How to Use ChatGPT for M&E
ChatGPT can draft indicators, analyze qualitative data, and write donor reports -- if you know how to direct it. A practical guide for M&E practitioners.
ChatGPT is a general-purpose tool, not an M&E specialist. The practitioners who get the most from it treat it like a capable assistant who needs clear, specific instructions, not a tool they hand vague tasks to and hope for the best.
What ChatGPT Is Actually Good At for M&E
ChatGPT handles some M&E tasks well and others poorly. Knowing the difference saves time and prevents bad outputs from entering your work.
Drafting and structuring
ChatGPT excels at turning rough notes into polished text: indicator definitions, TOR sections, evaluation report paragraphs, donor narrative updates. Give it structure and context, and it produces usable first drafts consistently.
Qualitative analysis support
Coding interview transcripts, summarizing FGD notes, identifying themes across multiple text documents. Works best on batches under 5,000 words. For larger datasets, process in chunks and merge themes manually.
Survey and instrument design
Generating question sets, checking question wording for bias, adapting surveys for different literacy levels, and converting open-ended questions to closed-ended formats. ChatGPT is faster than a blank page for first drafts.
Explaining concepts and frameworks
Translating M&E concepts for non-technical stakeholders, comparing framework types (logframe vs results framework), explaining statistical outputs in plain language. Useful for communication work.
Data organization and summarization
Cleaning small datasets, summarizing monitoring reports, extracting key findings from documents. For large datasets, use Python or Excel first -- ChatGPT context windows limit how much data it can process at once.
Prompts That Work vs. Prompts That Waste Time
The same task, two different prompts. The difference is specificity, context, and constraints.
Indicator Development
"Give me indicators for a health program." You get 10 generic indicators that could apply to any health program anywhere. None have data sources, definitions, or targets. All need rewriting.
Indicator Development
"Generate 5 SMART output indicators for a community health worker program in Ethiopia targeting 2,000 women of reproductive age. Each indicator needs: definition, data source, collection method, and baseline value placeholder. Align with USAID standard indicators where possible."
Qualitative Coding
"Code this interview transcript." You get a response that either summarizes the transcript or invents code categories with no explanation of the framework used.
Qualitative Coding
"Apply inductive thematic coding to this interview transcript. Identify 4-6 themes. For each theme: a short label (3-5 words), a 1-sentence description, and 2-3 direct quotes that support it. Format as a table."
Donor Report Narrative
"Write a progress report for my donor." You get a generic template that does not reflect your program, your results, or your donor's reporting style.
Donor Report Narrative
"Write a 300-word progress narrative for a USAID-funded water and sanitation program. Key results this quarter: 1,200 households with new latrines (target 1,000), 3 community hygiene training sessions completed. Key challenge: delayed procurement of materials due to import restrictions. Use active voice and quantify results where possible."
Rules for M&E Practitioners Using ChatGPT
Never paste beneficiary data
ChatGPT runs on external servers. Names, phone numbers, location data, and household identifiers must be removed or anonymized before any prompt. A data breach starts with a careless paste.
Name your donor and country
"USAID" or "FCDO" or "EU" instantly grounds the AI in the right reporting standards and terminology. "Ethiopia" or "Philippines" grounds it in the right context. Without these, outputs are generic.
Give it a role
Start with "You are an M&E specialist working on a [sector] program funded by [donor]." Role-setting improves terminology, depth, and relevance across the conversation.
Always verify outputs
ChatGPT hallucinates. It will confidently cite fake indicator definitions, invent baseline statistics, or misquote frameworks. Treat every factual claim as a hypothesis to check.
Iterate, do not start over
When an output is close but not right, ask for specific revisions in the same conversation: "Revise indicator 3 -- it needs to be measurable at community level, not household level." Iteration beats regenerating from scratch.
Copy-Paste Template for M&E Tasks
A single template that works across most M&E tasks. Fill in the brackets. The more context you provide, the better the output.
You are an M&E specialist working on a [SECTOR, e.g., 'food security'] program in [COUNTRY/REGION, e.g., 'rural Kenya']. Program context: - Funder: [DONOR, e.g., USAID / FCDO / EU / GATES] - Duration: [e.g., 3 years] - Target population: [e.g., 5,000 smallholder farmers] - Program objective: [e.g., 'increase household income through improved crop yields'] Task: [TASK, e.g., 'generate 5 SMART output indicators'] Output format: [FORMAT, e.g., 'table with columns: Indicator | Definition | Data Source | Frequency'] Constraints: [e.g., 'exactly 5 indicators, use USAID standard indicator terminology where available']
Ready to Try These Techniques?
Use our free AI-powered M&E tools and prompt library to put these approaches into practice immediately.
Related Resources
How to Write AI Prompts That Actually Work for M&E
The 4Cs Framework for prompts that produce donor-ready outputs.
Read guideHow to Choose the Right AI Tool for M&E
ChatGPT vs Claude vs Gemini -- which to use for which M&E task.
Read guideHow to Code Qualitative Data with AI
A step-by-step workflow for AI-assisted thematic analysis.
Read guideHow to Protect Data Privacy When Using AI
What's safe to share and what to anonymize before any AI prompt.
Read guide