How Nonprofits Should Write Prompts for AI Images

AI image generators can create campaign graphics, social media illustrations and website visuals within seconds.

The quality and safety of those images depend heavily on the prompt.

A weak prompt may produce stereotypes, misleading scenes, inaccessible designs or images that appear to document real beneficiaries. A stronger prompt gives clear instructions about purpose, audience, visual style, representation and ethical limits.

Start With the Purpose

Define whether the image will promote an event, explain a process, illustrate an abstract concept, support a fundraising campaign, create a report cover or provide a social media background.

The purpose determines what kind of image is appropriate.

Describe the Composition

Explain the main subject, background, camera angle, lighting, orientation, negative space, position of people or objects and focal point.

Harvard’s guidance recommends clearly describing the subject, setting, composition, style and other visual requirements, then refining through iteration. (huit.harvard.edu)

Specify the Visual Style

Define whether the nonprofit wants flat vector artwork, editorial illustration, infographic style, documentary photography, watercolor, minimal line art, realistic photography or abstract geometric imagery.

Realistic photography creates the greatest risk of confusing viewers. An obvious illustration is often safer.

Address Representation Directly

Consider race and ethnicity, age, gender, disability, body type, clothing, family structure, economic status and cultural setting.

Avoid prompts that define people only through vulnerability. Portray people with dignity and agency.

Do Not Generate Fictional Beneficiaries as Evidence

Nonprofits should avoid using realistic AI-generated people in ways that suggest they are actual participants.

WITNESS recommends consent, transparency and labeling when generated media could be misunderstood. (blog.witness.org)

Include Accessibility Requirements

Request high contrast, an uncluttered composition, a clear focal point, limited small details and space for readable text.

It is usually better to generate the visual without words and add approved text later.

Add Explicit Prohibitions

Depending on the campaign, prohibit visible logos, graphic injury, exaggerated sadness, identifiable children, stereotypical poverty imagery, text, distorted hands, medical equipment or recognizable public figures.

Use a Repeatable Prompt Structure

A practical prompt should define:

• Purpose
• Format
• Subject
• Composition
• Style
• Representation
• Accessibility
• Restrictions

Store Approved Prompts and Brand Rules

A secure AI Knowledge Hub such as Maisy can organize approved brand standards, image policies, consent requirements, accessibility rules and reusable prompt templates inside Microsoft 365 and SharePoint.

Maisy does not approve images automatically. It gives staff access to the policies and organizational knowledge required to make responsible decisions.