How to Draft Evaluation Reports with AI
Stop staring at a blank page. A 4-phase workflow turns your completed analysis into donor-ready evaluation narrative in hours, not days.
A structured drafting workflow means you spend less time fighting blank-page paralysis and more time on the analysis that actually matters. Every section starts from evidence, not from scratch.
The 4-Phase Drafting Workflow
Each phase builds on the previous one. Follow this sequence and your first usable draft arrives in hours, not days.
Structure
Organize analysis outputs into a 2-3 page summary per evaluation question: key statistics, coded themes, illustrative quotes, and triangulation notes. Structured inputs produce focused drafts.
Draft
Feed one section at a time (findings, conclusions, recommendations) with full context: program type, donor framework, audience, and word count. Generate 2-3 versions, then combine the strongest elements.
Verify
Cross-check every statistic against your original analysis. Confirm quotes are verbatim. Flag causal language the AI overstated ("caused" vs "contributed to"). If a claim lacks a data source, cut it.
Polish
Add contextual insights AI cannot know: political dynamics, cultural factors, implementation realities. Apply donor-specific formatting and citation styles. Run a final consistency check across sections.
What Strong Report Sections Look Like
Side-by-side examples showing how the drafting workflow transforms raw analysis into polished evaluation narrative.
Findings Paragraph
"The program improved water access. Many participants said water was easier to get. Survey data showed improvement. Women and girls benefited from reduced travel time." No numbers, no sources, no rigor.
Findings Paragraph
"Household access to improved water sources within 30 minutes increased from 34% to 71% (n=600). 11 of 12 focus groups cited reduced time burden for women and girls as the most significant change."
Recommendation
"The program should continue its good work and consider expanding to new areas while also improving sustainability and addressing gender issues." Vague, no priority, no timeline, no cost.
Recommendation
"[Critical] Develop a 12-month exit strategy with costed handover plan by Q2 2025. Only 22% of farmer groups can currently operate without external support, putting $2.1M in market linkage gains at risk."
Executive Summary
"A quasi-experimental impact evaluation using propensity score matching found statistically significant improvements in household resilience with heterogeneous treatment effects across subgroups." Jargon for the sake of jargon.
Executive Summary
"Households in the program were 20% more resilient to flooding than similar households outside it. The program cost $285 per household and produced gains valued at $620, a strong return on investment."
5 Rules for Stronger Report Drafts
Draft one section at a time
Target findings, conclusions, or recommendations separately. Each requires a different writing approach and prompt structure. Start with findings where evidence is clearest.
Anonymize before you prompt
Remove names, locations below district level, and any personally identifiable information before pasting evaluation data into AI tools. Use aggregated statistics and anonymized quotes only.
Specify your audience explicitly
"For USAID technical reviewers" produces different language than "for community stakeholders." The same finding needs different framing for each reader.
Set exact word counts per section
Say "300-word findings paragraph" not "write about this finding." Without length constraints, AI generates 800 words of padding when you needed 300 words of evidence.
Generate multiple versions and combine
Run 2-3 prompts for the same section with different instructions. Pick the strongest opening from version 1, the best evidence integration from version 2, the clearest conclusion from version 3.
Copy-Paste Report Drafting Prompt
Use this template to draft any evaluation report section. Fill in the bracketed fields and paste into ChatGPT, Claude, or Gemini.
I am writing the [REPORT SECTION: findings / conclusions / recommendations / executive summary] for a [EVALUATION TYPE: midterm / endline / impact] evaluation. Evaluation design: [YOUR METHODS, e.g., 'mixed methods, 400 household surveys + 12 FGDs + 15 KIIs'] Audience: [YOUR AUDIENCE, e.g., 'program managers and donor technical staff'] Data summary for this section: - Quantitative: [YOUR QUANTITATIVE DATA, e.g., 'access to clean water increased from 34% to 68% (p<0.01)'] - Qualitative: [YOUR QUALITATIVE DATA, e.g., 'community ownership theme mentioned in 8/12 FGDs'] - Key quote: "[YOUR KEY QUOTE, e.g., 'Before the project we walked 3 hours for water']" - [PARTICIPANT DESCRIPTION, e.g., 'female head of household, age 35'] - Triangulation: [YOUR TRIANGULATION, e.g., 'survey data confirms FGD themes on improved access'] Draft a [TARGET WORD COUNT, e.g., '1500']-word section that: 1. Presents evidence before interpretation 2. Integrates both quantitative and qualitative data 3. Uses neutral, evidence-based language (avoid causal claims unless supported) 4. Acknowledges limitations in the data
Put It Into Practice
Start with your strongest finding and use the prompt template above. Our free AI-powered M&E tools can help you structure your evaluation outputs.
Related Quick Guides
How to Write AI Prompts for M&E
The 4Cs Framework for prompts that produce donor-ready outputs on the first try.
Read guideHow to Clean M&E Data with AI
Turn 15 hours of manual cleaning into 2 hours with a 4-step workflow.
Read guideHow to Code Qualitative Data with AI
A structured workflow for coding interview transcripts with AI assistance.
Read guide