Design priorities: reduce cost, connectivity and device requirements while respecting local knowledge and choice. A photograph of our fieldwork is not evidence of anyone’s poverty, and digital technology is not automatically better than paper.
Fictional example: a small team needs a water guide for readers with different literacy needs. It drafts plain language, checks three messages with adult volunteers without collecting names, and revises misunderstood parts. It can be printed and discussed without an account or continuous connection.
AI may suggest simpler wording or a translation; a local reviewer checks meaning and terminology. Machine translation is not a reviewed published language, and hidden subscription costs must not be transferred to users.
Before choosing a tool: can the task work without it? What data leaves the device? What does connectivity cost? Can people opt out? Who corrects errors? Is there a paper or human-assisted alternative?
Practical exercise: compare a paper card and a digital explanation for a general question. Assess cost, clarity, privacy, training and update effort. Choose what actually serves the audience; the result may be not using AI.
Start with a useful free resource and general feedback. Do not request sensitive stories, health information or identity numbers, or automate decisions about eligibility for assistance.