Why a Weekly Review Is a Good Use of AI, and How to Actually Run One
What a generic AI-assisted weekly review actually catches and misses, four prompts that sharpen the review today, and what changes when the AI applies Cal Newport's documented deep-work planning discipline instead of generic reflection prompts.
Ask AI to help you run a weekly review and the default output is a tidy summary: here's what you accomplished, here's what's outstanding, here's a suggested plan for next week. It reads like progress. It's also mostly descriptive, a recap of what happened, with very little forcing you to judge whether what happened actually mattered.
The more useful review isn't a summary of your week. It's a specific check on whether your actual time matched what you'd decided mattered, and what to do differently if it didn't, run the same way every week regardless of whether the week felt good or bad in the moment.
What comes back when you ask AI to summarize your week?
Describe your week to an AI and ask for a review, and you'll typically get a well-organized recap: tasks completed, items still open, a reasonable-sounding plan for next week. It's accurate as a summary and largely silent on the actual question a review should answer, whether the time spent this week matched what you'd decided mattered before the week started.
Summarizing what happened and judging whether it should have happened that way are different tasks. A generic review prompt defaults to the first because it's the easier, more directly answerable one from whatever account of the week you give it, and it requires nothing uncomfortable of you in the process.
Where does AI actually help in a weekly review, and where should you check it?
Unaided, it's genuinely useful for structuring a review against a fixed set of questions, so you're checking the same things every week rather than whatever feels salient in the moment, and for catching inconsistencies in your own account, a stated priority that got zero actual time, for instance.
It's weak at knowing anything you didn't tell it. If your account of the week is generously edited, more focused time claimed than actually happened, an interruption downplayed, the AI has no independent way to catch the gap, since it has no calendar, no inbox, and no visibility into the week beyond your own description of it. The review is only as honest as your input.
What's worth running through AI before Sunday night?
These four run in Claude, ChatGPT, or Gemini as written.
1. The protected-hours audit. "Here's roughly how my week broke down: [describe it]. How many hours would you estimate were genuinely protected, focused time versus fragmented into meetings, messages, and interruptions? Be specific about where the fragmentation happened." Why it works: naming an actual, if rough, number forces a concrete accounting rather than a vague sense of having been "pretty busy," which is the vaguer, less useful read most people default to.
2. The misallocation check. "I said [stated priority] mattered most this week. Based on what I've told you about how the week actually went, did it get the time that statement implies it should have? Where did the gap come from?" Why it works: this directly tests the gap between stated priority and actual time allocation, which is usually where a week quietly goes wrong without anyone noticing until it's repeated for a month.
3. The carry-forward filter. "Of everything still open from this week, which items are actually still worth doing, and which have been carried forward multiple times without ever becoming urgent enough to finish? Be honest about the second category." Why it works: an item that's been carried forward for three weeks without becoming urgent is usually a candidate for dropping entirely, and naming that pattern explicitly is easier than noticing it buried in a growing list you've stopped actually reading top to bottom each week.
4. The next-week commitment. "Given this week's actual pattern, what's the one change to how I protect time next week that would matter most? Not a list, one specific change." Why it works: forcing a single priority change, rather than a list of good intentions, produces something that's actually more likely to survive contact with next week's inevitable pressure, since a list of five good intentions rarely survives past Tuesday.
What hasn't a single review fixed for good?
This exercise will make this week's review sharper and more honest than a generic recap. It won't give you an operating rhythm that survives a genuinely brutal week, the one where skipping the review entirely feels justified because there's no time for it, which is usually exactly the week it would have mattered most. One good review isn't the same as the discipline of running the same review every week regardless of how the week went.
How does Cal Newport's framework actually change what gets reviewed?
Cal Newport, documented across Deep Work and subsequent writing, argues that focused, high-value work is a scarce resource that has to be actively protected on a calendar, not a mental state you either summon through willpower or don't. His broader planning practice treats time blocks, not to-do items, as the actual unit worth reviewing, because a task list says nothing about whether the hours around it were protected or constantly fragmented.
Before, generic prompting: asked to review a week that felt busy but unproductive, a generic AI response lists what got done and suggests better prioritization next week, treating the problem as a task-selection issue.
After, Newport's framework applied: the framework asks a different first question entirely: not what got done, but how many hours were actually protected as deep, uninterrupted blocks, and what specifically consumed the rest. If the honest answer is that almost no time was protected, meetings and messages filled the calendar entirely, the framework's documented diagnosis is that no amount of better task prioritization fixes a week with no protected time to execute against; the actual fix is calendar-level, blocking specific hours before the week starts, not a smarter to-do list.
A generic "review my week" prompt has no built-in reason to ask that structural question, were hours actually protected, before any task-level review even starts. The framework asks it anyway, which matters more than a better-organized task list ever would.
How does this actually reach Claude, ChatGPT, or Gemini?
