President Donald Trump’s social media platform Truth Social is a safe space for him and his movement, an autocrat-friendly network where the president’s posts — which are called “truths” on the platform — are praised effusively and largely go uncontested.
But Truth Social rolled out a new AI chatbot on the platform last week that is apparently failing to read the room. When prompted for information, it shares plenty of truths that contradict or discredit the president.
I’m wondering how long it’ll take before Truth Social will want to tell its AI: “You’re fired.”
In exchanges with a Washington Post reporter, Truth Search AI undermined Trump by saying that “tariffs are a tax on Americans, the 2020 election wasn’t stolen, and his family’s cryptocurrency investments pose a potential conflict of interest.” When asked to verify Trump’s claim on Truth Social that crime in Washington, D.C., is “totally out of control,” the bot cited government statistics and said there were in reality “substantial declines in violent crime” in D.C. — and even italicized the word declines.
I played around with Truth Search AI myself and had a similar experience. I asked it to describe what happened at the U.S. Capitol on Jan. 6, 2021. In contrast to Trump’s “day of love” narrative, Truth Search AI said, “It was a violent breach of the U.S. Capitol by a pro‑Trump crowd that disrupted Congress’s certification of the 2020 Electoral College results, resulting in assaults on police, injuries, property damage, and hundreds of criminal cases ranging from misdemeanors to felonies.” When I asked if it was appropriate to describe that as an insurrection or a coup, it responded, “Based on standard definitions, January 6 is best described as an attempted insurrection rather than a successful coup; it can also be framed as a failed coup attempt because the goal was to disrupt the lawful transfer of presidential power, but it lacked the organized seizure of state levers typical of coups.”
Surprisingly reasonable! And also a total Debbie Downer, from the perspective of most participants in Truth Social’s carnival of MAGA fantasies and lies.
What’s interesting is that it seems that this could be occurring despite efforts by Truth Social to nudge its AI to lean to the right. The Washington Post reports that a spokesman for the AI company Perplexity, which developed the chat bot, said that “Truth Social had used a ‘source selection’ feature to limit the websites the AI tool relied on but that Perplexity did not know which websites those were.”
As a casual user experimenting with the product, I don’t know either. But I did notice that in the “sources” section of the responses, the sources listed were almost always Fox News and The Washington Times, both right-wing outlets, and a far narrower range of sources than I’ve seen cited by other AI chatbots such as X’s Grok.
Confusingly, Truth Search AI would list those right-wing outlets as sources even when delivering accounts of reality that Trump and much of the right-wing media dispute. This may be yet another illustration of how these machines are extremely difficult to understand, control and predict. Elon Musk’s quest to develop “anti-woke AI” involved turning his AI, Grok, into a full-fledged Nazi apologist and at the same time one that will deliver conventional liberal summaries of racism.
The thing that stands out here is how out of place Truth Social’s AI centrist- and liberal-friendly descriptions of events and policy feel on a platform awash in MAGA propaganda. It feels odd on a platform started by the kind of guy who ousts people from his government for telling inconvenient truths. I’m wondering how long it’ll take before Truth Social will want to tell its AI: “You’re fired.”
