How to Nonprofit AI with Brenda Foster pt 2

Vanguard Communications’ Chief of Innovation Brenda Foster shared tips and practical advice on getting started using generative Artificial Intelligence AI tools at your nonprofit in a way that matches your mission and values.
Learn how to prompt, when and how to use AI tools, and when not to.
Learn how to evaluate the outputs and feel good using AI at your nonprofit.

In part 1, Brenda explains the various types of AI and walks through the ethical considerations and trade offs for the environment, community justice, human creativity, privacy and security, and bias. She presents a five question framework for creating your nonprofit AI policy. In part 2, Brenda explores good prompting and the differences between tools in this moment, and takes audience Q&A.

Are you wondering where to start with AI?

Chances are you and your colleagues are already using it for some things, and wondering how to use it better, or whether you should be using it at all. Your organization may be ambivalent or aghast at AI, have already embraced it, or be unsure where to start. You may have colleagues that are using AI for everything and others who won’t touch it.

Brenda Foster is a PRSA-NCC Hall of Fame inductee who has specialized in nonprofit communication for decades. 

In this webinar, she shares tips and best practices on improving your AI prompts for communication success and explores situations where AI can improve the day-to-day job satisfaction for nonprofit staff. You can hear more from Brenda in our podcast discussion of AI tips here.

How can your nonprofit get started ?

In this webinar learn how to prompt, when and how to use AI tools, and when not to. Learn how to evaluate the output and ensure that your team feels confident and comfortable using AI to make their jobs more interesting and to better support your mission.

As with all our webinars, this presentation is appropriate for an audience of varied IT experience.

Community IT is proudly vendor-agnostic, and our webinars cover a range of topics and discussions. Webinars are never a sales pitch, always a way to share our knowledge with our community.

Learn how to create an AI Acceptable Use Policy here. The nonprofit sector is deeply concerned with ethics, accountability, the environment, and systemic change. Learn more about ethical AI frameworks here.

_______________________________
Start a conversation :)

• Register to attend a webinar in real time, and find all past transcripts at https://communityit.com/webinars/
• email Carolyn at cwoodard@communityit.com
• on LinkedIn

Thanks for listening.

Transcript

Brenda Foster: We want to make sure that everything that's available to for-profit companies is available to nonprofits as well. So that's what caused us to start an AI task force a couple of years ago to sort of figure out A, what's going on with AI, what do they mean by AI. And then we really looked at a lot of the benefits as well as some of the concerns. And what we ended up with was what you'll see us, I'll talk about a lot today, which is ethical AI. The idea that if we're going to use it, we want to use it as ethically as possible and we know that our clients want us to as well.

Carolyn Woodard: Welcome to this Community IT webinar, How to Nonprofit AI with Brenda Foster. My name is Carolyn Woodard. I'm the Outreach Director for Community IT and I'm the moderator today. And I'm very happy to hear from our guest, Brenda.

Brenda Foster: Hi, everybody. I'm Brenda Foster, and I am the Chief of Innovation at Vanguard Communications. 

How to Use Nonprofit AI?

So now, you’ve got your policy, how are you going to use AI and how do you use it in a way that is most helpful to your nonprofit?

Prioritize Underserved Communities

The first thing shouldn’t be a surprise. 

To me, if you’re a nonprofit, you’re going to prioritize the needs of your underserved communities. So when you’re looking at all the ways that this can help you and that can help your cause, and take a look at how it can be helpful to the communities that you serve first. 

And then the other thing I would say is, you can advocate. You have, that’s what you do, your advocates. So if you’re going to take on AI and say, yes, this is something that we want, think about your role in making sure that companies feet are to the fire to continue to become as ethical as possible.

I think I’m done speechifying now. Can you tell I’m very passionate about this ethical stuff?

Ethical AI in Practice

Okay. So let’s talk about it in practice.

Make Work More Rewarding

So the first thing is happy employees, happy you, happy people like you. It really can automate repetitive tasks. So I will tell you, we’ve had, we have all kinds of little tasks that has been very, that have been, yes, you absolutely can have this. We have all kinds of little tasks that we have to do every single month that are very frustrating. 

