Claude Reflect: How to Use Anthropic's New Monthly Recap to Level Up Your AI Skills

Anthropic just shipped something most AI tools have never bothered to build: a feature that tells you how well you actually use the product. Claude Reflect, rolled out this week to Settings > Reflect on web and desktop, generates a recap of your Claude activity and — more usefully — grades your habits against Anthropic's own 4D AI Fluency Framework. It's part "Spotify Wrapped," part skills audit.

For anyone treating AI fluency as a real, measurable skill rather than a vibe, this is the first built-in tool that turns your actual conversation history into a diagnostic. Here's what Reflect does, how the scoring works, and how to use it to find the specific gaps that are costing you time — or points on a certification exam.

What Claude Reflect Actually Shows You

Reflect lives at Settings > Reflect in Claude's web and desktop apps, currently in beta on Free, Pro, and Max plans. You can generate a summary covering the last 1, 3, 6, or 12 months, and it surfaces four things:

The topics you actually spent time on — named specifically, not "you used Claude for coding." Reflect pulls the real themes from your conversation history: the project you debugged for three weeks, the certification you were cramming for, the report you kept rewriting.
Your most active day and peak usage hour — useful for spotting whether you're doing your best thinking at 11pm the night before a deadline (a pattern worth fixing).
Behavioral observations — Reflect notices things like whether you tend to accept the first answer Claude gives you, or whether you push back and iterate.
A 4D Fluency breakdown — your interactions get categorized across delegation, description, discernment, and diligence (more on this below).

Anthropic built this on top of the memory feature that shipped earlier this year, and it inherits the same trigger: there's no separate on/off switch for the recap. If Generate memory from chat history is enabled in Settings > Capabilities, Reflect is enabled too. Turn memory off and the recap disappears with it.

Alongside Reflect, Anthropic also added Settings > Time and focus, which lets you set break reminders and quiet hours, and occasionally surfaces a prompt like "What's one thing you want to keep doing yourself, even if Claude could do it faster?" — a small nudge against outsourcing judgment along with the busywork.

The 4D AI Fluency Framework, Explained

Reflect's scoring isn't arbitrary — it's built on a framework Anthropic has been developing for a while through its AI Fluency course (free on Anthropic's Skilljar platform). The framework defines fluency as four distinct competencies, and understanding them is the difference between reading your recap as a novelty and actually using it:

Delegation — deciding whether, when, and how to hand a task to Claude in the first place. High delegation fluency looks like designing a workflow (define the steps, checkpoints, and review points) rather than just dumping a task and hoping. Low fluency looks like asking Claude to "handle it" with no structure and being surprised by the output. Description — how well you communicate context, constraints, and goals in your prompts. This isn't about clever prompt tricks; it's about giving Claude enough of the why behind a request that it can make good judgment calls on the how. Discernment — your ability to critically evaluate what Claude gives back. This is the competency most people skip. A confident, well-formatted answer is not the same as a correct one, and discernment is the muscle that catches the difference before it ships to production or shows up on an exam answer sheet. Diligence — whether you're staying accountable for the output: checking sources, disclosing AI involvement where it matters, and not letting speed erode responsibility.

Description and discernment are explicitly linked in Anthropic's framework: description is what you tell Claude you want, discernment is how you judge what you got back. If your recap shows strong description but weak discernment, that's a specific, fixable pattern — not a vague "use AI better" verdict.

How to Read Your Recap Without Getting Defensive About It

The instinct when a tool tells you how you work is to either dismiss it or take it as a personality verdict. Neither is useful. Treat it as a diagnostic:

Open Settings > Reflect and generate the 3-month view — long enough to show patterns, short enough to still be recent.

Read the named topics first. If Reflect surfaces something like "debugging a Next.js deployment issue" three separate times, that's not a coincidence — it's a recurring gap in either your own knowledge or your workflow setup.

Look at the 4D breakdown for the weakest category. Most people skew toward strong description and delegation (they're good at asking) and weaker on discernment and diligence (they're worse at checking). If that's you, the fix is concrete: before accepting Claude's next output, ask it to argue against its own answer.

