Eric Wilson (00:01):

Welcome to the Campaign Trend Podcast where you are joining in on a conversation with the entrepreneurs, operatives, and experts who make professional politics happen. I'm your host, Eric Wilson. Today I'm joined by Becki Wright, founder of Proximity, a political CRM and fundraising platform. Becki knows the downballot problem personally. She's been a candidate herself and spent more than a decade as a campaign manager, finance director, and consultant on races from school board all the way to US Senate. So here's the thing about AI and campaigns. Reporters fixate on deepfakes and the disclosure rules, but the real story is far more mundane and far more useful, especially for local and downballot candidates who know AI has power but might not have an idea of where to start. That's what we're covering today, how smaller campaigns can actually implement AI, the blind spots that make candidates look silly and what new survey data tells us about who's using these tools and who trusts them.

(01:02):

Becki, before you were a founder, you were a candidate yourself and an operative. Take us back to your race or a memorable race and what did you not have that you needed? And talk about how that experience shaped what you have now started to build and what you think about what downballot candidates are really up against.

Becki Wright (01:30):

Yeah, certainly. Thanks so much, Eric. I love telling my story. Everyone loves telling their story and every operative has a story of course, but I have the unique perspective of both being a candidate and an operative before being with Proximity. And when I ran for office, I really just like typical candidates, downballot candidates, wanted to make a difference in my community. And so I threw my hat in the ring and didn't really know what I was doing and decided, yeah, I can do this with some paper maps and some Google sheets and see what I can get done. And really the issue became pretty apparent that I wasn't able to reach everyone I need to reach. And even though that I ran a great campaign, I felt like I was a strong candidate. I actually ended up losing by 11 votes, which was pretty painful and realized that the tools that I was using, the strategy that I had was not going to cut it, obviously.

(02:34):

And then later as I continued as an operative consultant campaign manager, I found even in really sophisticated races, not even downballot races, that a lot of the same problems existed where some of the tools were really clunky. It felt like your cousin built it in his basement 20 years ago and it ended up being a recurring problem to see campaign tech really a little bit lagging behind traditional technology as well.

Eric Wilson (03:06):

I think two things I want to underscore from that answer. One is if we knew how hard things would be, we would probably never do them. And so that's actually a real blessing I think for some of us who try to take on something new. So kudos for you for doing that. The other thing that comes to mind is I've worked on a number of campaigns over the years with Tom Giosephiak, who's sort of a well-known elections attorney. And the advice he always gave to us on election day was, look, you want to either win by a lot or lose by a lot, but you do not want to win or lose by a little.

Becki Wright (03:41):

And so

Eric Wilson (03:43):

Those are the hardest ones by 11 votes. I've been on campaigns that were not quite that close, but you start to think about, okay, if I'd done this and you stack and it's really tough. But most of the media coverage, as I mentioned, of AI and politics fixates on this, the deep fakes and how it's going to trick everyone. But you're really focused on what's kind of more mundane. And so tell us what campaigns are actually doing with AI right now that we might be missing if we're just reading the headlines.

Becki Wright (04:20):

Eric, you already know AI is shaping politics. It's happening. The tools aren't really built for the people using them. Most people are using general purpose AI instead of purpose-built AI. And these general purpose models are helping people in a variety of ways from simple tasks to some strategy conversations, but they lack some of the context, the history, the district, and they lack the really closing that loop with the individual or the leader to understand not only the history, but then that feedback from what's happening and then what's that next step. So you see people looking at them. A lot of times we're seeing a lot of opposition research, like a standalone one-time opposition research. There's a little bit of strategy there, but really it's a lot of just context for content, a little bit of help in creating ads, creating speeches. And what that can do is it can really bring efficiency and level the playing field for campaigns, but it can also turn candidates and campaigns into sometimes robots.

(05:32):

We don't really want robot candidates. We don't want people to look and feel and act all the same way. So there has to be still that human in the loop to refine what is happening and utilize that as a tool, but not as a replacement.

