Baker beware: How I was fooled by an AI-generated recipe

By Brenda Goodman, CNN

(CNN) — It started with the cutest little desserts: chocolate acorns with nut-covered caps that popped up in my search for Thanksgiving cookies on Pinterest, a site I visit for inspiration and some step-by-step instructions. There were also chocolate-dipped strawberries that you could turn into little turkeys using pretzel sticks, with marshmallow pieces for the drumsticks, but the acorns looked easier.

“Who wouldn’t love these things?” I thought.

The recipe said you could whip up a batch in 45 minutes. I splurged and bought my favorite Lindt chocolate bars to make them.

It ended at 1 a.m., after about five hours of effort – including an emergency run to the grocery store for a missing ingredient – with parchment-covered baking sheets spread across my kitchen, holding misshapen globs of chocolate-dunked peanut butter sandwich cookies that didn’t come close to resembling acorns. Despite my many attempts at adjusting the temperature of the chocolate and MacGyvering the cookies — I cut them, stacked them and stuck them together with chocolate — they just could not match the acorns in the photo.

I’m usually pretty good at these things (pie crust being a frustrating exception). I’m an experienced baker, and this should have been an easy project, more assembly than baking. I wasn’t going for gourmet; I needed a lot of treats for a church event and was short on time.

I finally had to admit that no matter how many times I reread the recipe and attempted to follow the instructions, it wasn’t going to work. I was stumped.

Then it dawned on me: Could this recipe have been generated by AI? Had I been had?

Even as I had been reading through the instructions, I kept wondering how the blogger got the Nutter Butter cookies — which are flat, with a peanut-like curved shape — to turn into the nice rounded shape of the acorns. The recipe recommends holding the cookies by the pointier end “to create the more natural looking” acorn.

OK, I thought. Maybe I need to just trust the process.

“They’re ridiculously cute, surprisingly simple, and honestly? They’ve become the most photographed dessert at every family gathering I bring them to,” gushed food blogger Anna Kelly, who described herself as a professionally trained pastry chef who runs the website DessertsPro.com. “My sister-in-law now requests them specifically, and I’ve made them at least twelve times since I first stumbled upon the idea three years ago.”

I fell for it. I bought three packages of the regular-size Nutter Butter cookies and searched several stores for the bite-size ones, which were supposed to be for the caps. Then I noticed a problem: The bite-size cookies were much smaller than the ends of the big Nutter Butters. It didn’t seem like they would work to make the overhanging acorn cap.

I still didn’t question the recipe. Instead, I thought I had bought the wrong size cookies. Was there a medium-size Nutter Butter? Oh, the recipe said Kelly sometimes uses Nilla wafers. That made more sense, but it was getting late, and I wasn’t going back to the grocery store.

Thankfully, the chocolate melted like it was supposed to, but it was pretty thin, and I was losing hope that it would coat the cookies thickly enough to look like the acorns in the photo. I tried popping it in the freezer to thicken it.

The website DessertsPro.com says Kelly created it to share her favorite tried-and-true recipes with home cooks. Her “About Us” page promises “clear, simple instructions,” easy-to-find ingredients, “honest tips and tricks” and “recipes that actually work.”

The problem is that neither Kelly nor her creations appears to be real.

‘Bad actors’

I attempted to reach the creators of the site through a “contact us” email as well as through the contact form but got no response by my deadline. I also attempted to contact Kelly through another food blog set up with her name and profile, MuffinIdeas.com. This site uses the same photo and syrupy text that talks about Kelly’s nostalgia for her grandmother’s baking, but the bio is slightly different: It says she’s not professionally trained but self-taught by reading countless baking books by the likes of celebrated chefs like Dorie Greenspan, the very real author of “Baking with Dorie,” and David Lebovitz, author of “Ready for Dessert.”

Emails sent to both websites were returned as undeliverable.

“This site is almost certainly AI,” said Adam Gallagher, who runs the food blog Inspired Taste, which he started with his wife, Joanne, in 2009.