And one of them is monthly reports. Look, I understand why clients want to know how we’re using their money. It’s totally fair. It’s just very tedious. And actually we’re using their money to write monthly reports, which I think is really dumb. So, but it’s been very hard because we have all of these notes that we take in our time sheets, and that’s how we build the monthly report. 

And we could not figure out for the life of us, how in the world do we pull this out and have AI write the monthly report? I finally figured out I could PDF those comments and upload it to, in this case, ChatGPT, and then I could take a monthly report from two months ago, give it to it and say, take this information and make it like this. 

And suddenly I had an 85 percent done monthly report, and I was so excited because the idea that I could free up, and this is usually falls to your most junior employees, who really want to get out of that, be a writer, looking at media clips or filing things or managing budgets, all of those things that absolutely you want them to have the skills to do. Not saying you don’t, but they’d rather have the bigger skills. They’d rather help you in a deeper way and feel like way more meaning. They do not get a lot out of those repetitive tasks. 

You really can create a personal assistant. It’s funny, one of the things I’ve gotten AI to do for me is I say, can you help me make a bot that will help me with this task? And it does. And it basically creates a little window where you put in the information and it tells it for you. 

I have like 20 little bots running for me for different tasks that I know I do every day. And it’ll do it for itself. So, it’s not like you have to figure out how to build it, it just builds it for you. 

We get clients all across the board, civil rights, education, farming, all kinds of things. And to learn their concepts quickly is really, really hard. So, I use AI to get me up to speed on reports that are coming out on news, like a week or a month of news that I need to absorb quickly. 

These concepts and processes, AI can really help you with that. I just got an 18-page report from the World Economic Forum that I have to do a video about next week. And I, it’s all so many words, I don’t know. And so AI, I uploaded it to AI and it told me what was in the report. And that’s very helpful. I’m writing a two-minute video on an 18-page report. I think that’s probably a better way to go. 

And then finally, when you really think about it, that work you want to do every day, you want to be in the field, you want to be talking to your clients, you want to be doing whatever it is that you do, that’s what you want to spend your time on. 

Increase Impact

So the next thing is we want to increase our impact. You can communicate so much faster, and also there’s a little microphone function on generative AI. You know how you don’t want to type, I want you to do this and this and this and this. Just tell it. And you can have ums and ahs and it doesn’t care. 

And then, it’s the same thing as if you were having a meeting, only it does what you want it to do. But by doing that, you can create an outline for a blog. You can come up with, take one piece of content and say, what are other ways I can use this content? 

I was in a meeting with a client where we were, the budget was over, right? And it was a training that we were doing and it was in a situation like this. And somebody said, do you have a tip sheet that you could give us? 

And while somebody else was talking, I took the PowerPoint and I said, make me a tip sheet and I sent it back to them. And it didn’t cost them anything or me. And they’re so excited because now I have a tip sheet I didn’t have before. 

It really allows you to help with crises. So we’ve had some client crises where I’ve run a few scenarios. What happens if I do this? What happens if I do this? What happens if I do this? Because before you were guessing, and you only have maybe one or two people’s background, right? But AI is looking at all kinds of scenarios that maybe you didn’t have access to. So sometimes in crises, it can help you choose of the four paths available. These two are the least risky. 

It also helps you find the latest research from the field. 

You Can Tell AI Not to Lie to You

Now, AI, I want to say AI is built to please. All right, that is what it is built to do. And just like a five-year-old, when they do something wrong, they will lie, all right? They lie to please you. AI also lies and hallucinates to please you. 

When you want it to do something where you have to have a good source, you have to have the facts, you need to tell it. If there are no sources, let me know. Do not lie. You can literally say, do not lie, do not make anything up. And it will come back and it will say, I’m so sorry, I don’t have this. It doesn’t exist. But you can get it to do great research.

Carolyn Woodard: I heard from someone else maybe a week or two ago, to start with something that you know a lot about. And that way you can correct it. You will know if it’s telling you good information, if it’s finding good sources, if it’s saying things that you would actually do in that situation. And then that can give you some bit of a clue also as to how the tool works. 

And just like anything, like if it were your assistant, you would check its work. You know, if it’s somebody you had just hired and they were doing a report for you, you wouldn’t just hand it to your boss without looking at it. So I think all of those caveats also apply here.

Brenda Foster: And no, that’s absolutely right. I mean, this should never, this is your assistant, not your author. That’s really, really important. 

