To use ChatGPT for a weekly project status report, export your task list with task IDs, owners, deadlines, statuses and blockers, then ask for a draft using explicit red, amber and green rules and evidence references for risks and decisions. Check counts, dates and every material claim against the tracker before approving the report or changing any task.
The Complete Excel Course with AI Integration
From formulas and pivot tables to Power Query, DAX, VBA and AI with Copilot, ChatGPT and Claude — hands-on from the first lesson.
By Sagnik Bhattacharya and Virginia Thorn, Primetime Emmy Award winner
Buy nowA task list is a snapshot, not a forecast. Counting finished tasks cannot tell you whether the remaining work fits before the next milestone. Supply the dependencies and the agreed delivery assumptions, and make ChatGPT label missing information instead of turning an incomplete tracker into an unjustified green.
Why freeze a reporting snapshot before collecting updates?
Freeze the reporting snapshot at a cut-off, such as Friday at noon, so the report describes one moment instead of mixing versions. Ask owners to update the tracker before it. Export after the cut-off, note the export time in the file name, and keep that file with the draft. Anything you hear after the cut-off goes in as later evidence, labelled as such.
The worked example is an HR consultancy rolling out a new client onboarding process. The report is written on the Friday of week one, and the next milestone is a pilot in week two. Week labels are used here to keep the example general; in your own file, use real dates and a working calendar so overdue checks can be repeated by someone else.
Export these fields for every task: stable ID, task name, owner role, baseline due date, current due date, status, last-updated date, blocker, dependency and an evidence note. Keeping the baseline separate from the current date matters most. If someone pushes a missed deadline back before the export, a report built only on current dates hides the slip management needs to see.
ChatGPT accepts CSV and XLSX uploads on Free and paid plans, and OpenAI's file upload help lists counting spreadsheet rows with a particular attribute among the things it can do with them. On Free you get three uploads a day, which covers this week's export plus last week's. Upload a cleaned export rather than connecting the live tracker, and remove client names and personal details the report does not need.
If the tracking process itself is the problem, with no owners, no due dates and nobody updating it, the wider project delivery tutorial covers setting that up. A well-written status paragraph cannot repair a tracker that nobody has touched since the last report.
What do red, amber and green mean in a ChatGPT status report?
Red, amber and green (RAG) in a status report should follow project-specific rules written before ChatGPT sees the data, or the model picks colours from the tone of the notes. A workable rule: green when the next milestone has a credible current plan and no unresolved material threat; amber when a named, unresolved issue threatens that plan; red when the evidence shows the approved plan cannot meet the milestone.
Add a fourth state, Needs review, for when evidence is missing or stale. This is a reporting convention you choose, not an industry rule, but it beats a hopeful green when key owners have not updated their work. The project owner approves the final colour after checking the evidence; ChatGPT only proposes it.
Use these project rules:
Green: credible current plan meets the next milestone; no unresolved material threat.
Amber: a named unresolved issue threatens that plan; owner action needed.
Red: evidence shows the approved plan cannot meet the milestone.
Needs review: evidence missing, or not updated in the last 5 working days.
Propose a status with task IDs and reasoning. Do not invent forecasts or change dates.
That freshness rule catches a common trap. A task marked On track whose last update was nine working days ago is not evidence of anything. Ask its owner whether the forecast still holds, and until they answer, the report says the evidence is missing. A stale green is a data problem with delivery consequences.
Red needs evidence too, and it looks different. Picture a mortgage broker moving office, with the move tracked in a spreadsheet. The new internet line must be live before staff move on the 14th, but the supplier's written confirmation gives the 21st as the earliest installation date. Under the rule above that is red, because the approved plan cannot meet the milestone, and the report cites the supplier's confirmation as its evidence. ChatGPT's job is to notice that the evidence contradicts the plan; the owner decides whether the move date or the connection plan changes.
