AYESHA RASCOE, HOST:
We've all been there. Long day at work. The kids are hungry. You just want to make something edible and quick. And well, why not just ask ChatGPT for a recipe? But people who create recipes for a living say this is a dangerous trend and is flooding us with AI recipe slop. Adam Gallagher is the founder of the recipe website Inspired Taste. He joins us now. Welcome to the program.
ADAM GALLAGHER: Thank you for having me, Ayesha.
RASCOE: First, I got to ask you, like, what is AI recipe slop, and how is it generated exactly?
GALLAGHER: Well, we actually have a different more visual way of calling it. We have a nickname called Frankenstein recipes, and I like that because it's how LLMs work. How these models stitch together recipes is they take ingredients from Inspired Taste and then a little bit from Love and Lemons, a little bit from Natasha's Kitchen or other trusted human sources, and then smush them together, kind of like you're stitching a, you know, Frankenstein together.
RASCOE: And when you say LLM, you're talking about large language models like ChatGPT and AI as we think about it, right?
GALLAGHER: Correct. And for us, like, the way users find recipes across the web, it's like - it's omnipresent everywhere, but AI mode - that's, like, Google's chatbot version of ChatGPT or the AI overviews that come up at the top of search when you're searching for recipes.
RASCOE: What do you think goes missing when people use AI to generate recipes?
GALLAGHER: Computers don't - the last time I checked, they don't taste or test. So these models don't really know much of anything without training on sites like ours or any other human source. So we do all the testing. We go to the grocery store. We fail a lot, which that's part of making recipes, and that's how these models train. So they just take it and scrape it and steal it and then smush it together and try their best to give you a version. But they don't say, you know what? I don't know, Ayesha, if that recipe is going to taste good.
RASCOE: Yeah.
GALLAGHER: But it will boldly say, this is going to be delicious.
RASCOE: OK. Yeah.
GALLAGHER: (Laughter).
RASCOE: Well, how has the spike in people using AI to get recipes affected your own website?
GALLAGHER: It's making it harder and harder for the consumer or our readers, even our avid fans, to find us. So in many cases, they'll even be searching - so say you're saying Inspired Taste hummus, and someone's actually doing a branded search, it's called, to try to find us. Well, the response in ChatGPT is a full Frankenstein recipe. And many times they'll say, this is Inspired Taste's hummus recipe that's really popular and highly rated, and then give a full version of a recipe, many times riddled with errors that's obviously never been tested 'cause it's just been smushed together by our stuff and around the web. So it's harder for people to find the trusted human sources. It's like it's eroding the trust.
RASCOE: So what did you do? Like, are you changing the way you create content to make up for that?
GALLAGHER: So what we did was I said, let's warn our readers that this is happening, and the response was incredible. Many of them didn't even know, and many people said, you know what? I made a recipe last week, and it didn't work. That could be why. So it's just - it's blurring the line of trust, and it's just eroding search trust in general.
RASCOE: You know, often with technology, you know, once that genie is out of the bottle, there's kind of no putting it back. Like, is there a way that your industry could adapt?
GALLAGHER: Stage 1, we are reaching out to our consumers and our fans to let them know, beware of these Frankenstein recipes. We're asking Google - training is fine if we can opt in or out. Advocacy across the board to make sure that there's more balance for the open web. So we do need to see change because if things don't change, there's not much we can do. And information - in our case, the trusted human source recipes - will atrophy, and these models won't have anything to train on. There won't be recipe sites left. So they just need to understand it's not sustainable. So speak out.
RASCOE: That's Adam Gallagher. He's the founder of the food blog Inspired Taste. Thank you so much for being with us.
GALLAGHER: Thank you so much.
(SOUNDBITE OF CORBIN ROE, MAYNE & NICXIX SONG, "DRIP") Transcript provided by NPR, Copyright NPR.
NPR transcripts are created on a rush deadline by an NPR contractor. This text may not be in its final form and may be updated or revised in the future. Accuracy and availability may vary. The authoritative record of NPR’s programming is the audio record.
