TL;DR
Capture 2 to 5 raw bullets per day (90 seconds), then on Friday paste them into Claude or ChatGPT with one fixed prompt. You get a structured report in under 5 minutes: 3 wins, 2 blockers, 1 ask, 3 next-week priorities, under 300 words, plus one human-take sentence your manager actually reads. The AI does the bucketing and word-count discipline; you do the judgment. Below: the exact prompts, which model to use, and the fact-check that keeps AI from inventing wins you never shipped.
The Friday 4:30pm problem
It is 4:30pm on Friday and you cannot remember what you did on Monday. So you spend 45 minutes reconstructing the week from Slack, Linear, and your inbox, and the report still reads as a generic “kept making progress” paragraph. The reconstruction is the expensive part, and it is also the inaccurate part: rebuilding a week from memory on Friday misses 30 to 50 percent of the real signal because the small wins and the mid-week blocker you already routed around have evaporated.
This workflow inverts the problem. You capture raw notes daily, when the detail is still fresh, then let AI do the summarizing on Friday. The capture costs 90 seconds a day. The Friday step costs under 5 minutes. The output is a 3-2-1-3 report: 3 wins, 2 blockers, 1 ask, 3 next-week priorities, under 300 words, with one sentence of your own judgment that no model can write for you.
Who this is for
ICs and managers who file weekly reports, whether to a manager, a skip-level, or a recurring 1:1 doc. It is most useful for engineers, designers, PMs, and analysts who do a lot of small things in a week and find the granularity hard to summarize. Managers who file both an upward report and a downward report on the same week benefit twice: one daily log feeds both.
Reach for it every Friday or at end-of-shift before a manager check-in, and the night before a 1:1 if your team uses an async pre-read. The daily-log discipline also scales straight up to project-level reports that roll across multiple weekly cycles. Skip it the first week of a new job; you do not have enough signal to summarize yet.
Before you start
- Pick one daily-capture surface and stick to it: a single doc, a Notion page, a Linear “personal” project, or a plain text file. Switching surfaces breaks the habit.
- Decide your structure once. 3-2-1-3 (3 wins, 2 blockers, 1 ask, 3 next-week priorities) is a defensible default; adjust to match your team’s template.
- Set a 90-second daily reminder, around 4:50pm or at the end of your last meeting. Keep capture cheap: the bar is one bullet per item.
- Know your manager’s read pattern. Some skim the TL;DR; some read the asks section first. Order your report around what they actually read.
- Keep last week’s report open. Week-to-week continuity is part of the signal.
Which AI tool to use
Any current general-purpose model handles this task well, but they differ at the edges that matter for a weekly report: following the word limit, not inventing causal links, and producing prose you barely have to edit. The picks below are current as of June 2026.
| Tool | Best model (June 2026) | Why for weekly reports | Cost |
|---|---|---|---|
| Claude | Sonnet 4.6 (or Opus 4.7) | Tightest at honoring constraints (word count, “no buzzwords”, 3-2-1-3); least likely to slip a section. Strong for concise professional prose. | Free tier (Sonnet 4.6, limited) or Pro $20/mo |
| ChatGPT | GPT-5.5 | Fastest and most flexible across formats; good when you also want a Slack version, an email version, and a slide bullet from one input. | Free tier (GPT-5.5, tight limits) or Plus $20/mo |
| Gemini | Gemini 3.1 Pro | Best if your source-of-truth lives in Gmail and Google Docs; Gemini in Workspace can summarize threads directly. | Google AI Pro $19.99/mo (formerly Gemini Advanced) |
A practical rule: Claude tends to nail every constraint in the prompt, while ChatGPT is faster and more flexible if you want to fan one input out into several formats. For a single tight report, Claude’s Sonnet 4.6 is the safe default; for a one-input-many-outputs Friday, GPT-5.5 wins on speed. Both free tiers are enough to run this workflow weekly. See our ChatGPT vs Claude vs Gemini comparison if you have not picked a daily driver yet, and weekly planning prompts for the prompts that feed Monday’s plan.
Step by step
- Capture daily. Throughout the week, record wins, blockers, discoveries, and questions in a single raw doc. Plain bullets, no formatting, 2 to 5 bullets per day. Example:
shipped invoice export PR, took 2 days vs 1 estimate, postgres COPY slower than expected. - Paste and prompt on Friday. Drop the raw doc into your model of choice with a fixed prompt:
Convert this week's notes into a weekly report.
Format: 3 wins, 2 blockers, 1 ask, 3 next-week priorities.
Under 300 words total. Plain language, no buzzwords; do not use "leverage" or "synergy".
Do not invent causal links between items unless my notes state them.