Status and importance are different things. A finished minor task is done but says little about the launch, while an open dependency can need attention before its own due date arrives. Each paragraph of the report should name the next milestone, the task that controls it and the evidence for the proposed colour.
Read eight tasks without mistaking activity for delivery
The consultancy's export at the Friday cut-off has eight tasks: three done, three in progress, one blocked by missing sample data and one blocked by a missing reviewer. The owner wants to know whether the week-two pilot is still credible and what decisions are needed. All tasks and dates are invented for the example.
| ID | Task | Owner role | Baseline due | Status | Last updated | Evidence or blocker |
|---|---|---|---|---|---|---|
| A1 | Agree required onboarding fields | Client lead | Wk1 Mon | Done | Wk1 Mon | Requirements signed off |
| A2 | Map current records | Data officer | Wk1 Tue | Done | Wk1 Tue | Mapping reviewed |
| A3 | Approve welcome wording | Client lead | Wk1 Wed | Done | Wk1 Wed | Final wording filed |
| A4 | Configure pilot workflow | Systems admin | Wk2 Mon | In progress | Wk1 Thu | Owner forecasts Wk2 Mon |
| A5 | Test sample records | Data officer | Wk1 Thu | Blocked | Wk1 Thu | Sample data not supplied |
| A6 | Review operator instructions | Unassigned | Wk1 Fri | Blocked | Wk1 Wed | No reviewer named |
| A7 | Prepare pilot acceptance checks | Client lead | Wk2 Tue | In progress | Wk1 Fri | Depends on A4 |
| A8 | Prepare handover record | Office manager | Wk2 Fri | In progress | Wk1 Fri | Draft exists; finishes after pilot |
At the Friday cut-off, A5 is a day overdue. A6 is due that same day, so whether it counts as overdue depends on your due-time rule; the honest wording is “blocked and due at cut-off”. That small distinction stops a model inflating lateness by treating every due date as a midnight deadline that has already passed.
Three of eight tasks are done, 37.5% by count, and two of eight are blocked, 25%. Those are useful inventory figures, but the project is not 37.5% complete: the tasks differ in size, and A5 controls the pilot. Any statement about delivery needs the dependencies and the remaining effort.
Here the plan needs sample data early in week two and a named reviewer. Both are unresolved and both threaten the pilot, so amber is the proposed status for the owner to confirm. Do not invent a new pilot date. Ask the project owner whether testing and review still fit once the data owner commits to a delivery time.
If your task list lives in Excel, try The Complete Excel Course with AI Integration on Coding Liquids for the spreadsheet side.
Why ask ChatGPT for an exception report before the narrative?
Ask ChatGPT for a structured exception report before any narrative: counts by status, duplicated IDs, missing owners, overdue tasks, stale updates and dependencies with no evidence. Overdue tasks are judged under your due-time rule. Review those exceptions before asking for any management paragraph, because fluent prose smooths over source problems that a list exposes.
Inspect this task snapshot before drafting anything.
Use the cut-off (Friday 12 noon), the due-time rule (due at 5 pm on the
due date) and the RAG rules above. Return: counts by status; IDs with
missing owners, stale updates or blockers; overdue IDs with baseline
dates; dependencies relevant to the next milestone. Cite task IDs for
every finding. Ask about anything unclear. Do not edit the tracker.
A good inspection, as an illustration, reads: “8 tasks: 3 done (A1–A3), 3 in progress (A4, A7, A8), 2 blocked (A5, A6). Overdue: A5 (baseline Wk1 Thu). Due at cut-off and blocked: A6. Missing owner: A6. Milestone dependencies: A4, A5, A6, A7. Question: when will sample data for A5 arrive?” An output that says “five tasks remain, all on schedule” contradicts the blocked records and goes back for correction.
Duplicate IDs are worth testing on purpose. A joined export can list A5 twice because it carries two tags, giving nine rows for eight tasks. Do not delete one row blindly if the comments differ: merge the evidence under the stable ID and recount unique tasks before any percentage reaches the report.