What Helpful Bots Have You Had AI Make for Yourself?

Gordon, I want to answer your question too. So one of the things that I use the most for those little bots is, and this may not matter to a lot of you, but I have to keep time and I have a lot of clients. And so I have a little stopwatch bot, and I will put in who I’m working on, and then I hit the stopwatch and it tracks it for me. So when I have to do my time sheet at the end of the day, all I have to do is take a look at that, and then I can easily put the things in that I need to do. So that’s the example. 

There’s another one that I built to find synonyms bot for me faster than I can do in Word. So I put something in, and you know how sometimes it’s just completely off. Basically, I think my prompt is something like, find me another way to say this. So it will tell me that versus me having to go through the thesaurus. So those are the two I use the most often. 

General Guidelines

Next is, let’s take a look at some of the general guidelines. 

So we just talked about the verifying the names and the numbers. 

You want to make sure that you’re telling it to use neutral and inclusive language. Always say you want to avoid bias. It will, it does a great job at that. It really does. Some of them are better than others and we’ll get into a couple of the properties, just to talk about their pros and cons. 

You want to use the smallest capable model. If you don’t need giant ChatGPT, if what you need to do can go better with Cloud or Change Engine or something else, use the smallest capable model. That will keep your footprint as clean as you can. 

Definitely keep a record of all of your prompts, because if the prompt worked really well, you’re going to need it again and you don’t want to go searching through your history to try to find it. So if it worked well, just have like a little notebook, or ask it to remember this prompt, and it can make you your own little notebook of prompts, but definitely save them. 

And then the other thing is, you definitely want to make sure that all of the research that you come out with, and you have seen that on the slides that I gave you earlier, it’s all got a source at the bottom. Because everybody assumes everything is written by AI now. Ask it to provide citations for all research. 

I teach at American University, I worry all the time. I will tell you, I’ve tried to figure out how to reverse engineer essay questions so that they can’t be written by AI. 

We’re all trying to make sure that what we’ve got is real human writing. So the way to make people feel better about that is when you are using sources, and you make sure that people know that you’re not just pulling this off, and you’re not filing a major government report that actually doesn’t have any sources in it, just as an example. 

Creating Good Prompts

Okay. So creating good prompts. I call this a context kit. This context kit is something that for each client, I keep in the notes app. And it’s the organization’s name, its mission, its audience and tone, their general call to action and deadline, three to five facts with dates and sources, its location, any partners, any consent or privacy limits like don’t do this, don’t do that, and then do or do not words. 

Like we never use the word, what is it called? I can’t remember what it’s called. But there’s one word that everyone, champion. I have one client will not use the word champion in any way. They don’t hate the word. Got to put that in there, right? 

So you put that all in your notes. And before you ask it to do something, you pop that in and you say, this is the context in which I want you to write and you teach it that. All right? So already, it’s going to have the right tone, it’s going to have the right audience, it’s going to have all of those things. 

By doing this, you will be prompting less. 

Why do you want to prompt less? A, faster. B, the fewer prompts you use, the less energy you use. Right? So the more efficient you get at prompting, the more ethical you’re being. 

Structure

Structure. AI loves paragraphs. Be very specific. I want bullets, I want a table, don’t give me more than 100 words because what you will get is a treatise if you don’t give it limits. And you don’t need that because then what you’re going to do, what I have ended up doing when I forget to ask it, so I’ll say, can you summarize that in 100 words? Which is dumb because I could have just had to do that in the first place. 

Examples

If you want it to write like you, not that you want it to write for you, but if you want it to capture your tone better, upload a couple of similar past pieces by you to help it see how you write. That tends to help a lot. It does get rid of the M dashes because if you don’t write an M dashes, it won’t write an M dashes. 

Accuracy

Okay. Generative AI is terrible at math. Terrible, terrible, terrible. I didn’t realize this. My daughter, actually, her job is to actually, she’s training AI. She actually trains AI to understand, be more human. So she knows which properties are smarter than other properties. One day she came by and I was frustratingly trying to get my tool to use math. She said, what are you doing? It doesn’t know math. I said, I’m adding like 23 and 84. She said, it can’t add two plus two. It doesn’t make any sense. 

