Gains:
- Knows the parts of a good public report and produces a consistent report skeleton and executive summary with AI.
- It catches exaggeration and ghost line errors by comparing every number and claim in the executive summary to raw data.
- Recognizes misleading visualization in table and chart suggestions and verifies the source of each cell of the report.
Public institutions produce reports: activity report, audit report, briefing note, situation assessment, project progress report, parliamentary presentation, annual statistical summary. These documents are the institution's memory, evidence of its accountability, and the eyes of decision makers. But writing a report takes time: collecting the data, establishing the structure, writing the initial text, making different sources consistent, simplifying, formatting. Here, AI provides great speed in establishing the structure of the report, converting scattered notes into organized text, summarizing long text and translating technical language into managerial language. In this unit, you will see how to run the entire document production flow with AI, from raw data to the approved report, and how to stay disciplined on the most critical issue — the accuracy of every number and claim in the report.
Anatomy of a good public report
A public report usually consists of the following layers: executive summary (main findings and recommendations that the decision maker can read in 1 minute), scope and method (what the report examined and with what data), findings (numbers, tables, findings), evaluation (meaning of the findings), and recommendations/conclusion. AI helps at each of these layers, but their roles are different:
- Structure and executive summary: AI is very powerful — it summarizes scattered findings, highlights the important.
- Findings (numbers): AI should only edit verified data you provide; should not produce numbers.
- Evaluation and recommendation: AI produces draft; Comment and responsibility are yours.
Key principle: Don't make the AI produce numbers, make it edit the numbers you give it. Every figure included in the report must come from the institution's own records or official statistics; The AI just needs to convert them into sentences, tables, and summaries.
Tip: In the report prompt, set the rule "do not generate new numbers, rates or dates; only use the data I provide; if you need a number that is not in the data, write '[data required]'". This greatly cuts out the risk of fabricated statistics.
Step by step report flow
- Determine the purpose and readership. To whom is the report written and for what decision is it made?
- Collect verified data. The numbers are consolidated somewhere along with their sources.
- Have the AI build the framework. Headings, subheadings, what will happen in which section.
- Produce a chapter-by-chapter outline. Only with the given data; AI should not make up numbers.
- Print executive summary last. When the text is finished, a summary is made; so it is consistent.
- Validate and consistency check. Does each number match its source; Do the sections conflict?
- Simplify and format; Submit for approval.
three mini cases
Case 1 — Activity report period shortened. The annual activity report of a directorate was normally produced within 3 weeks, with the effort of a few people. When we had AI create the section skeleton and first drafts and the team focused only on data validation and interpretation, the time was reduced to 8 days; content quality was found "more consistent" in the audit.
Case 2 — Fake rate caught. In one draft, the AI wrote the sentence “satisfaction increased to 87%”; whereas the team provided no such survey data — the model had fitted a “reasonable” number. When the source was asked during the consistency check, it was seen that the number had no basis and was removed.
Case 3 — Translation into executive language. A technical infrastructure report was going to go to the city council, but most of the members were not engineers. AI translated the technical text into a jargon-free executive summary (preserving the meaning); The questions in the parliament decreased by two thirds, and the decision was taken in the same session.
Four copyable templates
1) Report structure skeleton:
Your role: public reporting specialist. Create a report framework for the following purpose and readership: headings, subheadings, and what information will be in each section. Don't text yet. Mark which sections will require verified data. PURPOSE: [purpose] READER: [who is the decision maker] SCOPE: [topic]
2) Data faithful section writing:
Write the text of this section with the VERIFIED DATA below. GENERATE new number, rate or date; Just use the data I give you. Write "[data required]" where missing data is required. Use formal but fluent language. SECTION: [title] DATA: [numbers and sources]
3) Executive summary (latest):
Create an executive summary from the full report text below: no more than 200 words, 3-5 most important findings and recommendations. Do not add any information that is not in the text. Write in a simple manner that the decision maker can read in 1 minute. REPORT: [full text]
4) Consistency and number checking:
Check the following draft of the report: (1) is a source cited for each issue, (2) is there any conflict between sections, (3) does the executive summary match the body, (4) is there an unsourced/dubious claim. Don't rewrite the text, just give a list of warnings.DRAFT: [text]
Weak prompt / Strong prompt
Weak: "Write an activity report with this data."