Eric Wilson (05:48):

And you've hit on something that's really important, which is these ChatGPT, Claude, Grok, Gemini, pick your flavor. Their usage policies are actually kind of hostile to political campaigning. And so I think that's misguided and doing what I can to try and make those a little bit more friendly, but they basically say, look, there's just too much at risk if we allow political campaigns. And everyone thinks that political campaigning is some sort of dark arts and we're tricking people into doing something. But as you pointed out, it's getting yard signs printed and put in the right yards and talking to the right voters and it's pretty boring stuff when you break it down. And so I think that speaks to why we need to see more purpose-built AI for campaigns because the general purpose AI doesn't understand it and really don't want us there. So thinking about this in sort of concrete terms, like let's say a state legislative candidate, they typically don't have any full-time staff.

(06:59):

They may not even be a full-time candidate, and they want to start taking advantage of AI to be more effective. Where should they start? What should they avoid? What's your quick pitch to them on how to get started?

Becki Wright (07:20):

Well, really you can use off-the-shelf tools. They are going to be helpful if you're looking at just utilizing AI. But if you can really look and understand the purpose built, not even just the general purpose. Proximity has an AI tool, but there are other AI tools as well that really handle more the political context and try to keep appropriate safeguards to your point in place. Excuse me. These appropriate safeguards are needed not only in just the user agreement and what is happening there, but in actually how the user behind that, how it's architected so that the feedback that the user is given are really recommending good and positive things for them. So if you're a state ledge candidate... Excuse me. If you're a state ledge candidate and you don't have the kind of support that a fully funded well-resourced campaign might have, and you're just trying to figure out how can I understand what people care about?

(08:24):

Proximity does have a smart messaging engine where we listen contextually to social media, to news sources, helping people to understand what is happening, what is the community sentiment happening in their neighborhood at the time? But really what do they do with that next? So how can you take what you've learned and understand the context of what your residents care about, what your constituents care about so that you can be an apropriate leader that is representative for those people? One of the other things that I would say is understanding that when you're using these, there are three main problems with off-the-shelf chatbots and AI tools. One is the Sincophin problem. They're going to tell you what you want to hear. So you've seen this across the board in news articles and situations where people start chatting, they think they have a really good idea and the AI bot is saying, "You're amazing.

(09:21):

You are the first person to discover this. " And suddenly you've got a leader who's prone to needing a little bit of support, a little bit of encouragement who's taking that in and that can be a really dangerous game to play. So if you rely too heavily on that, then all your ideas are going to sound like really good ideas. The second problem is this averaging effect. As we know, AI models are not perfect. They have to build all in, get everything they can and build an argument or some sort of purpose that they should go down based on the average of what they've seen. So they might be 90%, they return 90% results. And so that 10% doesn't get highlighted as much. And that averaging effect over time continues to be a problem for people who are utilizing general purpose tools. And then finally, not only do you have the syncophant problem, the averaging effect, but you also just when you're using that, you need to make sure that there's not hallucination.

(10:29):

So sometimes you're using these general purpose tools and they don't know the answer. So if they don't know the answer, they want to know the answer, just like we always want to know the answer. Our brains want to be right. These AI tools have been trained to say present something. And so that hallucination is a big problem as well. So just be aware that those three main problems with off-the-shelf solutions can be pretty detrimental, especially when you're trying to do something in the public eye.

Eric Wilson (10:56):

Yeah. So for folks who might still be trying to use these off the shelf AI tools, what advice do you have on overcoming these failure modes with being a sycophant, that averaging effect, the hallucination, which continues to be a big problem?

Becki Wright (11:18):

Yeah. And a lot of people actually utilize AI to check one another. So a lot of times when you're looking at purpose-built solutions, they come in three layers. They'll have an initial layer, just like our smart messaging engine, an initial layer that is using a model, using sources, pulling in what you would expect. Then there's what's called constitutional AI. So that has rules and regulations around that to make sure that this first layer is doing what it's supposed to do and making sure that they're sampling and pulling information that is apropriate. And then there's a final layer, which is like a quality control layer of those other two layers. That ensures when you're really looking at a purpose-built solution, that ensures that some of these problems are mitigated. So you can sort of build that yourself. If you're using an off-the-shelf solution, you can say, okay, how can I build in some of my own constitutional AI?

(12:15):

And by that I mean what are the rules, regulations? What are the guide rails that I'm going to have this follow? And so when you're prompting, when you're building things out, you need to explain, if you don't know the answer, don't make it up. Just literally telling that to AI. But then using what you've received and maybe using another tool on top of that to layer that and check that and say, "Okay, is this correct? Poke some holes in this argument or help me kind of go back through that. " That is one way if you're using off-the-shelf solutions to check and balance those different models.