Notes:
[paste raw doc]- Fact-check the output. Models sometimes invent connections that are not in the source (“this win directly enabled X” when X is unrelated). Verify every claim traces back to a specific bullet. This is the single step you cannot skip.
- Add your take. Edit for your voice and add the one thing AI cannot generate: one sentence with your opinion, your judgment call, or the thing you would say in person. This is where the human signal lives.
- Submit and save the template. Save the prompt and the input doc as next week’s starting point. Optionally archive a copy in a
weekly-reports/folder. - Optional pre-1:1 read. The night before your 1:1, drop the same notes in with a different prompt: ask for “the 3 questions my manager will probably ask me about this week.”
Make it reusable: Projects and custom instructions
Re-pasting the prompt every Friday is fine, but you can make the workflow one click instead of one paste.
- Claude Projects or ChatGPT Projects both let you set project-level instructions once. Create a project called “Weekly report,” paste the 3-2-1-3 prompt and your tone rules into the project instructions, and every chat inside it inherits them. On Friday you only paste the week’s notes. ChatGPT Projects keep their own Sources tab so you can attach a running notes file; Claude Projects do the same with an uploaded knowledge file.
- Linear’s AI summaries are worth knowing about if your team lives in Linear. Linear can distill project and initiative updates into a daily or weekly digest in your inbox, and (as of the May 2026 update) auto-create a Slack channel per project where updates post automatically. That digest is a strong source feed for your capture doc, though it summarizes project activity, not your personal cross-project week, so you still bucket and add the human take.
- Gemini in Gmail and Workspace can summarize email threads and project docs directly, which helps if half your week’s signal is buried in long email chains.
The point is the same in every tool: store the prompt and tone rules once, paste only the changing notes each week.
First-run exercise
Pick the week that just ended. Reconstruct a daily-log doc from Slack, Linear, and your inbox the manual way, this one time. Run the prompt and compare the AI output to what you would have written yourself. Most people find AI is better at structure and worse at voice. The takeaway: use AI for the bucketing and the word-count discipline, and spend your saved 2 minutes on the human-take sentence. Start the daily-capture habit Monday so next Friday is the real test.
Quality check
- Every claim in the report traces back to a specific bullet in the raw doc. AI invents connections; you verify.
- Word count is under target (300 words is a reasonable ceiling). If AI produces 500, ask for a re-trim instead of editing by hand.
- The ask is named, specific, and time-bound. “I need budget approval by Wednesday from finance” yes; “more support would be nice” no.
- The human-take sentence is yours, not AI’s. Read it aloud; if it sounds like a press release, rewrite it.
- Next-week priorities are 3 items, not 10. A list of 10 priorities has 0 priorities.
- Blockers actually block progress; “could go faster with help” is not a blocker.
Common mistakes
- Trying to remember the week instead of capturing daily. Friday-only reconstruction misses 30 to 50 percent of the real signal and feels like a chore.
- Submitting the AI version verbatim. Managers read a lot of weekly reports; they can tell within 2 sentences when it is AI-only.
- No human-take sentence. Without it, the report is bureaucratic noise; with it, it becomes a 1:1 conversation starter.
- Letting next-week priorities balloon to 6 or 8. Pick the 3 that matter.
- Listing every micro-task as a win. Three substantive wins beat ten trivial ones.
- Skipping a week because it was “slow.” Send the report anyway; cadence is the trust signal.
FAQ
- Which AI is best for weekly reports in 2026?: For a single tight report, Claude (Sonnet 4.6, or Opus 4.7 on the $20/mo Pro plan) honors constraints like the 300-word cap and the 3-2-1-3 structure most reliably. If you want to fan one input into a Slack post, an email, and a slide bullet, GPT-5.5 in ChatGPT is faster and more flexible. Both free tiers are enough to run this weekly.
- Can I do this on a free plan?: Yes. ChatGPT Free (GPT-5.5) and Claude Free (Sonnet 4.6) both handle a once-a-week 300-word report inside their limits. You only need a paid plan if you also run AI all day for other work.
- My manager wants a different format.: Adjust the prompt’s section count and order; keep the daily-capture format the same. Your input pipeline is portable.
- What if I work on too many small things to bullet?: Group by project. Each project gets one bullet per day with the day’s most important update. Three projects equals three bullets per day.
- What about confidential information?: Keep customer names and revenue numbers out of the prompt. Use role tags (“EU enterprise customer”) and round figures (“around $50K ARR”). If your employer requires it, use the enterprise/Team tier of your tool, which does not train on your inputs.
- Can AI write the human-take sentence?: It can draft one, but the value disappears. That sentence is the part of the report that proves you actually thought about the week.