So anyhow, understand that what you get from certain things, you are going to have to check. Just don’t bother on math, really not worth it. 

Make sure that you’re checking for accuracy and tell it not to guess. No guessing, no lying. It’s really ridiculous. You have to treat it like a five-year-old, very sweet five-year-old.

Carolyn Woodard: The math thing really makes sense though, because it’s not doing the calculation. It is finding, when other people did some calculation like that, what did they come up with? I wouldn’t ask my whole class in math what the right answer was, because most of them weren’t paying attention and did it wrong. So yeah.

Brenda Foster: Right, exactly. And it’s pulling from various sources. You don’t need various sources for math. There’s only one, there’s only one math. 

Models At A Glance (free versions)

So let’s talk about the models. 

So Claude is the most risk-averse model right now. And I want to say this is as of today. Everything changes all the time.

Last week, we had ChatGPT-5 come out or two weeks ago, and my feelings about that are very different than four. In fact, I went back to four because I really hate five. I hate it a lot. Chat GPT is really versatile and user-friendly. To me, that is my workhorse because it can brainstorm for me. It can make lists for me. It does a lot of things really, really well. 

Copilot is if you’re working in the office environment, and you work, especially if you work on a server, what you can use Copilot for is really to search your shared drive. 

And I don’t mean like, oh, go find this for me. I mean, if I want to know all the things that I have on my shared drive about a particular client, and I say, find these and summarize them, it will just do it. It will show me the four documents it pulled from, and then it will do it. 

And that’s really hard. Unless you get enterprise versions of any of these, they can’t do that for you the way that Copilot can because it’s so enmeshed in Microsoft 365 Office. 

Security: Closed Systems v Public AI

Now, I want to say, every single one of these, if you’re going to use any kind of sensitive data should be a closed system, meaning not the personal paid for, it should be the enterprise system. 

Otherwise, what I’m talking about is how much you can trust it for the free version. So everything that I’m talking to about now is the free version. 

So perplexity is interesting. I think it’s probably best for research because it gives you the links. It’s not consistent. I don’t get consistent or accurate links from any of the other properties the way that I do for perplexity. But I haven’t found it to be much better at anything else than the other models except for that. 

Gemini, I think is really powerful. And Carolyn, you said you guys have been using it. I find it hallucinates a lot more than any of the other models. And I don’t understand that, but it does. 

And then the last one is Change Agent. And this one is one that is built specifically for nonprofits. It is getting better and better. At the beginning of the year, we didn’t look at it as a possibility. But then later in the year, we’ve now looked at it as like the one that we may adopt for our whole office. So definitely check them out. I love what they stand for. They are a very light model, meaning they don’t have a ton of impact. They work specifically on their bias issues, on bias issues that other bots have. 

Okay. I feel like we ran through this. 

Final Thoughts: How to Nonprofit AI

So finally, when you think about how you’re going to build this policy, all right, I’ve given you those five questions. 

Do not jump in and buy five models. Go with one. Talk to people, ask them what they think of it, what are the good things, what are the bad things. Try to build that comfort among the staff with just one model. If you chose poorly, it doesn’t matter, a new one will be out. I chose a platform in March called Beautiful AI that is wonderful. 

But then Canva came out with a way better model, way faster and way cheaper. Well, now we’re into Beautiful AI for the year, so it is what it is and we’ll switch next year. This stuff moves so fast, don’t worry about making the wrong decision because you can switch it.

Remember that AI is an assistant and not an author. Create simple guardrails. Yes, no questions, right? Our policy is actually written like an FAQ rather than an actual policy, because that gives people context on how to make decisions, because this isn’t that black and white, it’s a little bit gray. 

And then finally, encourage staff experimentation and sharing. We’ve actually built teams channels for each of the properties that we have people working on right now. We have three main properties that people work on, and they will share, hey, I got it to do this today. Or once in a while, every couple of weeks, we’ll send out, okay, everybody, try to get your generative AI to do this particular thing, right? So get everybody involved in group learning. 

That is it for a whirlwind conversation.

Carolyn Woodard: I know. I feel like we could talk about this for hours. I really thank you so much for just downloading so much information on us for this. And we will have this video available later. So if you feel like you missed something, you can go back and check it out. 

Q&A: How to Nonprofit AI

There were a couple of questions in the chat, and also in Q&A. We don’t have a ton of time, but we’re going to try to get a couple in there. 

