Running a Weekly Review with AI

TL;DR: Running a weekly review with AI means feeding the model your own week, tasks done, numbers tracked, what stalled, not asking it to read your calendar on its own. Paste in the raw material, run one of the two templates below, and save the version that fits your role once instead of rewriting it from a blank page every Friday.

What Does Running a Weekly Review with AI Actually Involve?

Not what the phrase suggests. "AI weekly review" sounds like something autonomous: connect an assistant to your week and it hands back an assessment. That is not what is happening, and treating it that way is how you end up with a summary that sounds authoritative and is quietly wrong.

What is actually happening is closer to dictation with a very fast, very literal assistant. You gather what you did, what you tracked, and what stalled, in whatever rough shape that takes: a task manager export, a running notes file, a scroll through Slack, or five minutes of typing from memory. You paste that into a chat. The model's job is narrow but genuinely useful: find the pattern in a page of scattered notes faster than you can, ask the one question you should have asked yourself, and turn a loose braindump into something with a shape, three lines that matter, one line that does not, a next step that survives contact with Monday.

That is a real job and it is worth doing every week. It is just not the job of gathering the material in the first place, and the gap between those two jobs is where most people's first attempt at this goes wrong.

Picture the failure mode directly. Someone opens a chat and types "review my week," nothing else. The model has no notes, no numbers, no task list, and no honest way to say so without disappointing the person who asked. So it produces a reasonable-sounding generic reflection: some encouragement, a suggestion to prioritize better, a nudge toward work-life balance. It reads fine. It describes nobody's actual week, because nobody's actual week was in the request. The fix isn't a smarter model or a better prompt phrasing; it's giving the model something real to work from before you ask it to work at all.

What Do You Need to Gather Before You Open a Chat?

Four things, and you can gather all four in under five minutes because none of them need to be polished.

What you did. A done list, closed tickets, finished tasks, whatever your tool of choice already tracks. If nothing tracks it automatically, a rough braindump works: open a blank note and type every task you remember finishing, in whatever order they come to mind.

What you tracked. Two or three numbers you already look at, revenue, workouts logged, words written, tickets closed, outreach sent. Not a new dashboard you build for this exercise. If you have to calculate something you have never calculated before, skip it this week.

What stalled or got pushed. One or two lines, no explanation required yet. "Client proposal, still not sent" is enough. The model's job in a minute is to turn that into something actionable; your job right now is just to name it.

Nothing else. Do not add context about your industry, your role, or what a "good week" is supposed to look like. That is exactly the kind of thing the model will invent on its own if you leave it out, and it is exactly the kind of thing it should not be inventing.

What if you genuinely have nothing to report? Some weeks are slow, or sick, or eaten by something outside work entirely, and that is a legitimate answer to all three prompts above, not a reason to skip the exercise. "Nothing shipped this week, out sick Tuesday through Thursday" is a complete, honest entry. Padding it with invented accomplishments defeats the entire point of running a review in the first place: the value is in an accurate record, not in a record that always looks productive.

A Weekly Review Template You Can Paste Right Now

This one is built for a single person reviewing their own week, no audience beyond yourself.

You're helping me run my weekly review. Here's my raw material for the
week of [WEEK OF DATE]:

WHAT I DID:
[paste your task list, done column, or a rough braindump]

WHAT I TRACKED:
[paste 2-4 numbers you actually track: revenue, workouts, words written,
tickets closed, whatever you already look at]

WHAT STALLED OR GOT PUSHED:
[one or two lines, no need to explain why yet]

Using only what I gave you above, and nothing you're inferring about my
calendar, industry, or role:

1. Name the one thing that actually moved the needle this week, and say
   why you picked it over the others.
2. Name one thing I said I'd do and didn't. Don't soften it.
3. Turn "what stalled" into a single sentence I can act on Monday morning.
4. If anything above is ambiguous or missing, ask me one question about
   it instead of guessing.

Keep the whole reply under 200 words.

The last line matters more than it looks. Without a length cap, a model asked to review your week will generate exactly as much text as the request seems to deserve, four paragraphs of even-handed observation whether your week produced one real insight or none. Capping the reply forces it to rank what it found instead of listing all of it.

A filled-in version, for a hypothetical week, looks something like this once the brackets are replaced with real material:

WHAT I DID:
Shipped the invoice-reminder feature, onboarded two new clients, wrote
the March newsletter draft, ran four sales calls.

WHAT I TRACKED:
$3,400 in new revenue, 4 sales calls booked, 1 blog post published.

WHAT STALLED OR GOT PUSHED:
Client proposal for the Chen account, still not sent. Third week in a
row it's been pushed.

A model working only from that has something to actually reason about: three weeks of the same stalled task is a pattern worth naming outright, not a vague nudge toward "better time management." That specificity, not a cleverer prompt, is what makes the output worth reading.

A Version for When Someone Else Reads the Output

The personal template above is for your own eyes. The moment someone else reads the result, a manager, a client, an investor, the job changes: brevity and honesty matter more, and every claim needs to trace back to something you actually wrote down.

Turn my week into a short update for [WHO READS THIS: my manager / my
team / an investor].

