How AI shaped the 2024 election: From ad strategy to voter sentiment analysis
Generative AI hasn’t had an outsized impact on 2024 presidential campaigns, but it still played a pivotal role in many aspects of political campaigns.
With Election Day finally here, it’s worth looking at some of the ways ad-tech firms, political startups and political agencies have used large language models and machine learning. While AI has posed new dangers for election misinformation — something companies and governments have been all sought to detect and prevent — LLMs and ML were put to use helping political campaigns with content creation, audience analysis, voter targeting, and ad-buying.
Generative AI arrives for its first political season
Predictive models have been used for years, but 2024 is the first U.S. election in which generative AI has gained traction. Smaller campaigns have been using startups like Battleground AI, which helps progressive candidates in down-ballot races use AI to create and scale text-based ads for search, social, YouTube, and programmatic ads. Other new features help streamline ad creation and approval with editable ad mockups and an approval link feature for collaborations. A new content template store also offers designs from political advertising agencies like Adapt Digital, Studio Mosaic and Uplift Campaigns.
Candidates are using the platform to scale organic and paid media about topics they care about, according to Battleground AI co-founder Maya Hutchinson. In an interview with Digiday last month, she said total clients have quadrupled since the platform expanded beyond beta this summer. Although Meta has been used the most for testing and reaching voters, she noted the importance of diversification while also caring about the quality of conversations – not just quantity.
“[Candidates] want to keep iterating and finding better ways to talk to people and make a lot of variations,” Hutchinson said. “How can we start to test those different policy issues that we know are important to our voters? Let’s allow more time to create messaging around multiple issues that resonate with our voters, not just hammering one thing all the time [that] people can easily drown out.”
One Battleground client is Pixels and Persuasion, which tested the platform for a school board race to see whether AI or humans could write better copy. After ingesting campaign materials, they found the AI-generated copy that was then edited by humans performed the best and even led to the AI drafting unexpected phrases.
“There’s one that was talking about human flourishing, which isn’t necessarily language that we would use in a political ad necessarily,” said Pixels and Persuasion CEO Myles Bugbee. “I’ve been in the [political] space for a while and I haven’t seen the word ‘flourish’ used, but that was a word we used in a headline that’s performing well so far…It raises concepts and things that we may not necessarily have thought of.”
Another Battleground AI user is the political firm Blue Dot Consulting, which used it to help first-time candidate Kiana Fields campaign for Kentucky’s state senate. After first running a summer poll to see what voters cared about, Blue Dot analyzed that info along with voter data and Meta user data to create a $20,000 social ad campaign while using AI to create more content faster and cheaper.
“We see a lot of folks that are well heeled come in and self-fund or are from the donor class,” said Blue Dot CEO Taylor Coots. “It’s easier for them to go out and find digital experts and pay whatever they’ve got to pay in terms of a retainer or minimum buy. But there are these things that prohibit some folks from getting into race. We see BattlegroundAI and tools like it as a way to get into the space.”
How LLMs and ML helped analyze political ads
Arguably the oldest form of AI, machine learning has been used in politics for years, but it was used more prominently in campaigns that wanted to analyze ads and voter sentiment. One firm, XR Extreme Reach used LLMs and ML to analyze and forecast political ad spend and messaging trends. XR’s interactive dashboard tracks election dynamics weekly, highlighting top topics by party and state, candidate share of voice, overall ad sentiment, and party-specific messaging. For instance, after Vice President Kamala Harris unveiled her economic plan, XR observed a Republican shift toward economic messaging.
AI tools were applied to assess voter sentiment and engagement, which then helped campaigns refine negative messages. To educate its AI platform, XR trained multiple AI models on general key interest categories and then scored LLMs using typography from PEW Research. It then used an API that runs data analysis every night and analyzed both words and images that appeared in ads.
According to XR, in the last month about 45% of political ads used negative messaging, which focused on critiquing opponents instead of promoting a politician’s own platform. XR noted that 35% of political ads in swing states focused on the economy, 25% addressed healthcare, and 15% talked about national security.
New political tools for niche undecided audiences
Campaigns are using predictive AI to measure real-time voter sentiment and gauge viewers’ receptivity to political ads across CTV channels, said Robin Porter, head of political ad at LoopMe. With so many unprecedented events this presidential cycle, she said historical voter survey data wasn’t always helpful in understanding where voters were leaning at various points in the race. After President Joe Biden dropped out of the race, LoopMe found 20% of Biden supporters shifted toward Trump or became undecided, then helped campaigns target niche audience segments of undecided voters.
LoopMe has also been in talks with candidates who are already looking to the next election that want to use AI to gauge state voters and see prospects for higher office — such as from the U.S. House to the Senate.
“[Candidates want to] get into the hearts and minds of their state voters to understand who they’ll be looking for,” Porter said. “What type of leader would resonate with them most? It could not only inform if they should or shouldn’t run, but how they should potentially shape that leadership style and messaging in order to kind of guarantee a win.”
Other ad-tech firms are using ML to analyze the context of content on platforms for political and non-political advertisers to either run ads within or to avoid. One hurdle on YouTube for political advertisers is they’re not allowed to use voter registration files — the tried and-true way to run political ads. Instead, some are relying more on contextual data to reach certain audiences or avoid others.
Pixability has gotten into the business of solving this, helping candidates plan campaigns, optimize creative and analyze competitors’ ad strategies on YouTube and CTV. The firm’s also working with advocacy groups to know when to run ads to reach audiences at certain times.
“People are talking very passionately on one side or the other, so it’s important to understand the context of things, which is very difficult to do without AI,” said Jackie Swansburg Paulino, Pixability’s chief product officer. “Obviously, you can’t just use keyword-matching for those kinds of [topics] that fall on both sides [of the aisle].”
More in Media Buying
Influencer spending increases in Q4 after election blackout as brands integrate channels, commerce
While brands paused or reduced their influencer investments during the presidential election, they also face the ongoing impact of inflation and the need to balance their media budgets going into next year.
AI Briefing: How Google and DOJ’s latest antitrust arguments address AI
Lawyers for the Department of Justice and Google filed lengthy written arguments in the ad-tech antitrust trial ahead of closing arguments later this month in which they each highlighted AI.
Media Buying Briefing: How generative AI is being applied in big and small settings
As generative AI begins to settle into everyday workings across the media and marketing landscape — we’re definitely moving past the shiny new object phase — it’s worth looking at some micro- and macro examples of its application.