Give ChatGPT the decision the report must support and the notes or records behind it. First separate facts, opinions, gaps, and proposed actions. Then draft a report that leads with the answer, explains the evidence, keeps limits visible, and labels recommendations as proposals. Check every important claim against its source before sharing it.
What you need before you start
Use this for a project update, event report, trial review, operations report, or short business recommendation. It is most useful when the source material exists but is scattered or hard to turn into a clear story.
Bring: Supply the audience, report purpose, covered period, required format, and labeled notes or records. Include figures with units and dates, source locations, approved decisions, and anything still unknown. Tell ChatGPT whether it may use outside sources.
Start with these steps
- State the question the reader needs to answer and the period the report covers.
- Label the supplied records, then ask for an evidence map separating supported facts, opinions, gaps, and proposals.
- Resolve conflicting figures and choose a short outline that serves the decision.
- Draft the report with a summary, findings, limits, recommendation, and next steps. Keep unapproved actions clearly proposed.
- Check calculations and claims against the records, then make a clean version for the reader.
A prompt you can adapt
Write a [length and type] report for [audience] answering [decision question], covering [period]. Use only the labeled records below. First identify facts, opinions, gaps, and conflicting figures. Then draft a report with a brief answer, evidence-backed findings, limitations, a clearly labeled recommendation, and proposed next steps. Keep sources beside key claims. Do not invent causes, costs, approvals, owners, or dates. Show the basis of percentages. If the evidence does not answer the question, say what is needed. RECORDS: [paste].
Replace every bracket with real information. Remove fields that do not affect your task.
What this looks like in practice
Our fictional training pilot has 24 participants from 40 invited staff and 18 feedback responses. Fourteen responses say it was useful. A good report says 14 of 18 respondents found it useful, not that 78 percent of all staff did. It also treats a fall in support requests as an observation whose cause is unproven.
From records to a decision
Counts, notes, dates, and costs
What the records actually show
What remains unknown
A proposed next step with a reason
Write for a decision, not for the appearance of a report
A report should help someone understand a situation or decide what to do. Before drafting, write the reader's question in one sentence: Should we repeat the training pilot? Why is the project late? Which option fits our needs? This question tells you which evidence belongs in the report.
Name the audience and the covered period. A team member needs enough detail to act. A manager may need the conclusion, costs, risks, and choice. A report about last month's work should not casually mix in this week's numbers without identifying the change in period.
Choose a format that fits. A one-page decision report may need a short summary, findings, limits, and recommendation. A longer project report may need a timeline and separate sections. Do not ask for a twenty-page document if the useful evidence fits on two pages. More headings do not create more insight.
Tell ChatGPT what a finished report must let the reader do. For example: The manager should be able to decide whether a second small pilot is worth planning, while seeing what the first pilot cannot prove. That instruction is more helpful than make it detailed and professional.
Prepare a small source pack
Gather the records before asking for a polished draft. They might include meeting notes, attendance, survey totals, invoices, logs, or previous reports. Give each one a short label and date. Labels such as A, Attendance and B, Feedback let you trace a claim without filling every sentence with long filenames.
Keep numbers with their meaning. Twenty-four participants is different from twenty-four registrations. Eighteen responses is different from eighteen positive responses. A cost of 180 might cover materials only, rather than the entire event. Include those distinctions in the pack.
Separate direct records from someone's interpretation. A support log can show a count. A team member may believe training reduced that count. Both can matter, but they are different kinds of evidence. Ask ChatGPT to keep the opinion attributed rather than presenting it as an established cause.
Remove personal details that the report does not need. An attendance total may be enough without employee names. Use approved records and keep the originals available for checking. The report should carry the relevant results, not expose every private detail from the source pack.
A worked example: should a training pilot be repeated?
The following fictional records describe a small training session. We want a short report for an operations manager deciding whether to plan another trial. The records contain useful evidence, a tempting causal claim, and several gaps. All dates, costs, and figures below are invented for this teaching example.
RECORD A, Attendance, 10 September:
40 staff were invited. 24 attended a 90-minute training session about handling routine support questions. Attendance was optional. No test was given before or after the session.
RECORD B, Feedback, collected by 12 September:
18 attendees completed a feedback form. 14 marked the session useful. 3 marked partly useful. 1 marked not useful. 9 respondents asked for more hands-on examples. The form did not ask respondents to demonstrate a skill.