MY RAW NOTES FOR THE WEEK OF [WEEK OF DATE]:
[paste your notes, ticket list, or standup recap]

Write three short sections:
- Shipped: what's actually done, not "in progress"
- Blocked: what's stuck and what you need to unstick it
- Next: the 2-3 things you're committing to next week, not a wishlist

Rules:
- Every line has to trace back to something in my notes above. If you
  can't find it in my notes, don't include it.
- No adjective that isn't backed by a number or a name. Write "closed 14
  tickets," not "had a productive week."
- Flag anything in my notes you don't understand well enough to
  summarize confidently, instead of smoothing over it.

If you already send a recurring client-facing status update, Free Client Update Email Generator (Weekly Status) has a template built specifically for that shape. If the audience is closer to a manager expecting release-note structure, Stakeholder Updates & Release Notes on Repeat covers that cadence directly.

What Cadence Actually Works?

There is no single right day, but there is a wrong pattern: reviewing sporadically, whenever you remember to, then wondering why nothing about the practice ever compounds. Pick one slot, put it on your actual calendar, and treat a skipped week the same way you would treat a skipped rent payment, something to notice, not something to let become the default.

None of these is objectively correct. What matters is that the slot survives contact with a bad week, since a bad week is exactly when the review is most useful and most tempting to skip. Whichever slot you pick, treat the first two or three weeks as calibration, not commitment: you're finding out how much raw material you actually have to paste in on a normal week, and adjusting the template's expectations to match, rather than assuming the first draft of the prompt got it right.

What if you skip a week anyway? Don't try to reconstruct it from memory two weeks later; memory is exactly what this exercise is supposed to stop relying on. Note in the next review that last week was skipped, and move on. A gap in a weekly log is honest information. A guessed-at retroactive entry is worse than nothing, because it looks like data and isn't.

What Can't a Weekly Review Prompt Do for You?

Worth being direct about the ceiling here, since it's easy to oversell a template like this once it starts working. It cannot tell you whether you're on track for a goal it was never told about; if you want that judgment, the goal has to be in the prompt every time, not assumed. It cannot notice a number you never wrote down, so a week where the real story was something you didn't track, hours spent in unproductive meetings, say, will come back looking fine, because nothing about it was ever visible to the model. And it holds no memory of last week's review unless you paste last week's output back in yourself; each run starts from zero, which is exactly why the template asks for what stalled every time instead of assuming continuity.

None of that makes the exercise less useful. It makes it a tool for organizing what you already noticed, not a substitute for noticing in the first place.

How Do You Stop the Model From Just Being Nice About Your Week?

A model asked to review a thin, vague input does not respond with "there isn't enough here." It responds with four confident-sounding observations anyway, some of them quietly invented to fill the gap you left. That is not a flaw specific to any one tool; it is what happens whenever you ask for a confident answer and only provide a partial question. Why Does ChatGPT Make Things Up? covers the mechanism in more depth than fits here.

Three things fix most of it in practice. First, the instruction already in both templates above: work only from what I gave you, and ask a question instead of guessing. Second, read the first draft against your own memory of the week before you trust any specific claim in it, the same way you would read a first draft of anything else you didn't write yourself. Third, if a sentence in the output describes something you don't remember happening, it probably didn't; go back to your notes and check before you repeat it to someone else.

A concrete version of the failure: paste in three vague lines, "did some marketing stuff, talked to a few clients, worked on the website," and ask for insight, and you'll get back something specific-sounding anyway, a claim about which channel performed best, a number that was never in your notes. Nothing in the input supported that specificity; the model produced it because a confident, specific-sounding answer is what the request seemed to call for. The templates above head this off by giving it real numbers to reason from instead of vague categories to elaborate on.

How Do You Turn a One-Off Review Into a Prompt You Reuse Every Week?

This is the part that makes a weekly review the clearest possible case for a saved prompt template, more than almost anything else you'd write. The structure never changes: what you did, what you tracked, what stalled, three or four fixed instructions underneath. Only the details inside the brackets change, and they change every single week without fail.

The practical version: write the prompt once, run it for two or three weeks, and notice which parts of your first draft you keep editing by hand every time. Fold those edits into the template itself. What is left after that, the boilerplate outside the brackets, is the version worth keeping permanently rather than rewriting from a blank page each Friday.

A concrete example of that folding-in: if you notice yourself adding "and don't use the word 'productive,' just tell me what happened" to the end of the prompt three weeks running, that's not a one-time correction anymore, it's a rule that belongs in the template itself, permanently, so you stop typing it fresh every Friday. That is the entire mechanism by which a rough first draft becomes a template worth keeping: not a rewrite, just noticing which edits repeat and moving them from the bottom of the chat into the prompt's fixed text.

Where you keep it matters less than doing it at all, but it's worth being precise about what "free" actually includes here. Prompt Architects' free plan gives you five architected prompt generations a day, and that's real and documented, but it does not include Refine mode, Shorten mode, or a saved prompt library; those start on the Pro plan, which caps your library at 50 saved prompts, with Advanced and Team removing the cap entirely (verified on the pricing page, September 2026). A weekly review template is exactly the kind of single, high-value entry worth one of those slots. For the broader question of what belongs in a library once you're deciding among more than one candidate, How to Build a Personal AI Prompt Library goes further than a single template.

If you'd rather build this from an existing habit than start from the templates above, Journaling and Reflection Prompts with AI covers the more personal, less output-driven version of the same underlying idea.

None of this requires a special tool to start. A plain chat window and a template pasted from a notes file gets you every benefit described on this page; a saved library just removes the one remaining friction point, opening a notes app and finding the right file, once you've already decided the template is worth keeping past a single Friday.