Skip to main content

Automate a Weekly Report with AI

info

This content was developed with AI assistance and is regularly reviewed for accuracy.

Snapshot: October 2026. Model names, prices, and product features change monthly; What's New in AI is the maintained chronology.

This playbook takes a report you write every week and turns it into one the assistant drafts on a schedule, from sources it reads itself. You will design the report around what your reader needs, connect the documents and inboxes it draws on, prove the brief works by hand, and then schedule it. The last step is the one that keeps automation honest: you review every run before anything reaches your reader.

Last tested with Claude scheduled tasks (Pro plan), Claude Code Routines (research preview), and ChatGPT scheduled tasks (Plus plan) in October 2026.

What you'll accomplish​

  • Design a report your reader can absorb in two minutes
  • Connect the sources the report is built from
  • Write and test the report brief by running it manually
  • Schedule it as a recurring task that drafts but does not send
  • Review each run and improve the report from feedback

Prerequisites​

  • A report you produce on a regular schedule, or one you need to start
  • A paid plan with scheduled tasks: ChatGPT Plus and above, or Claude Pro and above. On a free plan, do Steps 1-3 and run the brief by hand each week.
  • Connectors for where your inputs live: Google Drive, Gmail, Google Calendar, or a team chat app where your plan offers one
  • A running notes document you add to during the week
  • 60 minutes to set up, then about 10 minutes a week

The playbook​

Step 1: Design the report for its reader​

Goal: A structure your reader values, not a log of what you did.

Action: Start from the reader's decisions. What do they use this report for? What makes them stop reading? Ask the assistant to propose a structure, then cut it down:

I send a weekly status report to my director, who reads 10+ of these a week on her phone. She uses them to spot projects at risk and decide where to step in. My current report is long paragraphs she skims. Propose a structure she can absorb in under two minutes, with the most decision-relevant information first. Explain what you cut and why.

A strong default: a one-line status (on track, at risk, blocked), what needs her help, what shipped, and next week's top three.

Checkpoint: A four- or five-section template that leads with what the reader needs to act on.

Step 2: Connect the sources​

Goal: Let the assistant gather the inputs instead of you pasting them.

Action: List where the week's evidence lives and connect each source. In Claude, connectors are under Settings; ChatGPT's connected apps are on paid plans. Typical sources:

SourceWhat it contributes
Running notes doc in DriveYour own one-line entries, the most reliable input
Sent emailCommitments and updates you sent others
CalendarMeetings held, launches, deadlines
Team chatDecisions and blockers raised in channels

Keep the running notes doc even with every connector on. Two minutes at the end of each day ("Fixed login timeout in prod; client demo went well, asked about export") beats any amount of after-the-fact reconstruction, and it is the input you can trust most.

Connect only what the report needs. Every source you add is something a scheduled job can read without you watching.

Checkpoint: The assistant can answer "What's in my running notes doc for this week?" correctly.

Step 3: Write the brief and run it by hand twice​

Goal: Prove the brief produces a report you would send before you automate it.

Action: Combine the structure and sources into one brief. The old approach chained three prompts (organize, draft, review); current assistants do that work inside one well-specified request.

Objective: Draft my weekly status report for [director's name].

Context: [Project] status report. She reads it on her phone and acts on risks.

Resources: My Google Doc "[running notes]" for this week, my sent email from Monday to today, and my calendar for this week.

Constraints: Under 250 words. Plain language. Only report progress that appears in the sources. If a section has nothing, write "None this week."

Output: Status (on track / at risk / blocked) with one line why; Needs your help; Shipped; Next week's top three.

Verification: After the report, list each item with the note, email, or event it came from. Then list anything from my notes you left out and why.

Run it by hand this week and next. Check every item against its source. Adjust the brief where the draft was off; most fixes are a sharper constraint or a clearer source.

Checkpoint: Two consecutive manual runs produced reports you would have sent with only light edits.

Step 4: Schedule it to draft, not send​

Goal: A draft report ready every Friday without you starting it.

Action: Choose a scheduler that fits where you work:

OptionGood forNotes
ChatGPT scheduled tasksChatGPT Plus usersRebuilt June 2026; runs at most once per hour
Claude scheduled tasksClaude Pro usersRuns in the cloud, even with your laptop closed
Claude Code RoutinesPeople already using Claude CodeResearch preview on Pro and above; schedule, API, or GitHub triggers; uses your claude.ai connectors
Zapier, Make, n8nReports that need approval steps or an audit trailSee Workflows vs Agents

Paste your tested brief as the scheduled instruction and add a schedule and a delivery rule:

Every Friday at 2pm: [your tested brief]. Save the draft as a new Google Doc titled "Status report draft [date]" in my Reports folder. Do not email or share it with anyone.

That last line matters. A scheduled task acts without you present. Routines in particular include all your connected connectors by default, and Anthropic's documentation warns that the assistant can use every tool in an included connector, writes included, without asking. Remove any connector the report does not need.

Checkpoint: The task appears in your scheduled list with the right day, time, and a draft-only delivery rule.

Step 5: Review every run​

Goal: Catch errors before your reader does.

Action: When the draft lands, run a two-minute check: the status line matches reality, each "Shipped" item exists in the sources, and nothing in "Needs your help" is stale. Then send it yourself.

Do not mistake a completed run for a correct one. Anthropic's Routines documentation is blunt that a green status in the run list does not mean the task in your prompt succeeded. A run can finish cleanly and still miss a source that failed to load or a week with thin notes.

Keep a one-line log: date, what you fixed, why. Three weeks of the same fix means the brief needs changing.

Checkpoint: Three scheduled runs reviewed and sent, with a log of what you changed.

Step 6: Improve from feedback, then look ahead​

Goal: A report that gets better instead of staler.

Action: After a month, ask your reader what they use and what they skip, then hand the feedback to the assistant:

Here is my current report brief and three recent reports. My director says "Needs your help" is the most useful section, she never reads "Shipped", and she keeps asking about client feedback. Rewrite the brief to reflect this.

This report is the simplest kind of loop: an agent that wakes on a schedule, gathers inputs, and produces work for review. The next steps are loops that check their own output against a goal and keep going until it is met, with budgets and approvals to keep them safe. That is the subject of Loops and Harnesses.

Checkpoint: An updated brief that reflects real reader feedback.

Common pitfalls​

  • Automating a report that does not work by hand: Scheduling a bad brief produces a bad report every week. Prove it first.
  • Letting the task send the report: Draft to a document; you send. Keep the irreversible step yours.
  • Connecting everything: Each extra connector is more for an unattended job to read, and in Routines, to write.
  • Trusting the green checkmark: A finished run is not a correct report. Review every one.
  • Dropping the running notes: Connectors see email and calendars, not your judgment about what mattered.

Key takeaways​

  • Design for the reader's decisions, then build the brief around that structure
  • Connect only the sources the report needs, and keep a running notes doc
  • Run the brief by hand twice before you schedule it
  • Scheduled tasks should draft to a document, never send
  • Review every run; a completed run is not the same as a correct one

Next steps​

Continue to: Set Up Your Personal Agent to combine this report with a daily briefing and other recurring jobs.

Continue to: Loops and Harnesses for agents that check their own work and keep going until a goal is met.

Continue to: Workflows vs Agents: Which Automation Do You Need? if your report needs approvals or an audit trail.