Claude installs it as a Skill: Settings → Skills → Add skill, upload the .zip. It auto-invokes once your question matches the skill's description, or gets forced directly with /cal-newport-framework. In Claude Code, the package sits in ~/.claude/skills for every project, or a project's own .claude/skills folder for just that one.
ChatGPT has no Skills equivalent, so the plain .md becomes a Custom GPT's Instructions: Create a GPT → Configure → Instructions, pasted in directly. Skip the default Custom Instructions fields, they cap at 1,500 characters; a Custom GPT's Instructions field runs to roughly 8,000, enough for the full framework.
Gemini has no dedicated upload option, so the content goes into the system prompt if your interface has one, or opens the first message instead: "Here is a thinking framework to apply throughout this conversation. Read it carefully, then answer using this framework's approach," with the framework text pasted below it.
Where does this lead next?
If a weekly review is a habit you're trying to actually build rather than run once, Founder Weekly Review is a $129 tool built to guide exactly that cadence, week after week. Newport's documented method is one of several in the Performance & Mindset category, alongside James Clear and Jim Collins, each an .md file for Claude, ChatGPT, or Gemini.
Frequently asked questions
Isn't a weekly review just journaling with extra steps?
Journaling is open-ended reflection; a structured weekly review is a specific, repeatable operating procedure, what shipped, what didn't and why, what's actually the priority next week, run the same way every week. The distinction matters because journaling can drift into whatever feels salient that day, while a real review checks the same fixed set of things regardless of mood, which is what makes it useful for catching a pattern rather than just processing a feeling.
Why does my AI-assisted review always end up feeling like a status report?
Because a generic "summarize my week" prompt produces exactly that, a description of what happened, with no forcing function toward judgment about what it means. The fix is asking specifically evaluative questions, was this week's use of time actually aligned with what mattered, not just descriptive ones like what did I do, which most default review prompts fall back to without you asking for it.
What's the biggest mistake people make running a weekly review with AI?
Treating it as a summarization tool instead of a diagnostic one. Feeding it a list of completed tasks and asking for a recap produces a recap; asking it to identify the gap between planned deep work and what actually got protected produces something you can act on next week. The value is in the second question, not the first, and most people never get past the first.
Do I need to track my time carefully all week for this to work?
Some tracking helps, but a rough honest account is enough to start. Even an approximate list of what got real focused attention versus what got fragmented into meetings and interruptions is enough for the review prompts here to surface something useful; precision matters far less than being honest about the fragmented parts of the week you'd rather not admit to when describing it.
How is 'protecting deep work' different from just saying I need to focus more?
Cal Newport's documented argument is specific: deep work is a scarce, schedulable resource, not a state of willpower you either have or don't, which means the actual unit of analysis is hours protected on a calendar, not a vague intention to concentrate harder. A weekly review built around this asks how many hours were actually protected and what specifically interrupted them, not whether you felt focused.
Can this work for someone who doesn't have knowledge-work deep tasks, like a manager whose week is mostly meetings?
The framework adapts rather than becoming irrelevant. For a meeting-heavy week, the equivalent question becomes which meetings actually required your specific judgment versus which could have been handled by someone else or skipped entirely, and Newport's underlying point, protect scarce high-value attention deliberately, still applies even when the high-value unit is judgment exercised in a meeting rather than solo focused work at a desk.
Should the weekly review look backward at last week or forward to next week?
Both, and doing only one is the most common way a review becomes useless. A review that only looks backward turns into a status report with no consequence; one that only plans forward without checking last week's actual pattern repeats the same misallocation every week without ever noticing it's a repeat. The four prompts here deliberately alternate between the two directions rather than picking one.
What's the honest limit of using AI for a weekly review?
It has no independent knowledge of what actually happened this week beyond what you report, so a dishonest or overly generous self-account produces a diagnosis that's equally generous and equally useless. Use it to structure the questions and catch inconsistency in your own account; the honesty about what actually happened, including the parts you'd rather gloss over, still has to come from you.
Written by Gareth Hoyle. Last updated 26 September 2026. Part of the authority.md guides library.
More guides.
Bringing AI Into a Creative Project Without Ruining It
What generic AI brainstorming actually gives a creative project and what it quietly flattens, four prompts that help today without taking over, and what changes when the AI applies Rick Rubin's documented listening practice instead of generic idea generation.
AI for Hiring Decisions: Where Claude and ChatGPT Actually Help
A grounded look at what generic AI use gets right and wrong when you're deciding on a hire, four prompts that genuinely sharpen the decision, and what changes when the AI applies a documented evidence-based hiring framework instead of generic interview advice.
Using AI to Prep for a Negotiation: What Actually Works
Generic AI negotiation advice is genuinely useful and genuinely limited. Here's what Claude and ChatGPT get right before a negotiation, four prompts that work today, and what changes once the AI has a documented framework instead of a generic one.