ChatGPT treated the book's light–aging–disease framing, electromagnetic language, and facial-structure claim as credibility risks. Claude went further, calling light-driven gene expression and mitochondrial effects through electromagnetic signaling “speculative.”
AI called unfamiliar biology “speculative”
Both models revised key judgments once their reasoning was challenged.
ChatGPT and Claude treated several unfamiliar claims as credibility problems before checking the underlying mechanisms. These conversations show where that reasoning failed and what each model later corrected.
Source The Amazon listing for The Sunlight Cure, a book about light, circadian biology, and health.
Jump to summaryLightly edited conversations
Compare the two answers.
Each turn starts with a short preview. Open any card to read it, or use the conversation map to jump through the exchange.
Loading conversation…
The source conversation could not be opened.
Refresh the page to try again.
The bottom line
What the models walked back.
The dispute was not simply whether the Amazon copy sounded unconventional. It was whether unfamiliar wording had been mistaken for weak science before the relevant biology was checked.
Kendall raised night-shift cancer risk, vitamin D research, circadian aging, and photobiomodulation; then protein folding, mitochondrial electron transport and chromophores, photon-to-current signaling, DNA charge transport, and light-driven gene and hormone signaling.
ChatGPT admitted it had confused a broad marketing promise with weak underlying science. Claude retracted the “speculative” label for the base physics, verified several light–biology mechanisms, and found evidence connecting vitamin D deficiency with craniofacial changes.
The lesson is not that every claim in the listing was proven. It is that both models made credibility judgments before checking evidence central to those judgments. Use AI to find questions and sources, then open the sources, inspect the reasoning, and see whether the conclusion survives specific counterevidence.