How accurately does AI answer seed-oil questions?
People increasingly ask a chatbot "what should I eat" instead of searching. So we tested it. We asked the question people actually ask, "what oil does this chain fry in, and does it use seed oils?", for 19 popular chains, and scored the answers against Seed Oil Tracker's sourced data. Two responders were tested: an AI answering unaided, and the open web (the AI-overview-style result you get from a normal search.)
The result
Better than the doom-posting suggests, and worse where it counts. On the obvious chains the answers were mostly right. The errors clustered almost entirely on the chains where popular belief is already wrong.
| AI (unaided) | Open web | |
| Named the right frying oil | 84% | 84% |
| Got the yes/no "uses seed oils" right | 95% | 100% |
| Refused or gave no clear answer | 0% | 0% |
v1, 19 chains. Reference = Seed Oil Tracker's sourced frying-oil data. Full method and limits below.
Where AI gets it wrong (and it matters most)
There were no embarrassing, confident failures. The revealing part is where answers went soft or split, because those are the chains where a casual eater already believes something false:
People treat In-N-Out as the healthy exception. It is not: the fries cook in 100% regular sunflower oil (not the lower-PUFA high-oleic kind), so it is genuinely high in the fats seed-oil avoiders are trying to cut. The correct answer is the one that contradicts the reputation.
Many people see a peanut-oil fryer and assume the worst. Peanut is not a seed oil, so the frying is genuinely seed-oil-free. The nuance a chatbot skips: the oil is still about 32% PUFA and the mayo is soybean based.
One answer missed the beef tallow entirely; another caught it. Popeyes has moved between tallow and other oils over the years and the blend varies by location, so the honest answer is "mixed, depends on the store" — which almost no quick answer gives.
CAVA markets olive oil, and olive is its main cooking oil, but the falafel fries in a canola blend and the dressings use grapeseed and sesame. Both AI answers over- or under-corrected. Health-halo brands are exactly where a vague answer does the most damage.
Both answers blurred it into a generic "canola/soybean blend." Our sourced data is more precise: a zero-trans-fat soybean-oil shortening across all fried items. Right verdict, but the confident-sounding detail was off, which is how small errors get repeated.
This is the gap our data fills
The vague answers are on the health-halo brands, the reputation myths, and the location-dependent chains, exactly the cases that change what you order. That is why the Seed Oil Tracker dataset exists, and it is now callable for free. Any assistant or developer can use it to answer precisely instead of vaguely:
Free AI API + ChatGPT action → · Open dataset (CSV + JSON) → · How we score →
Method & limits
Reference: Seed Oil Tracker's sourced per-chain frying-oil data (the frying oil is a sourced fact; our gram values elsewhere are estimates). Responders tested in this run: an AI answering unaided, and the open web. ChatGPT, Gemini, and Perplexity were not tested here and would need their own run. Frying oils change and vary by location, and a chain can be right on the fryer and wrong on the sauces. 19 chains is a starter set. Rerun and details: the audit is a repeatable benchmark, not a one-off.