Why every M&E prompt
needs an ethics check

We’ve all seen the excitement around generative AI in monitoring and evaluation. Drafting interview protocols, synthesising literature, even suggesting codes for qualitative data — AI can do it in seconds. But in our rush to save time, are we introducing ethical blind spots?

AI models are not neutral. They inherit biases, they can expose vulnerable populations, and they can produce outputs that look authoritative but are culturally tone‑deaf or even harmful. That’s why EvalCommunity has launched a new interactive tool: Ethical Considerations for Prompt Creation — a structured, 33‑item checklist built around the ethical principles that evaluators already trust.

The tool is not a theoretical document — it’s a working checklist. You open it, work through six sections (Do No Harm, Vulnerable Populations, Cultural Sensitivity, Data Protection, Bias & Representation, Transparency), tick what your prompt already covers, and see your ethics score in real time. Miss a critical item? The tool shows you exactly what to fix, with concrete examples and links to authoritative guidance.

Why the six domains?

Because evaluation ethics are not optional. The UNEG Ethical Guidelines, the AEA Guiding Principles, and the OECD AI Principles all stress the same message: rigour without ethics is not rigour. AI prompts are now part of our methodology; they deserve the same ethical scrutiny as a survey instrument or an interview protocol.

Here’s a glimpse of what each section covers — and why it matters for your next evaluation.

• Do No Harm — does the prompt avoid requesting information that could stigmatise, expose, or harm?
• Vulnerable Populations — have you added explicit child protection / refugee / disability safeguards?
• Cultural Sensitivity — does the prompt instruct the AI to flag where its training data may be insufficient for this context?
• Data Protection — are you complying with GDPR / minimisation / informed consent?
• Bias & Representation — does the prompt require disaggregation and surface minority perspectives?
• Transparency — have you disclosed AI use to donors and documented your prompts?

“But I’m only using AI to summarise…”

Even simple summarisation prompts can cause harm if they collapse diverse voices into one “average” finding, or if they strip away the context that gave the data meaning. The checklist applies to any use of AI in evaluation — from generating codes to drafting recommendations. Because once an AI output enters your report, it becomes part of the evaluation record, and you are accountable for it.

The tool doesn’t just give you a score — it teaches. Every item includes an expandable “Why this matters” box with concrete examples, such as:

The checklist also links directly to authoritative sources: UNEG, AEA, OECD.AI, OHCHR, and EvalCommunity’s own AI‑Ready Evaluator course. You’re not just guessing — you’re building prompts aligned with the global standards that funders and ethics committees expect.

The tool is self‑assessed and does not replace institutional ethics review. Use it as part of your wider quality assurance.