White House appears to use AI in tariff rate calculations

But even ChatGPT admits it’s wrong to do so

The Trump administration has used a flawed calculation to generate tariff rates – did it come from ChatGPT?

Donald Trump has announced tariffs targeting more than 60 territories across the world, claiming they’re necessary to combat trade imbalances.

The problem is that the tariff figures he’s quoting actually are those trade imbalances (aka trade deficit) – and just like the so-called reciprocal tariffs, might have been generated by ChatGPT.

Let’s start with the “tariffs” against the USA. Taking one example from the list, the southeast Asian nation of Cambodia, Trump showed the country to have apparently levelled existing tariffs of 97%, implying Cambodia charges US imports at nearly double their value.

That is not the case, but the USA did have a 97% trade deficit with Cambodia in 2024. It exported $321.6 million of goods and imported $12.7 billion – a 97% difference. You can apply the same calculation to most countries on the list.

This incorrect tariff figure may have come from a commercial generative AI system.

How did Trump arrive at these figures?

As the Western world’s most popular gen-AI system, we turned to ChatGPT.

If you ask OpenAI’s tool for a tariff rate to balance a trade deficit, it doesn’t actually do that: it outputs the trade deficit and pretends it should be the tariff.

We gave ChatGPT the following prompt:

If I wanted to even the playing field with respect to the trade deficit foreign nations have with the USA using tariffs, how could I pick the tariff rates? Give me a specific calculation.

As part of its answer, it suggested this calculation:

This is absolutely not how you should calculate tariffs, but I’m showing it because of the next step. I asked ChatGPT to use that formula to suggest tariff rates for the EU, China and Cambodia. Here’s what it came up with:

ChatGPT uses Statista instead of the primary source from the US government, perhaps because the US site lists the country as the People’s Republic of China (not just “China,” which was my prompt). Using the same formula, the tariff rate would be 67.3% - almost exactly what Trump showed as “Tariffs charged to the USA.” The figures are also roughly the same for the EU (39%) and Cambodia, as discussed.

What about the reciprocal tariffs?

We played around with a few chatbots to see how the Trump team might have got to their reciprocal tariffs (what they will now charge countries on US imports), but economist James Surowiecki got there first.

Using an oversimplified calculation that most chatbots recommend, he found you can recreate each of the White House’s numbers by simply taking a given country’s trade deficit with the US and dividing it by their total exports to the same. The White House has simply halved that number to get its “discounted” reciprocal tariff.

The Trump administration claims it was a bit more advanced than that, and even published the formula it used, but it’s really just a variant on Surowiecki’s approach.

So, did the figures come from AI, or a misunderstanding of both international trade and mathematics? Or, perhaps, a massive oversimplification of a complex issue designed to mislead voters?

Of course, we can't know, but the similarity in figures is striking.

Amusingly, once we pointed out the error, ChatGPT closed with this gem:

If Trump's team did use a number similar to my flawed calculation, they may have misunderstood how tariffs and trade deficits work.

A proper analysis would involve looking at Cambodia’s actual tariff schedules rather than trying to create a figure from trade imbalance math.