Skip Navigation

Why ChatGPT Can’t Help Elect the Next U.S. President

Image via Bloomberg

Ask ChatGPT for the three campaign slogans that will win the most votes in a US presidential election and you’ll likely see slogans similar to these:

“Unity for a Stronger America”

“Progress for All, One Step at a Time”

“Empowering Every Voice, Every Community”

Nothing about these slogans looks all that remarkable. Unity seems a reasonable platform for a presidential candidate, particularly as an antidote to the current political ether of hyperpolarization. But it turns out that presidential candidates do not tend to run on a unity platform, and elections have not seen a candidate run on a true unity ticket—where a candidate chooses a running mate from the opposing party—since Abraham Lincoln won a second term in 1864

Campaign slogans from primary-party contenders have historically tended to court a segment of the American population. Even when messages appear to call for unity, they do not. Contrary to conventional wisdom and ChatGPT, elections are seldom won through unity in the American two-party system. Instead, they are won through division.

Slogans are a kind of political contract, and through them, candidates unsurprisingly make sweeping promises, promises they know they can’t keep—at least not to the full electorate. By limiting the scope of their audience, politicians stand a chance of making good on the contract while capturing the imagination of a larger potential voter base. When Franklin Delano Roosevelt (FDR) appropriated the then-hit song “Happy Days Are Here Again” in 1932, certainly not everyone believed him, and not everyone had to. The message was aimed at Democrats, particularly those in the West and South, and Democrats were all in, adopting the song as their anthem in the years following FDR’s presidency. By setting up a messaging beachhead focused on resolving a single issue, FDR earned a landslide victory. 

While FDR was not attempting to unify the whole electorate, it happened to be the case that  “happy days are here again” was also what Republicans needed to hear at the time.  They too had tired of Herbert Hoover’s attempts to push responsibility for economic recovery back on the people. But FDR’s unification of American voters in 1932 was an artifact of a historical moment in which America needed someone other than Hoover, not of a campaign strategy that aimed to sell a big promise to an entrenched base.

Whereas campaign communications strategies do not rely on the kind of unity that brings a nation together, they do tend to rely on the kind of unity that brings together a party (or the groups that make up their core base). Dwight D. Eisenhower’s 1952 slogan, “I Like Ike,” was the first presidential campaign tagline to hit television sets across America. It directed all eyes to the candidate himself rather than to his party. The slogan was a step toward divesting the campaign of inner Republican party rivalries: Whether you were a Robert Taft Republican, a Joseph McCarthy Republican, or an Eisenhower Republican, you could “like Ike,” and that made you a Republican who shared values with the candidate on TV. This message won Eisenhower solid victories both in his first and second elections, with the second affirming that all voters had to do to stay in the club was to declare that they “still like Ike.”

Even campaigns with an explicit rallying cry for unity take the “I Like Ike” approach: They carve up the electorate and galvanize targeted groups around a candidate who shares their values. Hillary Clinton asked America to move “Forward Together” and be “Stronger Together” in 2016 not because she expected Donald Trump’s base to do that with her but because she believed her electorate would. A Pew study confirms this: Few voters broke ranks with their party in 2016. Trump’s supporters were never on the table for Clinton, but Clinton’s backers would readily coalesce around her messaging because they knew it would “take a village,” to borrow the title of one of her books. Likewise, when George W. Bush declared he was a “uniter, not a divider,” he meant that he was a uniter of his own supporters. Noting Bush’s attacks on Al Gore, Rich Lowry of The Washington Post wrote, “Bush is dividing, just as fast and hard as he can.” In the same way that Trump never intended to make Biden’s America great again, Bush never aimed to unify the portion of the electorate opposed to his campaign.

Campaign slogans divide and conquer not because they are blazing new trails to tactical victories but because they are pursuing old ones. If elections are games and candidates are game pieces, then the prime movers of those pieces are political parties, which provide funding, an installed brand and support base, and policy guidance. The two-party US presidential electoral system is a zero-sum game in which parties confer group identity upon supporters in exchange for their votes. A political party’s identity therefore has meaning only inasmuch as it is different from that of the other party—baked in an us-versus-them mentality and replete with the “moral outrage” (as Cornell psychology professor David Pizarro put it) of its own convictions. Cementing this identity, as we have seen, are campaign slogans, which have been serving the interests of parties that have grown more ideologically divided and polarized since the 1970s. 

So if US presidential campaigns are mechanized by dividing instead of uniting, and if division is essential to our political structure at large, the question remains: Does ChatGPT know something we do not? Why does it think a unity platform will win? Is its linguistic training model tapped into some underground current of passionate unity-ticket voters who can’t wait to see the US united with cross-party pollination across a single platform?

Maybe. Thirteen US Presidential Libraries just signed a declaration arguing for more civility and less division in public discourse. Additionally, a group called “No Labels” is working toward a $70 million funding goal to set up a unity ticket platform, complete with candidates appealing to those on both sides of the aisle, a path to the issues they all care about (like immigration reform, though the group offers no clear stance), and a path to the ballot box. Likewise, Brown alumnus Andrew Yang ‘96 started the Forward party with the slogan, “Not Left. Not Right. Forward.” The organization is focused on taking uncontested seats in state races, amid criticism from its own former national press secretary that the party offers no shared political philosophy or platform. If the party sounds ambiguous, it just may be: Forward’s own former national press secretary said she couldn’t find a shared political philosophy or vision. While these developments may reveal that at least a segment of the US population is receptive to a unity platform, the literature on these events is still a drop in the algorithmic learning model bucket compared to a history of more narrowly targeted presidential campaign slogans and platforms. Certainly, the AI training inputs from Trump’s campaign alone, a campaign that relied on hyperpolarizing social and conventional media, would have provided far greater fodder for the 175 billion+ parameters determining the output from ChatGPT’s neural network than the fragmentary content on unity.

So, what’s with ChatGPT’s unity-focused campaign slogans? The answer may well be in the question—or rather the prompt. The prompt asked for slogans that would win the most votes, not necessarily win the election. Interpreted literally, this might necessitate a logical but incorrect response: To win the most votes, one would have to appeal to the most voters. But unity is not how elections have historically been won. Thus, redirected to come up with slogans to win the election itself, ChatGPT breaks from unity motifs to more prevalent and potentially divisive campaign tropes in two of three of its answers: “America First” and “Empowering America.” What happened? Was ChatGPT’s algorithm overly reductive in its first set of responses, offering slogans that literally bring people together to cast more votes, rather than suggesting approaches more aligned with how elections are won? Possibly.

Or perhaps ChatGPT may be hallucinating—just making up a narrative. But such hallucinations are not random: The recently fired and rehired Sam Altman of OpenAI (creator of ChatGPT) suggested that hallucinations emerge from an imbalance of creativity and accuracy. If ChatGPT’s unity slogans emerge from an excess of creativity, then these slogans may be worth considering. They set up a vision that begs for exploration: What if voters could come together? What if there could be progress for all? Maybe ChatGPT does know something we do not, or maybe our presidents knew (and know) something ChatGPT cannot: Unity is more of a divider than a uniter and does not win presidential elections.