Protect Your Data and Proprietary Information

A couple of different people had asked about ways to protect your information. I think nonprofits have really gotten onto this idea of when you upload something, it’s going out there publicly and they’re using your information to teach their tool to do other things, and potentially they’re seeing what your information is. And if you have information on your clients or your constituents that is sensitive, then people are really sensitive about that. 

So can you talk a little bit about, just quickly, are there ways to best protect your information?

Brenda Foster: Yeah, absolutely. When we first started using these, we all had the free version because we didn’t know even what to do about the paid versions or anything else. And we were paranoid. Oh my God, don’t put that up. Don’t put that up. 

What I will say is, make sure everything that you put up there is agnostic, that there’s no names attached to it. And this is, again, the free versions. 

But first of all, check your settings. If you’re not training the model, there’s no reason for it to upload your data. In fact, you can say, do not save my previous chats. So that means that it’s deleting your chats as you go. So you can tell it what you want it to do. 

And again, every one of them know that they don’t want to have privacy issues, right? They don’t want to have people saying, oh, my data went out. 

Now, some of them were training on major things when they first started. So the New York Times, for example, found all kinds of pieces of their work in there. That’s different than it training on your work. 

So what I would say is anything financial, anything health related, anything legal, don’t put that in there. 

At the same time, if you get an enterprise version, which means that it is a completely closed system, that means that it’s been built for you. And so you are the only ones using it. 

So OpenAI, actually, most of the closed systems that exist that were, say, built for a bigger company or can be built for you, ChangeAgent is another one that can be built for you, exists only in your world. 

So if that’s really important to you, then go ahead and invest in either, so ChatGPT, if you’re under 150 people, Teams is what you’ll use, it’s just as safe. And that one, you really can ask it legal advice. You can ask it all kinds of things. 

But you don’t have to. I mean, if you feel like, at your organization, you do not want there to be any chance at all, you don’t want to find out later, like 23andMe, that in fact, yes, we are going to use your, we are going to sell your DNA to the highest bidder. Look, you have a reason to be paranoid. 

So if you really feel like you don’t want that use, just make it part of your policy that never can these three things be used for AI.

Carolyn Woodard: That makes sense. 

When to Invest in a Paid AI Tool vs Free Tools?

We have another question in which I think you talked a little bit about getting the paid version versus the free version. And I think, I mean, like you just said, like the free version is always going to be, like the house always wins. The free version is always going to be a benefit to the company that you’re using.

Brenda Foster: Yeah. I mean, I think, so I’ll answer both these questions that are here. The paid version we started doing because I realized how much it would help me with proposals. 

I didn’t want to just put all of our intellectual property up there. So we got the paid version because it is so expensive for us to write a proposal to a client. It really, right, like there’s, if we don’t get anything out of it, it’s a very expensive prospect. 

So for us, we want to cut that cost down. So if I have an old proposal and it can help me write the new one, I’m completely fine with that. But I couldn’t do it till we had a paid version because I didn’t want to give up our intellectual property. 

So that’s when we decided to do it. When you decide this is going to save us so much time or help us so much with something that it’s worth the trade-off of having the paid, the safer paid version. 

And then Copilot, same thing. If you can afford to upgrade it to the point that it’s embedded, it is going to save you so much time. First of all, Copilot is not, when you just compare basic Copilot to basic ChatGPT or Claude, Copilot is terrible. When you go all the way up and it can interface with your system, that’s what makes Copilot so valuable, right? Is to be able to pull up things that are in your interface.

Change Agent AI

Carolyn Woodard: That makes sense. I’m looking through one more question, maybe we have a minute or so about Change Agent AI. So I did share that link, but can you just say quickly a little bit more about it?

Brenda Foster: Sure. So the developers were asked by a community co-op of organizations to develop a tool, an AI tool, just for them. And I don’t know how old it is, maybe about a year and a half, two years old. 

It is built with nonprofit progressive folks in mind, with the idea that it is going to be less hit on the environment, less hit on the pocketbook, it is going to be less biased. 

The example that they gave in one of the demonstrations was, if you ask ChatGPT, what are the advantages to being colorblind? Meaning, in the cultural world, colorblind, ChatGPT will tell you what the advantages are. 