Cross-reference peak hour with output quality. If your busiest Claude hour is also your latest hour, you're doing your most delegation-heavy work when your own discernment is at its weakest. That's a scheduling fix, not an AI fix.

Repeat monthly. A single recap is a snapshot. Three recaps in a row are a trend line.

Why This Matters More If You're Studying for an AI Certification

If you're preparing for the Claude Certified Architect (CCA-F) or any AI certification that tests practical AI collaboration skills, Reflect is effectively a free diagnostic for the exact competencies those exams probe. CCA-style exams don't just test whether you can write a prompt — they test whether you know when not to delegate a task, whether you can spot a hallucinated citation, and whether you understand the accountability tradeoffs of shipping AI-generated work.

Those are delegation, discernment, and diligence, respectively. If your Reflect recap shows you're weak on discernment — accepting Claude's first answer without pushback — that is very likely the same gap that will cost you points on scenario-based exam questions that ask "what's wrong with this AI-generated output?"

Use your recap as a study plan input: whichever of the four Ds shows up weakest, deliberately drill that competency for a week before you touch another practice question.

Privacy: How to Turn Reflect Off

Because Reflect runs on the memory pipeline, disabling it is a one-step process: go to Settings > Capabilities and turn off Generate memory from chat history. This disables both future memory generation and the recap feature tied to it. There's currently no way to keep memory on while opting out of just the recap, so if you want memory's context-carrying benefits without the usage analysis, you'll need to wait for Anthropic to split the toggles — or accept the tradeoff for now.

Frequently Asked Questions

Is Claude Reflect available on all plans?

It's in beta on Free, Pro, and Max plans, on web and Claude Desktop. There's no indication yet of a Team or Enterprise rollout timeline, though given Anthropic's pattern with other consumer-first features (like memory), that typically follows within a few months.

Does Reflect work on the Claude mobile app?

As of this rollout, Reflect is limited to Settings on web and desktop. Mobile access to Reflect specifically hasn't been announced, separate from the broader Claude Cowork mobile expansion that shipped the same week.

Will Reflect read my private conversations to build the recap?

Reflect works off the same memory pipeline Claude already uses to carry context between chats — it's not a new data collection mechanism, but a new lens on data Claude already has access to if memory is on. If you're not comfortable with that, turning off "Generate memory from chat history" removes both.

Can I export or share my recap?

Anthropic hasn't announced an export feature for Reflect recaps at launch. For now, it's a Settings-page view only, though this is the kind of feature that commonly gains a share or export option after beta feedback.

How is this different from just asking Claude "how have I been using you lately"?

Asking Claude directly gives you an answer generated in the moment, based on whatever it can recall or infer from the current conversation. Reflect is a structured, purpose-built analysis run against your actual usage history and scored against a fixed framework — closer to a real audit than a conversational guess.

Key Takeaways

Claude Reflect (Settings > Reflect) is a new beta feature on Free, Pro, and Max that generates a monthly recap of your Claude usage, including named topics, peak activity times, and behavioral patterns.
Recaps are scored against Anthropic's 4D AI Fluency Framework: delegation, description, discernment, and diligence.
Discernment — critically evaluating AI output rather than accepting it — is the competency most users score weakest on, and the one exam questions on AI certifications most often probe.
Reflect requires memory to be on (Settings > Capabilities); there's no separate toggle for the recap alone.
Use three consecutive monthly recaps to spot a trend, not just a snapshot, and turn your weakest D into a deliberate one-week drill.

Next Steps

Reflect will tell you where your AI fluency gaps are. Closing them takes deliberate practice — the same kind that AI certification exams reward. If discernment is your weak D, AI for Anything's CCA-F Practice Test Bank is built specifically to drill that skill: every question includes distractor answers designed to look plausible, forcing you to evaluate rather than guess. Run your next Reflect recap after a week of deliberate practice and see if the pattern actually moves.