explaining context

If the report flags that, a Project often helps immediately.

Source: Anthropic · A new way to reflect on how you use Claude (2026-07-09). Screenshots from the official blog.