RECORD C, Support log:
The team recorded 36 support requests in the seven days before the session and 30 in the seven days after it. The log does not show how many staff were working each day, changes in workload, or which staff submitted requests.
RECORD D, Costs:
Materials cost 180 example currency units. The room had no separate charge. Preparation took 6 staff hours. Staff time was not given a monetary value. No budget for another session has been approved.
RECORD E, Team note:
A team lead believes the training reduced support requests and suggests repeating it monthly. No decision, owner, or date has been agreed.Build the evidence map before the prose
The evidence map is a small bridge between raw notes and the report. It tells you what each record supports and what it does not support. This is where you catch denominator errors, missing costs, and conclusions that run ahead of the evidence.
In this example, attendance was 24 of 40 invited staff, or 60 percent. Feedback coverage was 18 of 24 attendees, or 75 percent. Positive feedback was 14 of 18 respondents, about 78 percent. Those are three different measures. Calling the last figure staff approval would make the evidence look broader than it is.
Support requests fell by 6, from 36 to 30. Dividing 6 by the original 36 gives about 16.7 percent. The arithmetic is straightforward, but it does not establish why the count changed. A week with less work could also have fewer requests. The records do not let us compare requests per staff member or per unit of work.
| Claim | Supported by | Limit |
|---|---|---|
| 24 of 40 invited staff attended | Record A | Optional attendance; not a random group |
| 14 of 18 respondents marked useful | Record B | Feedback is not a skill test |
| Requests fell from 36 to 30 | Record C | Workload and staffing were not tracked |
| Recorded materials cost was 180 units | Record D | Total cost excludes unvalued staff time |
| Monthly sessions are suggested | Record E | No approval, owner, or date exists |
A complete mini report built from those records
Here is an illustrative finished report. It answers the decision question early, keeps the evidence close to its source, and does not hide the gaps. The recommendation adds judgment, but it is plainly a proposal rather than a record of something already agreed.
TRAINING PILOT REPORT
Purpose: Decide whether to plan a second small trial after the 10 September session.
Summary
The first session received mostly positive feedback from respondents, but the records do not establish that it improved skills or caused the fall in support requests. Consider planning a second small trial with more hands-on examples and a simple skill check. A monthly program is not yet supported by the available evidence or an approved budget.
Participation and feedback
24 of 40 invited staff attended, a 60 percent attendance rate. Attendance was optional. [A]
18 of the 24 attendees returned feedback, a 75 percent response rate. Of those 18 respondents, 14 marked the session useful, 3 partly useful, and 1 not useful. Nine respondents asked for more hands-on examples. The positive rating is about 78 percent of respondents, not 78 percent of all invited staff. [B]
Operational observation
Support requests fell from 36 in the preceding seven days to 30 in the following seven days, a decrease of 6, or about 16.7 percent. The log does not track changes in workload, staffing, or who submitted requests. We cannot attribute this change to the session. [C]
Costs and limits
Materials cost 180 units. There was no separate room charge. Preparation took 6 staff hours, whose monetary cost was not recorded. The full cost of the session is therefore unknown. No budget for a repeat session has been approved. [D]
The pilot did not include a before-and-after skill test. Feedback came from 18 attendees and may not represent attendees who did not respond or staff who did not attend. [A, B]
Recommendation, proposed
Plan one further small trial before considering monthly sessions. Use practical support examples and a short task completed before and after training. Define what improvement would count as useful before running the trial. Estimate the full cost, including staff time, and ask for approval before booking it.
Proposed next steps
Ask the manager whether a second trial should be planned. If approved, agree an owner, participant group, success measure, budget, and date. Track staffing and workload if support-request counts will be used again. Owners and dates are not yet assigned.
Decision needed
Approve planning for a second trial, request more evidence first, or stop after the initial pilot.Fix a report that overstates the result
A weak illustrative draft might say: The training improved staff performance by 17 percent, received 78 percent staff approval, and cost only 180 units. We recommend a monthly rollout starting next month. Each claim sounds businesslike, but each goes beyond the source.
The 17 percent figure describes the change in a request count, not a measured improvement in performance. The 78 percent figure applies to survey respondents, not all staff. The 180 units exclude staff time. A monthly rollout has no approved budget, date, or agreed owner.