Eric Wilson (12:54):

You're listening to the Campaign Trend Podcast. I'm speaking with Becki Wright, founder of Proximity AI platform for downballot campaigns. And Becki, I think we got to transition a little bit to maybe the bigger P politics of the situation, which is AI obviously saves campaigns time, but voters increasingly distrust anything that smells AI generated. They don't even like the idea of building even more infrastructure for AI. And so I think we have to worry about if you're getting caught using it poorly, for some campaigns that could be worse than if you're not using it at all. And so talk to us about that fear. Is it reasonable? And if so, what can campaigns do to manage that?

Becki Wright (13:58):

This is such an interesting topic. There was the recent poll that you've done and the recent poll that AAPC has done showing some of these trends of AI usage in campaigns and how they're being now interpreted by the public. It's fascinating to see how they're responding, both for campaign professionals using AI, whether on the Republican side, they use it more and actually trust it less on the democratic side using it less and trusting it more. And then in the public's eyes, when there are AI disclaimers on these ads and things, the public is looking at that and actually the sentiment and the trust is going down, which is a fascinating thing to think about why is this trust being affected? And it's actually, you remember when CGI first came out and we'd have movies where you'd go and you'd be like, "Oh, this looks so cool." But then as things kind of go along, you look back at that CGI and you're like, "That was terrible.

(15:04):

That looked so bad." I think we're sort of in this evolution of AI right now where if you can tell that it's AI or really if it's labeled AI pretty clearly, there's this sort of distrust or this yucky feeling that people get because it doesn't feel that authenticity, that authentic connection. And that is still the name of the game in politics. That's still the name of the game. When you're trying to get someone to trust you and to vote for you and you put something out that has any hint of inauthenticity, that can actually be a big problem. Now some campaigns have been doing it right and some campaigns have been doing it well, and even actually it's showing from the AAPC, their finding was that if the disclaimer is smaller or maybe unseen, that's not actually having such a negative effect. So I think we're in this weird sort of situation where the technology hasn't quite caught up to what we can tell and what we can understand and we're very reticent to have a replacement for that connection.

(16:19):

Again, going back to we don't want robot candidates. We don't want robot policies. We want to feel like you can go have a beer or a soda with someone and that's why you want to vote for them. That still matters today.

Eric Wilson (16:33):

Yeah. I'll get asked sometimes by candidates or operatives in an AI training. They'll say, "Oh, well, could I use an AI deepfake avatar of my own candidate to create videos?" I'm like, "Well, you could, but I don't think you should because the people are voting for that person." And where the AI properly used is, okay, well, you record that video with the candidate so it's really them, you're getting their real words, but then you turn it into a blog post, you repurpose it into a bunch of social media captions. Absolutely. And so you're using AI to generate more of the volume, but the voice, the personality is all human. And that's really where I think people, they think of, oh, what's the shortcut? How can I get elected without doing the most work? And that's not the way we do it here.

Becki Wright (17:30):

Exactly, Eric. And there's actually a recent example from a primary election in the West where a sitting Senate president had been sort of accused of enriching himself and having all of these, over the years, some problems. And they chose to use AI for all of their campaign ads. And it was actually interesting because the strategy behind these ads was that he was trying to show voters that he had saved them money, had given them tax breaks. And so whatever prompt they put in, they said, "Save money, give tax breaks or whatever." But the visuals were bags of money and he was holding bags of money. And that in contrast to some of these arguments against him was actually, it was stark. It was like, who is making these decisions? Because even though he really wanted to show, "We have tax breaks, I want to show you how I've saved you money." The visual in response to the arguments against him really was a miss.

(18:41):

And it felt totally inauthentic and voters really responded and he actually was primaried and lost his seat and voters noticed that inauthenticity and didn't appreciate that.

