Reports, sitreps and minutes: writing faster without losing accuracy

1The right workflow

For a report, the classic mistake is to ask AI to "write the report". The right workflow has four steps:

• 1. Gather the material: monitoring data, field notes, meeting minutes, previous report, donor template. You bring the facts.
• 2. Structure: ask for an outline matching the required template, then validate it.
• 3. Draft section by section, from the material provided only, requiring that every figure be reproduced as is and that gaps be flagged.
• 4. Check and finalise: fact-check, consistency between sections, tone, length, then review by a second person for binding reports.

With this method, teams often report significant time savings on drafting, but not on checking, which remains fully necessary. Plan for it in the schedule.

2The situation report (sitrep)

A sitrep must be short, factual, dated and sourced. AI helps turn scattered notes into a structured document:

• Key points (3 to 5 lines);
• Situation: changes since the last report, with figures and sources;
• Needs and gaps;
• Response: what has been done, by whom, where;
• Constraints: access, logistics, funding;
• Next steps.

Standard request. "From the notes below, write a sitrep in the following format [format]. Reproduce the figures exactly. Clearly distinguish what is confirmed, what is reported by a single source and what is estimated. Give the date and source of each figure. Do not mention any person's name or any team location. 400 words maximum."

Watch out for security information: a widely shared sitrep must contain no routes, no times, no names and no incident details that could identify people. That information stays in security channels.

3Meeting minutes

Coordination, cluster or team meetings produce often disorganised notes. AI is good at turning notes or a transcript into structured minutes: participants (by organisation), points discussed, decisions, actions with owner and deadline.

Points of vigilance:

• Recording and transcription: only record a meeting with the participants' agreement and according to the organisation's rules. Some tools (for example the note-taking features of Google Meet with Gemini, depending on the subscription) do it automatically: inform participants.
• Decisions and actions: check that they correspond to what was actually decided. AI can turn an idea that was mentioned into a decision.
• Sensitive remarks: before circulation, remove remarks about individuals, incidents or negotiations.

Useful request. "List separately: 1) the decisions taken; 2) the points discussed without a decision; 3) the actions with owner and date. If the owner or date is not clear in the notes, write 'to be confirmed'."

4Checking: the four-check method

Before sending a text written with AI, apply four checks:

• 1. Figures: every number is found in the source, with the same unit, the same date and the same area. Recalculate totals and percentages.
• 2. Names and terms: organisations, places (official spelling), people's titles, acronyms, donor terminology.
• 3. Statements: every sentence asserting a fact or a result must be supported by the data. Remove unjustified adjectives ("remarkable", "unprecedented").
• 4. Omissions: has the AI left out a problem, a delay, a gap against targets? Reports must present difficulties honestly.

Tip: ask the assistant itself, in a new conversation, to compare the final text with the source data and list any difference. It is a useful aid, but it does not replace your own check.

Control calculation. A report states "3,450 people reached, i.e. 115% of the target of 3,000". 3,450 / 3,000 = 1.15: correct. But the source table shows 3,450 contacts, including people counted twice. The right indicator is the number of unique people: the text must be corrected, not just the calculation.

5Adapting to each audience

The same content often needs to be adapted:

• Donor: compliant with the template, focused on results, indicators, variances and budget.
• Management: one page, decisions to be taken first.
• Cluster: aggregated data, geographical coverage, gaps.
• Communities: plain language, local language, what has been done, what will change, how to give feedback or complain.

AI excels at these rephrasings: "Turn this report into a one-page note for management, with three decisions to be taken at the top." or "Summarise these results in 120 words in easy French, end-of-primary-school level, for a poster at distribution sites."

For communities, the final version must be validated by people who speak the language and know the context (week 8).

6What not to do

• Let AI "fill in" missing data or estimate figures you do not have.
• Copy "case studies" or testimonies invented by AI. A testimony must be real, collected with consent and anonymised if necessary.
• Send a report in which you have not read every sentence.
• Standardise all your reports with the same phrases: donors quickly spot generic texts, and it damages credibility.
• Paste named beneficiary tables to "calculate the indicators".

A fast but wrong report costs more than a slow one: loss of credibility, requests for clarification, and sometimes refunds.