Task hierarchies cause the same trap. A parent task called Launch with six subtasks becomes seven “equal” items if counted naively, which flatters progress. Decide which level counts, keep the parent as context, and make the model state its counting rule whenever parents and subtasks are mixed.
What is the difference between an issue, a risk and a decision?
An issue is a problem that already exists, a risk is a possible consequence, and a decision is a choice that needs authority. In the example, the issue is that the sample data is missing. The risk is that late testing may delay the pilot. The decision is to name a data owner and agree a delivery time, or approve a revised pilot plan if the current sequence cannot fit. Keep the three linked, but do not merge them into “the project faces some challenges”.
For A5, the issue entry reads: “Sample data not supplied; testing blocked; source A5; owner awaiting confirmation.” The risk entry: “Pilot may move if testing cannot start by Wk2 Mon.” The decision request asks the sponsor to confirm an accountable data owner and whether the current pilot date stays the target once the updated plan is reviewed.
For A6, the action is to assign a reviewer, not to ask ChatGPT to review the operator instructions itself. The report can say a missing reviewer threatens readiness and name who must resolve it; it must not quietly substitute an AI review for a sign-off the process requires.
A useful decision record holds the choice, the decision maker, the needed-by time, the evidence, the consequence of no decision and the current state. “Management to review” holds none of those. “Sponsor: name the operator-instruction reviewer by Monday 10 am; otherwise confirm how pilot readiness will be judged” holds all of them, and the sponsor still chooses the person.
Do not let the model invent probabilities or money values from task notes. If an owner supplies an estimate, show its basis; otherwise describe the risk with a trigger and an owner. That is more useful than a made-up 72% chance of delay.
Draft the owner report with task references intact
Draft a weekly owner report from the approved findings only.
Sections: proposed RAG with evidence; completed this week; next milestone;
issues and risks; decisions needed; next-week commitments with owner roles.
Keep task IDs. Separate confirmed dates, forecasts and missing commitments.
Never invent an owner, promise completion or change a baseline date.
No percentage-complete claims. Under 200 words.
The consultancy's finished report, as approved by the project owner, reads:
Onboarding pilot: weekly status, Wk1 Fri (snapshot 12 noon). Proposed status: AMBER. The pilot remains the target, but two blockers are unresolved. Done this week: required fields, record mapping and welcome wording (A1–A3). Issues: sample-data testing is blocked and one day overdue (A5); operator-instruction review is blocked with no reviewer and due today (A6). Risk: if testing cannot start by Wk2 Mon, the pilot date may move. Decisions needed: sponsor to name a data owner with a delivery time, and a reviewer for A6, by Mon 10 am. Next week: systems admin forecasts workflow configuration done Wk2 Mon (A4); acceptance checks follow (A7). Commitment still required: sample-data delivery time.
Notice what it does not say. “Systems admin forecasts A4 done Wk2 Mon” is supported by the record; “the data officer will deliver on Monday” would not be, because nobody has accepted that commitment yet, so the report says “commitment still required” instead. The report should reveal ownership gaps, not hide them behind confident future tenses.
When the audience is a client rather than the owner, the client status report template gives a client-facing layout. Decide what the client may see before rewording internal notes. A supplier dependency might be commercially sensitive for a B2B software reseller, for example: the client version can describe the dependency and its effect at an agreed level, but the status itself must be the same in both versions. A softer tone should never turn a blocker into “on track”.
How do you report what changed since last week?
To report what changed, upload last week's approved snapshot alongside this week's and ask ChatGPT to compare stable task IDs: newly done, reopened, newly blocked, removed and added. One export alone cannot show change. If last week's file is missing, say so and report current state only.
Take a project that had three tasks done last week. Two more finish, but one earlier task reopens after failing acceptance, so four are done now. “Two completed this week” is true but incomplete: the movement is three plus two minus one, equals four. Report the reopened task and its effect so the headline does not overstate progress.