Strong: "Your role is public reporting specialist. First, build a report skeleton for the city council (headings and subheadings). Then write each section with ONLY the following verified data; do not produce new numbers, rates, or dates, write '[data required]' where missing data is needed. Finally, produce a 200-word executive summary. When finished, provide a checklist for intersectional conflict and unsourced numbers. Data: [numbers and sources]."
Difference: strong prompt chains structure, data fidelity, number-shaming ban, summary, and self-control; The output is a verifiable and consistent outline.
AI trust by report section
Section
AI trust
rule
Structure/skeleton
high
Use freely
Executive summary
high
Limit to text
Language / simplification
high
Meaning must be preserved
Findings/numbers
low
Data given only
Review / recommendation
medium
Draft + human comment
Table, indicator and executive summary
The impact of a public report is often determined on the first page: the executive summary (the section that summarizes the report's main findings, conclusions, and recommendations on one page) may be the only part the higher authority will read. AI is powerful at reducing a long report to a faithful executive summary; but you need to check line by line that the summary accurately reflects each number in the report — because when summarizing, the AI may round numbers, drop a condition, or overstate a finding. Likewise, AI can generate table skeletons and chart suggestions from your raw data; However, the source of every cell in the table must be found in your data, and any rows that the AI adds "just because it seems reasonable" should not make it into the report.
Mini case — magnified the summary issue. While the actual success rate was 63 percent in an annual report, the executive summary produced by YZ increased the perception to 90 percent by saying "the vast majority of applications resulted positively." The audit replaced this expression with the actual number; Because an exaggerated positive perception in the public report puts the institution in a difficult situation in subsequent audits.
Mini case — ghost row into table. AI had added an "Other" category and a made-up number to the monthly application sheet, which did not exist at all in the institution; It totaled but the line was not real. Comparison with raw data caught this ghost line.
Template that produces verifiable tables and summaries:
Task: Produce a table and 5-item executive summary from the following raw data (use only this data).Raw data: [paste table/CSV]Rules:- Do not add any rows/categories to the table that are not in the data.- Show with parentheses which row in the table each number in the summary comes from.- Write "data insufficient" where in doubt, do not guess.Output: (1) Table (2) Executive summary (3) Basis for each summary sentence.
Report item
AI contribution
Mandatory human supervision
Executive summary
Drafting and simplification
Does each issue/claim comply with the report?
Table
skeleton, form
Correspondence of each cell in raw data
Chart suggestion
Appropriate chart type
Misleading no scale/axis?
Suggestion section
Option generation
Authority and regulatory compliance
Tip: Aggregate the raw data yourself in a separate location before printing the report to the AI; then compare the AI's summary to these correct totals. Inspection is done simultaneously with production, not after it.
Common mistakes
- Making AI produce numbers. The most dangerous mistake; Only verified data enters the report, AI edits it.
- Writing the executive summary first. If the summary is omitted before the body is finished, it will contradict the text; leave it for last.
- Not checking consistency between departments. Numbers in different sections should match.
- Not citing sources. Every table and figure must be traceable to its source; control looks for this.
- Leaving the technical language as is. If the decision maker does not understand, the report loses its function; Simplify but do not distort the content.
- Skipping the confirmation step. The report is not published without the reading and approval of the responsible person.
In summary
In report production, AI provides great speed in building structure, drafting, summarizing and language simplification. But every number that goes into the report must come from the institution's verified data; AI should not produce numbers, only edit them. The executive summary should be prepared last, inter-departmental consistency and resources should be checked, and the document should be approved. AI speeds up the report; You carry the accuracy and signature.
Application task
Build a skeleton for your upcoming real report with the "Report structure skeleton" template. Print a section with a few verified data using the "Data faithful section writing" template and check if the AI is making up numbers despite the rule. Finally, apply the "Consistency and number checking" template to your own draft.
checklist
- [ ] Every number included in the report came from a verified source; AI did not produce.
- [ ] I set up the structure skeleton and printed the sections faithfully to the data.
- [ ] I made the executive summary last, limited to text.
- [ ] I oversaw interdepartmental consistency and resources.
- [ ] I simplified the technical language without spoiling the content.
- [ ] I submitted the report to the person in charge for approval.