AI for Social Good

By the end of this lesson you’ll be able to describe, in your own words, how AI is being used to help people rather than just to sell things — and you’ll have used a free AI tool yourself to do one small piece of good-in-the-world work. You’ll also be able to spot the difference between a genuine AI-for-good project and marketing dressed up as one.

The idea in plain English

“AI for social good” is a plain label for a simple thing: using AI tools to help people who need help. Not to boost a company’s quarterly numbers, not to write better ad copy. To translate a medical form for someone who doesn’t read English. To help a two-person animal rescue write grant applications they’d never have time to write. To let a blind person point a phone at a can of soup and hear what it is.

Here’s the part that surprises most beginners: almost none of this requires new technology. The same free ChatGPT or Claude account you might have poked at once is the engine behind an enormous amount of this work. What changes isn’t the tool. It’s who’s pointing it and at what.

That matters for you specifically. The last module covered AI in your everyday life — your phone, your email, your maps. This lesson is about the same skills pointed outward. If you volunteer anywhere, sit on a board, help at a church or a school, care for an aging parent, or just have a neighbor who’s drowning in paperwork, you already have somewhere to point these tools.

One honest caveat before we go further: AI is genuinely useful here, and it is also genuinely oversold here. Some of the most-hyped “AI saves the world” projects have delivered very little. The real wins tend to be smaller and less glamorous — a translation, a summary, a draft, a description — repeated thousands of times by people who didn’t have the hours to do it by hand.

How it actually works

Most AI-for-good work falls into a handful of patterns. Once you can name them, you’ll see them everywhere.

Removing a language barrier. Refugee-resettlement volunteers, hospital intake desks, and school offices use tools like ChatGPT and Google Translate to move between languages on the spot. AI translation isn’t perfect and shouldn’t be trusted for a legal contract or a diagnosis, but for “here is what this letter from the landlord says,” it’s the difference between understanding and not.

Turning something dense into something readable. A lot of the paperwork that governs vulnerable people’s lives is written badly on purpose or badly by accident: benefits letters, insurance denials, lease agreements, discharge instructions. Asking an AI to rewrite a document at a sixth-grade reading level is one of the highest-value five-minute tasks that exists.

Giving small organizations big-organization capacity. A food bank with three staff can’t afford a grant writer or a communications person. AI drafts don’t replace those roles, but they take a volunteer from “I have no idea how to start this application” to “I have a rough draft I can fix.” That’s the actual unlock — beating the blank page.

Describing the world to people who can’t see it. The app Be My Eyes pairs blind and low-vision users with AI that describes photos out loud — the expiry date on a carton, which door is the entrance, what’s on a restaurant menu. This is one of the clearest, least-hyped wins in the whole field.

Finding the signal in a pile of data. Conservation groups use AI to sort millions of camera-trap photos and flag the ones with animals in them. Researchers use it to pick out illegal-logging chainsaw sounds in rainforest audio. Humans still make the calls; AI just does the sorting no human has time for.

The mechanism underneath all five is the same: AI is very good at handling a large volume of language, images, or sound quickly and roughly, and quite bad at being certain. So the shape that works is AI drafts or sorts, a human checks and decides. When you hear about an AI-for-good project that skips the human check — especially one making decisions about someone’s benefits, sentencing, or medical care — that’s the warning sign, not the innovation.

Try this now

You’re going to do a real piece of this work. It takes about ten minutes and you need nothing but a free account at ChatGPT, Claude, or Gemini. Any of the three is fine.

1. Pick a real cause you have some connection to. Your kid’s school, a local shelter, your church’s food pantry, a rescue you follow. Real beats hypothetical — you’ll be able to judge whether the output is any good.
2. Do the plain-language rewrite. This is the single most useful skill in this lesson. Find a chunk of dense text — a real one if you have it, or use the sample below — and paste this in:

   You help nonprofits communicate clearly with people who
   have low literacy or who are reading in a second language.
   
