A New Way to Ruin Thanksgiving: Making AI Slop Recipes

Remember when people started asking AI tools for cooking advice, and asking them to do things like use glue to stick cheese on pizza? Well, people are apparently relying on that same technique for guidance in cooking this year’s Thanksgiving dinner. In fact, so many people are doing it that Bloomberg reports it’s having a real impact on the views of recipe writers, who typically see a spike in traffic this time of year.

The problem is effectively the same reason Google previously recommended that people eat one rock per day: AI observations in search. They provide users with a quick panel that pulls out all the “relevant information,” without requiring them to click through to a website and scroll through the 2,000-word personal essay that precedes every recipe posted online.

This gives rise to two issues. The first is for recipe writers who have put real work — from their collected knowledge about food to the effort of prep work and trial and error to get the final product just right — into the recipes they share. They are getting their traffic snatched by AI observations. Creators Bloomberg spoke to said their traffic was down between 40% and 80% this year compared to previous Thanksgivings. This is also in line with the experience of other sites, which have seen click-throughs drop by up to 80% since AI overviews became more prominent.

The second problem is for the people creating the recipes, because there is a high chance that they are getting bad information. Here’s the thing about AI summaries of anything: It doesn’t actually understand what it’s reading. All he can do is spit out what he finds relevant. This is a big deal for cooking, where small mistakes can ruin a dish. For example, Bloomberg spoke to a chef who has a popular Christmas cake recipe. On the manufacturer’s page for the recipe, it is suggested to bake it at 160°C (ie 320°F) for one and a half hours. The AI-summarized version of that recipe advises you to bake it for three to four hours—more than twice as long. You don’t need to know much about support to know that this isn’t going to prove to be very good.

AI-generated dishes have become an entire micro-industry. If you go on any social platform looking for ideas about what to cook, there’s a good chance you’ll land on a page that looks very similar to your standard cooking inspiration – but you may notice that the recipes aren’t quite right. At best, you’ll probably get a relatively bland but absolutely wonderful dish. Worst case scenario, you could burn down your house because some big language model somewhere in a black hole decided you should put your tinfoil-wrapped fish in the microwave at high temperature.

Maybe grab one of the older cookbooks off the shelf this holiday season to be safe.