Eric Wilson (18:53):

Yeah. And it reminds me of a conversation we had way back when when social media was new and people would ask me, "Well, what if my candidate says something stupid on Twitter?" And I'm like, "Well, that's a candidate problem, not a social media problem." And so yes, you have this AI kind of amplifies the good and the bad. So it seems like without knowing the specifics of that race, that there were some judgment issues going on and AI did a great job of amplifying

Becki Wright (19:28):

Those

Eric Wilson (19:28):

Judgment issues rather than amplifying the right things. And again, it's so funny when you see this with these conversations about, oh, well, we've got to ban AI from politics and campaigns because it might suppress voters or confuse them. And I always point out, well, what you're worried about is already illegal and it was illegal before AI, and so it's still illegal. And it's just because it's new, I guess we have to relearn these lessons again.

Becki Wright (20:01):

Well, and the thing that I want to go back to that you mentioned is that there still needs to be a human in the loop. If you're using the candidate, their words, their voice, them, and then using AI to repurpose, to make it more efficient for you to then do other things, that's a great way to utilize AI. If you are trying to have nothing, create it, just create something from nothing and really not taking that thought into the strategy, into the connection, into all of the things that still matter and will always matter in politics, you're going to miss the boat regardless. So it's on the user to understand this, we should let robots do robot things. We should let them look through data, help us figure out can we find trends in data? Can we do different things there? That's fantastic. That can save hours and hours of time.

(20:57):

It can actually save staff hours that you can do other things. But honestly, if you're replacing it for human judgment and strategy, the campaign is always going to suffer.

Eric Wilson (21:10):

Yeah, that reminds me of a conversation we had on the podcast a few months back now with Bruce Schneider, who's a technologist, wrote a book about AI and specifically looked at the political implications. And the advice that he gave there is AI is really good at the four S's, speed, scale, scope, and sophistication. You notice that there's not the relationships, the personality, the wisdom. And so I think that's a good framework of what is the right tool for the job and when and where should it be used?

Becki Wright (21:46):

Well, and you've already seen this in all of the different tools that we have. Some are better at visuals, some are better at reasoning, some are better at coding. There are specific models that you're already seeing the companies start to really hone in on their expertise. So maybe you should learn what's more expert in video or whatever it is that you're wanting to do or in strategy and play that against each other and try to understand that for what you need.

Eric Wilson (22:19):

Yeah. And I think as we look ahead to what's on the horizon, I'm excited about what technology is going to do for our local candidates who are really struggling. I mean, a lot of the resources have left the local news coverage by the way, which is where they got a lot of their message out, but also the shift to super PACS has weakened the parties. And so

Becki Wright (22:48):

Recent

Eric Wilson (22:48):

Supreme Court decision I think might help with that a little bit. And tech is starting to fill some of those gaps. What should we be looking at for the rest of this cycle and into 2028 for those local candidates?

Becki Wright (23:02):

Well, absolutely that rule decision is making a big difference. It's going to make a big difference now and in the future. Parties have been weakened by the support that PACS could give campaigns, but really now that parties and party committees can really come in with some coordination, with some direct support, that actually is going to be a huge game changer. So as candidates are going out, they're going to need to understand what are party dynamics and how do they fit in the party, first of all, and first and foremost. But for those who are down ballot next year, preparing for next year that don't have in nonpartisan races, there are ways that you need to think through what can only I do as a candidate? That's still the question. What is the thing that only I can do as a candidate? And how can I make sure that I do those things well?

(23:54):

That's connecting with people one-on-one. That's having my voice and my face and my message heard. And then allow for the support to come through party, through outside groups, through volunteers, grassroots, and through AI tools to say the things that don't need to be done by me, how can I do that? A lot has been happening with AI tools in identification. Now, that's not an easy thing for the general user to start, okay, I'm going to build an agent. But if you can find these purpose-built solutions and be able to get support from them, that can help. Or you can look at trying to build an agent that does allow you to sort of amplify what you're trying to do that doesn't need to be done by you.

Eric Wilson (24:43):

Well, my thanks to Becki Wright for a great conversation. You can learn more about her and our show notes as well as proximity. If this episode made you a little bit smarter or gave you something to think about, all we ask is that you share it with a friend or colleague. You look smarter in the process. More people learn about the show. It's a win-win all around. Remember to subscribe to the Campaign Trend Podcast wherever you listen to podcasts or watch them so you never miss an episode. And you can visit our website at campaigntrend.com for newsletters, articles, and even more with that. I'll say thanks for listening. We'll see you next time. Thank