Elon Musk bought Twitter and used it to help elect Trump. He also controls a popular chatbot, which he might use to exert political influence when he needs to.

Musk’s purchase of Twitter and the following tumble of the platform’s market value became a subject of ridicule - until the 2024 US presidential election. X was an important piece of the puzzle in Musk’s effort to help elect Trump. In short, it became a private propaganda tool to promote views aligned with the owner’s view.

Here’s an article on how Twitter became Elon Musks propaganda machine, and here’s a paper on potential algorithmic bias on X during the 2024 US election. Even Washington Post published a piece about how on Elon Musks X, Republicans go viral as Democrats disappear, unfortunately it’s paywalled, but see the author’s tweet linked below the image for one more chart.

Source: Drew Harwell, Washington Post

Anyway, who cares about market value when you get an office in the White House and so much influence that some call you “president Musk”.

Source: The Insider

Grok

Now X has its own chatbot, Grok. It was mostly a meme model until Grok 3 unexpectedly topped the LM Arena leaderboard, the leading chatbot benchmark, around February 2025. Unlike the other commercial chatbots, there is no API - the model seems to be available only through X.

Grok has gained popularity among X users opposing Musk because of the chatbot’s apparent “wokeness”. It has no qualms about criticizing Musk or Trump. For example, here’s a paragraph from Grok about Musk’s role in the election:

He formed and heavily funded the America PAC, donating nearly $300 million of his own money to bolster Trump’s bid for re-election. Musk’s acquisition of X (formerly Twitter) in 2022 further amplified his influence, allowing him to shape public discourse and expose what he and Trump framed as government inefficiencies and media distortions.

However, this apparent honesty is a two-edged sword. It’s relatively easy to align a language model the way the developers want. You can prompt it, you can finetune it, you can adjust the probability of output tokens, or even edit the knowledge it stores. What Grok says today might not be what it says tomorrow.

Here’s a dangerous scenario with a high probability of becoming reality. People on X learn to trust Grok. When Musk needs to steer the public opinion again, for example before the next election, he instructs the developers to modify the model so that it says what Musk wants it to say, more or less subtly. People conditioned to trust Grok’s objectivity gobble it up like young pelicans.

In 2016 it was Cambridge Analytica and Facebook. In 2024, it was X. The next time, it might be X again, this time with Grok doing the influencing too.

P.S. Try an experiment. Like this famous post by Pekka Kallioniemi about Musk being pro-Russian. Come back the next day and check if your like is still there.