   Rewrite the text below so a 12-year-old could understand it.
   Rules:
   - Short sentences. One idea each.
   - No jargon. If a term is unavoidable, define it in
     parentheses the first time.
   - Keep every deadline, dollar amount, and phone number
     exactly as written. Do not round or change numbers.
   - Keep the same meaning. Do not add reassurance that
     isn't in the original.
   
   At the end, list anything you were unsure about, so a
   human can check it.
   
   TEXT:
   "Applicants must submit documentation substantiating
   household income for the preceding 90-day period.
   Failure to provide requisite verification within 14
   calendar days of notification will result in
   administrative closure of the pending application,
   requiring reinitiation of the application process."

   Read what comes back. Notice that it flags its own uncertainties at the end — you asked it to, and that turns it from a black box into something a volunteer can actually check.

3. Now do the capacity task. Draft something your cause genuinely needs. Fill in the bracketed parts with real details:

   I volunteer with [name of organization], a [size, e.g.
   "3-person"] nonprofit in [city] that [what it does in one
   sentence]. Our biggest problem right now is [the real
   problem, e.g. "we get 40 donation offers a month and no
   one has time to reply"].
   
   Do two things:
   
   1. Write me a short, warm email template we can reuse to
      respond to donation offers. Plain language, no
      corporate tone, under 150 words.
   
   2. Then list 5 other tasks like this one where a free AI
      tool could save us a few hours a week. For each, say
      what could go wrong if we don't check the output.
   
   Ask me any questions you need before you start.

4. Check the output like it’s going out under your name. Because it is. Every number, date, name, and phone number gets verified against the source. Every factual claim about the organization gets read by someone who knows the organization.
5. Offer it to a real person. Email the draft to whoever runs that cause with a note: “I made a first draft of this, feel free to trash it.” Ten minutes of your time, and the blank page is gone for them.

Common mistakes

• Letting AI make the decision instead of the draft. Using AI to summarize forty grant applications is fine. Asking it which applicant deserves the money is not. AI has no way to tell you it’s confused — it produces fluent text either way. Fix: AI narrows and drafts, a human decides. Always.
• Pasting other people’s private information into a chatbot. Volunteers do this constantly and don’t realize it’s a problem. A client’s medical history, an immigration case number, a donor list, a child’s name and school — once pasted, it’s on someone else’s server. Fix: strip names and identifying details before pasting, or replace them with placeholders like [CLIENT]. If the information is legally protected, don’t paste it at all — check the organization’s policy first.
• Shipping the AI’s voice instead of the organization’s. AI defaults to a bland corporate register that makes a scrappy local charity sound like a bank. Donors notice. Fix: paste in two things the organization has actually published and add “match this voice.” Then cut every sentence that doesn’t sound like a person.
• Trusting invented specifics. Ask an AI for grant statistics and it will sometimes produce a very convincing figure that does not exist. Same with citations, funder names, and deadlines. Fix: treat every number and source as unverified until you’ve found it yourself. Use Perplexity when you need something with a link you can click through to.
• Assuming the small stuff isn’t worth doing. People skip this work because it doesn’t feel like curing anything. But a translated letter and a readable benefits notice are the actual texture of most AI-for-good work. Fix: do the small thing this week rather than planning the big thing forever.

Key takeaways

• AI for social good is mostly the same free tools you already have, pointed at people who need help instead of at profit.
• The five recurring patterns: removing language barriers, simplifying dense documents, giving small organizations capacity, describing the world to people who can’t see it, and sorting large piles of data.
• The reliable shape is AI drafts or sorts, a human checks and decides — a project that skips the human check is a red flag, not a breakthrough.
• Never paste someone else’s private information into a chatbot, and never trust a number or citation you haven’t verified yourself.
• Small and repeated beats ambitious and hypothetical. One rewritten letter this week is real work.