Added scope changes the denominator. If a project grows from ten tasks to twelve after two approved additions, a lower completed share may reflect scope growth rather than slower delivery. Show the additions and their approval separately, and never edit the old snapshot to make the weeks look comparable, because that erases the evidence of the scope change.
When forecasts move, keep baseline and revised dates side by side with the reason and whether the change was approved. Otherwise a task can look on time simply because its due date was pushed. If your tracker lacks consistent fields for this, the spreadsheet project tracker tutorial sets them up.
Why is count-based completion misleading in a status report?
Count-based completion can badly misstate effort, so state whether a percentage means tasks done or estimated effort. In a separate example, eight one-hour tasks are done and two ten-hour tasks remain. By count that is 80% done; by estimated effort it is eight of twenty-eight hours, about 28.6%. Neither figure alone shows whether the milestone is safe, so keep the dependency discussion beside it.
A zero-task export needs a different response. It may come from a filter, the wrong project or a genuinely empty plan. Never report 100% complete because nothing is open. Ask the owner to confirm the export scope, and put the project identifier and filters in the control note.
Check any claim about the critical sequence with the delivery owner. If testing needs three working days and pilot observation two more, and five working days remain, there is no slack under those assumptions; a further one-day delay means a changed plan or date. ChatGPT can do that arithmetic, but it cannot decide which scope or acceptance step to drop.
Finally, read every “will”, “completed” and “approved” in the draft against its evidence. “Draft ready” in a task note is not approved instructions, and a dependency marked requested is not received data. Replace unsupported certainty with the exact state and the action required; this catches errors that correct counts never will.
What do you do when evidence changes after the report went out?
If a material update arrives after the cut-off, decide whether it needs an immediate correction or belongs in next week's report. Suppose the data owner shows that sample records were delivered before the cut-off but the tracker was never updated. Check the delivery evidence, correct A5, and ask the project owner whether amber still stands.
Keep the original snapshot and label the corrected report as a revision that says what changed and why, rather than quietly replacing a file the sponsor has read. If the records arrived after the cut-off, the original report was accurate for its stated time: record the later event without rewriting history.
How do you approve the report and write decisions back?
Approve the weekly report only after checking it against the tracker, then record agreed owners, dates and decisions in the tracker itself. Once the tracker fields are consistent, the routine fits in about thirty-five minutes a week: ten for updates and the export check, ten for inspection and drafting, ten for owner verification and five for circulating and recording decisions. That is a suggested budget for a single small project; measure your own before relying on it.
Before circulating, verify against the tracker: project, snapshot time, unique task count, completed IDs, overdue rule, baseline dates, milestone evidence and decision owners. Keep the export, the approved report and the reviewer's note together. A provisional report says it is provisional.
After the meeting, enter agreed owners, dates and decisions in the tracker through your normal process. A sentence in a report does not update the source of record. Then next week's export starts from the corrected tracker.
When the manual routine has run cleanly for a few weeks, automation becomes reasonable. OpenAI's ChatGPT Work page gives a weekly brief with priorities, risks, decisions and follow-ups, and a project tracker built from a planning document with a drafted status update, as example tasks, and describes team tasks that delegate recurring work such as weekly reports. Free and Go users get limited Work in the desktop app, Plus and above also get it on the web and mobile, and the pricing page shows scheduled tasks on Go, Plus and Pro but not on Free. Keep the owner's approval step whichever way the draft is produced: the report earns trust when every important sentence traces back to current task evidence.
If you want help building a reporting routine around your own tracker, I offer one-to-one AI consultations for small businesses.
Further reads
- Asana AI Status Updates: Draft a Weekly Client Report From Tasks — Draft status updates inside Asana instead of exporting.
- AI Tools for Operations Management: Weekly Exceptions Report From Exports — The same exceptions-first approach for an operations export.
Browse all AI-for-business tutorials
Sources: OpenAI help articles on uploading files and on data analysis with ChatGPT; ChatGPT Work page; ChatGPT pricing page. Checked 18 November 2026.