A focused correction prompt would say: Replace these conclusions with the source-supported measures. Name the denominator for feedback, describe the request-count decrease without claiming a cause, and label materials separately from total cost. Remove the unapproved start date. Keep the proposed second trial as a recommendation, not a decision.
After that repair, inspect the summary as well as the findings. An overclaim can survive at the top even after a detailed section is corrected. Every version of the conclusion should tell the same evidence-based story.
Keep facts, interpretations, and recommendations distinct
A fact in this report is that 24 people attended. An interpretation is that the feedback suggests attendees who responded saw value. A recommendation is to plan another trial. The three can sit near each other, but readers should be able to tell which is which.
Use clear wording rather than formal labels in every sentence. The attendance record shows introduces a fact. This suggests introduces an interpretation with limits. We recommend introduces a proposal. Avoid proved, confirmed, and guaranteed unless the evidence actually supports that strength.
Recommendations should connect to a finding and a limit. More hands-on examples respond to nine requests in the feedback. A skill check responds to the lack of evidence about learning. A cost estimate responds to the missing value of staff time. This makes the proposal easier to assess.
Include a reasonable alternative when the decision has one. The manager could ask for more evidence before another session or stop after the pilot. A report should clarify those choices rather than making the proposed option seem inevitable.
Choose detail that helps the reader
Keep the conclusion near the top, but do not squeeze all the evidence into it. The reader should quickly understand the answer and then find the support below. A summary can mention positive feedback and unproven impact, while the findings show the actual counts.
Use tables when they make a comparison easier. A table of measures, source, and limit works well here. A table with long paragraphs in every cell can be harder to read than a short section. Choose the form for the information, not because reports are expected to contain tables.
Define unfamiliar terms in plain language. Denominator means the group you divide by, such as the 18 people who answered the survey. A baseline means the starting measure used for comparison. Often you can avoid the terms entirely by writing 14 of 18 respondents.
Keep supporting detail that another reader needs to reproduce a result. Show the starting and ending counts for a percentage, the covered dates, and which costs were included. Remove decorative sentences that merely repeat the summary in different words.
Handle gaps and conflicting records honestly
Missing evidence does not always block the whole report. It changes what the report can conclude. In this example, the full cost is unknown, but the recorded materials cost is still useful. The report can identify the gap and make estimating staff time part of the next step.
If two sources disagree, record both figures and their labels. An attendance sheet might show 24 people while a meeting note says 25. Ask which count is final and what each includes. Perhaps one counts a trainer, or perhaps one contains an error. Do not let ChatGPT average them or pick the more convenient one.
Use an explicit placeholder when the reader needs a fact you cannot yet supply: date to be agreed or total cost not recorded. A placeholder is better than an invented date. Before sharing, decide whether the placeholder is acceptable or whether the missing fact needs resolution first.
Do not turn every uncertainty into a long defensive paragraph. State the gap where it affects the conclusion. The reader needs to know what is known, what is not, and whether the missing piece changes the decision.
Check and share the report
Check the report from the source pack outward. Verify names, dates, units, totals, percentages, and any statement about approval. Then check the reasoning: does the recommendation follow from the findings, and do the limitations weaken a stronger conclusion elsewhere?
Ask ChatGPT for a claim list if the report is long. Each row can show a key claim and its source label. Use that list to navigate your own records. It is a checking aid, not independent proof that the report is accurate.
Read the report once as the intended audience. Can you identify the decision, the strongest evidence, the main limit, and the proposed next step? If the answer is buried, shorten the opening and move the essential point upward. Preserve the evidence rather than adding more formal phrasing.
When a decision is made, update the action record with the actual approval, owner, and date. Writing proposed next steps did not assign them. A useful report ends with a clear choice and a truthful account of what happens next.
Your next useful step
If the report's next steps are approved, turn them into an action list with real owners and dates. Keep the report's proposed actions separate until that decision is made.
Check the result before you use it
- Each important factual claim traces to a supplied record.
- Percentages name the right group and calculations match the figures.
- Findings, interpretations, and recommendations are clearly different.
- Limitations appear near the conclusion they affect.
- Proposed work is not presented as approved, assigned, or completed.
Watch for: Reports often become too certain when notes become polished prose. A small trial does not prove a wider effect, an opinion does not establish a cause, and a suggested action does not create an approval.