I’ve never felt comfortable with AI searches, I always feel like I’m waiting for it to say, “I’m sorry Dave, I’m afraid I can’t do that”. ^[1] However, I decided to give ChatGPT a try to research potential side effects of a supplement recommended for chronic fatigue. Usually I spend hours trawling through multiple papers from clinical trials, but I was hoping AI would streamline this for me. I didn’t keep a copy of the exact conversation, but it went something like this:

Other than being slightly disappointed he didn’t say, “No, I’m sorry Dave…” I found it deeply concerning it was so wrong. How many people using AI would stop digging after being told there was a paper providing evidence? Or even stop after the first answer? Doing the research myself later showed that there was no evidence whatsoever that this herb caused joint pain, neither from trial results nor likely from the mechanism of action of the herb’s active ingredient.

A 2025 study by a Salesforce AI research team discovered that around one third of all AI answers are completely unsubstantiated, with that figure rising to 47% with Chat GPT. ^[2]

In discussing AI with some friends, few had realised that AI is a large language model (LLM), it is not reasoning nor has any intelligence whatsoever. In creating ‘answers’ to questions posed by a user, it scans massive directories of data and creates an answer based on the most statistically likely results. Each word it adds to that answer is based on a statistical likelihood. In reality, you are getting the most popular answer in the most statistically likely word order, rather than the most accurate or evidence-based answer.

It’s also worth bearing in mind that the data that it scans is primarily what is available on the internet. There is a giant wriggling mass of free nutrition advice available on websites and social media. A 2023 systemic review of quality and accuracy of online nutrition-related information concluded, “Online nutrition-related information is often inaccurate and of low quality. Consumers seeking information online are at risk of being misinformed".^[3] With all this information being used by AI tools, any chance of finding accurate and evidence-based nutrition information is minimal.

Despite this, the facts show that people are using the internet and social media as their ‘go to’ for advice on how to eat. In 2024, myfitnesspal released a survey about TikTok as a source of nutritional advice. A report on the survey noted, "The MyFitnessPal survey found that 87% of Millennials and Gen Z TikTok users have turned to the platform for nutrition and health advice, while 57% report that they are influenced by or frequently adopt nutrition trends they've found on the platform. In fact, of those who are influenced by nutrition and health trends on TikTok, 67% report that they adopt at least one of these trends a few times a week".^[4]

The part that worries me the most is that the way we eat has a profound influence on our long-term health. Like massive influence, not just changing how long we live but how well we are in the process. Research has found that 80% of chronic diseases can be avoided through lifestyle changes, especially a healthy diet and healthy weight.^[5] If we are following unsubstantiated or inaccurate advice or trends, what risk are we taking with our health? With our life?

The answer is to always check the source. For tips on checking the evidence behind any nutrition advice, see our science page