Use ChatGPT Work to Turn a Messy Monthly Report Request Into a Reviewable Brief

Evidence label: documentation-based guide with an illustrative example. The sample inputs and output below are fictional. No ChatGPT Work execution is claimed.

If a monthly report starts as a vague message like “make this presentable,” the first output is often polished but hard to check. ChatGPT Work is documented as an experience for longer, multi-step work and finished deliverables. That makes it a reasonable surface for a structured report brief, provided the source files and review rules are explicit.

Prerequisites

You need eligible access to ChatGPT Work on the web or mobile, or in the desktop app where your plan allows it. The current OpenAI help page describes Work as an agent for research, analysis and finished documents, spreadsheets, presentations or reports. Feature availability depends on the plan and workspace.

Use fictional or approved internal files only. Do not upload confidential data until your organisation has approved the tool and its data handling.

Fictional input

You are preparing a customer-support operations brief for September:

• tickets.csv: 240 tickets with ticket_id, opened_date, team, priority, first_response_hours, resolution_hours, status
• targets.txt: first response target = 8 hours; resolution target = 48 hours
• notes.txt: “Weekend coverage changed on 14 September. Treat tickets without a closed date as open, not resolved.”

The desired artifact is a two-page brief with an exception table, not a decision to change staffing.

The workflow

1. Start a Work task with a bounded deliverable. Attach the three fictional files or your approved equivalents. Say that the report must preserve the original values and that calculations must be shown.

2. Paste this brief:

Prepare a monthly support-operations brief from the attached files.

Deliverables:
1. A five-sentence executive summary.
2. A table of tickets that miss either target, with ticket_id, team, metric, actual value, target, and source file.
3. Counts by team for total, open, resolved, first-response misses, and resolution misses.
4. A “missing or ambiguous data” section.

Rules:
- Use only the attached files. Do not invent values.
- Treat tickets without a closed date as open, as stated in notes.txt.
- Show the formula or filtering rule behind every count.
- Separate observed facts, calculations, and suggested questions.
- Do not recommend staffing changes or send anything.
- Before finalising, list five checks a human should perform.

3. Review the draft against the files. Check one total manually, inspect a few exception rows, and confirm that open tickets were not silently counted as resolved. If the brief invents a number or hides an assumption, ask for a correction rather than accepting a smoother paragraph.

Acceptance check

The brief passes when every reported number can be traced to one of the three inputs, the exception table identifies its metric and target, open tickets remain open, and unresolved questions are visible. “Looks professional” is not an acceptance check.

Common failure fixes

• A plausible total has no method: ask for the exact filter and calculation, then recompute a sample outside the tool.
• The report fills missing dates: restate “missing means unknown” and require an ambiguity list.
• The output becomes a slide deck before review: request the plain table and formulas first.

TRY / SKIP

TRY this for a bounded report package where a person can check the source rows. SKIP it for regulated conclusions, personnel decisions, or any workflow where the files cannot be reviewed.

For software-development tasks, OpenAI documents Codex as the separate coding experience; Work is the better fit only when the deliverable is an office report rather than code. Read the current ChatGPT Work and Codex documentation.