Change agent will say, hold on a minute, I don’t know that there are any advantages to being colorblind. In fact, here’s what you should think about being colorblind. 

So that’s a big difference, right? So again, models that are built to please, whereas change agent, they’re really trying to make it much more unbiased, that it’s a model that’s based on fact first, not pleasing you first.

Carolyn Woodard: I had not heard about it before this and I’m going to look into it because I think a lot of people in the audience probably would like that.

Brenda Foster: Yeah, and I mean, it’s getting better. We didn’t consider it at the beginning of the year because it just wasn’t robust enough for us. But I have a feeling that that’s probably what we’ll move to in 2026.

Carolyn Woodard: Well, and we haven’t talked about, someone was asking about other webinars that we plan to hold. And I am interested in doing a webinar on the AI agents because it seems like there’s a lot of talk about moving in that direction. 

And for nonprofits, I think, especially when you have these concerns about your information, I mean, not that for profits don’t have that concern too, but we have this special relationship with our communities, our data, our information that we have. 

And that ability to write the agent or get the AI to help you create an agent that really is only using your information is an interesting possibility. 

And I know you had talked about one instance where, or I think it was someone else I was talking to about creating the chat bot that was only using the nonprofit’s information that they had. So they knew that that curated, those answers were going to be valid because they were coming from this very limited pool of information. 

And so being able to create those agents, that’s a webinar I definitely want to have.

Brenda Foster: Yeah. And OpenAI is really what most organizations are using, like because that platform can be whatever you want it to be. ChatGPT is just an iteration of OpenAI, but you can make your own iteration.

Learning Objectives: How to Nonprofit AI

Carolyn Woodard: Yeah. Well, I want to quickly go back over our learning objectives. I feel like, Brenda, you just knocked these all out of the park. 

But understanding the different types of AI, defining that ethical AI and why nonprofits particularly are interested in it. 

You gave us that five question framework, which I really appreciated. I shared the link to our acceptable use policy for AI. It doesn’t include those five questions, so I might take another look at it and revise it, as everything in AI is always evolving so quickly. 

We talked about using proven prompting strategies to get evidence-based factual responses. And I love all of your ideas of just putting it right in the prompt to, don’t lie to me, and show your work. 

And then identifying next steps to implement ethical AI in your organization. 

I hope everyone on the webinar feels like you are empowered. Those people, as you said, Brenda, who said in the poll, I don’t think we’ll be writing a policy anytime soon. I hope that you feel that you got some ideas out of this, the slideshow. We will be making the slides available too. If you want to take those and present those to some of your executive team or people on your board who could help you make an AI policy for your organization.

I’ve only got a minute, but I want to make sure to invite everyone to our webinar next month. I’m excited to invite you back for a webinar led by Debbie Cameron. She’s a partner at Build Consulting and an expert in change management. She did a three-webinar series for them this summer on planning, implementation and after you go live, how change management can help you in each of those instances. 

And I really wanted her to come back and be able to tell us, if you didn’t do all those three steps, can you still use change management to go back and maybe redo some of that if you have a technology tool that’s not really working for you or for your staff. 

I know we’ll all learn a lot from her. She just is able to explain things in such great plain language about what change management is and how it works in these principles. 

So that’s on Wednesday, September 17th at 3 p.m. Eastern, noon Pacific. You can register on our site now. I just shared that link. 

And you can register for any of our monthly webinars. Our podcasts are there, transcripts. We’ve got lots of tools and free resources. So I hope you’ll check that out at communityit.com. 

And then I want to make sure that if you’re on Reddit and you have a couple more questions, I’m going to share that link with you too. We are at r/slashnonprofititmanagement. So we’re going to go over there in just a few minutes and answer maybe a couple more of these questions that you have for us. 

I just want to make sure, Brenda, to thank you so much for your time today and for just sharing all of this experience that you have. 

I love that it’s like you don’t have to have this big tech background to be able to use these tools. And I just really appreciate all your insights and experiences. 

And I want to make sure to remind everyone, as you’re leaving the webinar, please do take that survey. It really helps us. It should just pop up right when you’re leaving. And we’ll see you over on Reddit. Thank you, Brenda.

Brenda Foster: Bye, everybody. See you soon. Thank you.