Harnessing the Power of Conversational and Generative AI, Part 2: Spotlight on Strategic Campaigning
Explore practical applications of AI like ChatGPT in union and community organizing. Discover how it aids in strategic campaigning, drafting press materials, demand letters, phone banking scripts, and more.
In the second installment of this series on the application of generative AI in union and community organizing, we’re going to dive right into practical use cases.
If you haven’t already, I strongly recommend reading the first part of this series where we discussed important security information and caveats for using AI tools like ChatGPT. You can find that article here. Before inputting any sensitive data it’s crucial to consult your organization’s appropriate-use policy for AI technologies, should they have one. Although this guide uses union-specific examples, the use cases presented here are quite general making them broadly applicable across various organizing contexts.
So that being said, let’s get right into it. Below are 5 additional ways for you to harness the power of generative and conversational AI in your organizing practice. This time, however, we will focus on strategic campaigning.
To skip to one of the 5 methods now, I have curated a list for you.
1. Building a campaign and escalation plan
For this first use case, let’s assume that I need to build a strategic escalation campaign plan. I might have little experience with doing so, or, perhaps, I have run several campaigns but am experiencing a sort of “writer’s block.” In either case, ChatGPT can help. I start with a very simple prompt: “Build a strategic escalation plan for a union contract campaign.”
This is a great start. However, let’s say that I am going to be building this campaign with a group of workers who are new to union campaigning. So, I want to create a visual that can show them what “escalation” means. So, I asked ChatGPT-4 to visualize the above steps in a graph.
This is great! However, I want to show a chart in a slightly different format that explains the way that escalation plans move from low- to high-risk actions. Notice how I also include preferences for colors and direct it to “make it beautiful.” (you, of course, can be the judge of how successful it was on that last part)
This is a lovely start. However, notice that the text does not fit into the purple boxes. I now prompt ChatGPT to correct the error, which it can do by editing the code directly. It might take a few tries but we can guide our new AI assistant along the way.
Now we have a more presentable chart. If I wanted to, I could further prompt ChatGPT to change other aspects of the chart, including words, colors, and even the type of chart we are using. It’s important to remember that although generative AI is powerful, it isn’t perfect. With a little problem-solving, we can work together with it to correct errors in its code and produce our desired output. And, by learning its limitations, we are able to better write our original prompts to avoid them.
Tip: If you don’t want ChatGPT to explain what it’s doing, just include the following after your prompt: “Don’t talk, just execute.”
2. Drafting press materials for actions and campaign milestones
For our second use case, assume that I need to create press materials for a union boycott. All I need to do is feed it the correct information with the prompt. The more information I give it, the more detailed and relevant the response. Note that I included the purpose, name of the union, name of the company, reason for the action, and other relevant information. I also included the time, date, and location.
Interestingly, ChatGPT knew to include a quote from the impacted worker. Although it created one for Maria, our imaginary union leader in this scenario, the union could replace it with an actual quote from their leader. And, if you already have a quote, you can include it in the initial prompt.
3. Drafting demand letters on behalf of workers
Now we get to my personal favorite: demand letters. While such letters should always be vetted by counsel or a designated “approver,” ChatGPT can be used to draft templates and to provide a starting point. Notice how I include specific information about the violation and then instruct it to “cite all relevant sources and references.”
And the result? A well-written, detailed demand letter complete with references to federal labor law and fill-in-the-blank prompts for the organizer to fill out with the appropriate information. As a bonus, it even included suggested attachments, such as proof of official recognition and relevant sections of the NLRA.
4. Drafting phone banking scripts
ChatGPT can also create phone banking scripts. In this case, I ask it to draft a script that our organizers can use to call the entire membership in order to mobilize them to support the janitor’s strike. Note that I want this script to include the details we already have, so I copy and paste details of this hypothetical strike, such as start time and location, directly into the prompt.
For a language model, the resulting script is pretty impressive. It includes an introduction, a brief discussion of the issue at hand, a call to action, and a closing. It even included a clear question asking if they can attend an event or not.
5. Drafting one-on-one templates
Here I will use a more general prompt and list a simple purpose. Because one-on-ones require more sophistication, with back and forth in conversation, I prefer to have it generate a more general template that I can edit a little more heavily.
Among the many impressive parts of this template is the way it knew to begin with building rapport and identifying issues and to end with “calling the question” for a commitment and providing the next steps. Clearly, ChatGPT was trained on some organizer training materials!
If your union or organization prefers to use its own template, then you can simply copy and paste it directly into the text box and prompt it with something like: “Use the following template as a guide to draft a one-on-one conversation. The purpose is [PURPOSE] and the aim is [AIM].” Note that you can make the prompt as detailed as you like.
Sample workflow and best practices
Remember, generative and conversational AI like ChatGPT can be a powerful tool, but it’s not perfect. Learning its limitations can help you better write your original prompts to avoid errors and produce your desired output. It’s always a good practice to verify sources, references, or case law that it cites.
So let’s put review and verification into context and see how we might ensure accuracy before publishing. In our demand letter example, we can imagine a sample workflow as follows.
- Write a prompt into ChatGPT-4 instructing it to generate a demand letter from the union to the employer. Include relevant details in the prompt.
- Review the output for any errors or undesired elements. Instruct ChatGPT to correct those errors, change the tone, or make another change with new prompts. Repeat until the output is satisfactory.
- Move the output from ChatGPT into a Google document. Fill in any missing information and independently verify any citations, references, or case law it generated. Send to legal, or the designated staff member, for approval before sending to the employer.
While ChatGPT can seem rather omniscient to the new user, it’s important not to overstate its abilities. It does not replace the need for you, the human expert, to verify its output and add necessary content before sending or publishing. The reader should be careful to avoid misconstruing AI’s abilities by thinking that such abilities signify a deep, conceptual understanding of the topic. While AI can assist in developing templates and visuals and citing relevant laws, it does not have the conceptual understanding as would an organizer or campaigner with years of hard-earned experience. Therefore, the reader must remember that generative AI does not replace human creativity or reasoning; it merely augments it and helps in providing a starting point.
If you want to try these tools out yourself, check out ChatGPT by clicking here. Note that depending on what you want it to do, you might need to use ChatGPT-4 with code interpreter, which does charge $20 a month for a subscription. However, the majority of these tasks can be done in ChatGPT-3.5, the free version, and in cases where it cannot produce the desired output, such as charts, it can guide you on how to do it yourself.