The Gallaghers have been outspoken about the growing volume of AI-generated images and recipes meant to look like genuine food blogs that have permeated social media sites like Facebook and Pinterest.

They say that not only are these fake sites siphoning traffic that used to go to real bloggers, AI technology is scraping sites like theirs and emulating their style to create the computer-generated content it churns out.

It’s not that artificial intelligence, as a tool, is necessarily a bad thing, says Tom Critchlow, executive vice president for audience growth at Raptive, a company that helps bloggers and other creators monetize their content. It would be one thing, maybe, for a real chef or food blogger to use some AI to help refine their writing or get new ideas.

But that’s not what’s happening here, Critchlow says.

“These are bad actors generating hundreds of thousands of brand new, fresh domains that have never been seen before, using AI to generate hundreds and thousands of AI-generated recipes, images, et cetera, and covering the websites with ads,” he said.

“Pinterest has been overrun by these spammers,” Critchlow said. “Almost any kind of search term or any kind of exploration on Pinterest in the food and recipe space leads you to these websites, which are just spam top to bottom.”

Even sites with worthless content can make money when people click and scroll through them, just as I did when trying to figure out what I was doing wrong.

On average, content creators earn about $30 to $50 for every 1,000 page views to their sites, according to Raptive.

One food blogger might spend hours creating, testing and shooting a recipe to earn that rate, but AI-generated sites can earn the same money for content created in minutes.

“When you look at the scale and volume of these things, there are whole networks of sites, hundreds of sites, made by a single individual,” Critchlow said, noting that videos on YouTube also show how to do this.

Raptive says it kicks creators and websites out of its network if it discovers that they’re using AI tools to make spam recipes. This year alone, the company says, it’s blocked nearly 600 sites – of about 6,000 in its network – for AI-related issues.

The AI clues I missed

When I read back through my apparently fake recipe, I find clues I initially missed.

In the introduction, “Anna Kelly” writes that her nephew calls her Aunt Sarah, rather than Anna. The text says the site was created in 2019, but it – like MuffinIdeas.com – was copyrighted in 2025. Had I clicked around on some of the other recipes, I might have noticed that some of the photos were suspect: The blueberries in a loaf of sourdough bread are impossibly round purple craters, and the cupcakes or muffins (cuffins?) in Kelly’s profile pic have frosting that magically fades into the cake.

The recipe I attempted didn’t have any ratings or comments – another red flag, since I normally rely on the cooks who’ve gone before me for tips.

“People have a shiny look. It’s almost like everyone’s a little plasticky. That’s a telltale sign,” Gallagher said. “It’s like, everything’s in a little bit of Saran Wrap.”

I found the recipe on Pinterest, but the site links to accounts on Instagram, Facebook and X, though these profiles have little to no followers.

“Consumers are being duped by all the platforms,” Gallagher said.

He says platforms like Pinterest aren’t really motivated to filter out AI spam because they make money no matter where the recipe comes from, as long as it keeps people scrolling.

Meta, the parent company of Facebook and Instagram, didn’t respond to a request for comment by deadline.

Pinterest did not offer a direct comment but said it disagrees with the idea that it isn’t motivated to help users who want to avoid AI-generated content. It says it will certainly hurt its business if people get frustrated by the content that’s on the site and stop coming.

It also says user feedback led the company to create an AI filter that users can turn on in their settings, which can help keep computer-generated content out of their feeds. It added the filter to the Food and Drinks category in mid-December. AI images are also supposed to be labeled when you enlarge them. The chocolate acorn photo I clicked on doesn’t have such a label, though.

Gallagher and Critchlow said they’ve tried the filter and didn’t notice a difference in their feeds. The company says the tool is new and should get better over time.

Pinterest also pointed me to an audit, performed by the digital security reporting site The Indicator, testing the promises of platforms to label AI-generated content, which was posted by the reporters. The audit found that five major platforms repeatedly failed to label AI-generated content on their sites, including content created using their own AI tools.

Pinterest was the most successful at labeling AI images, the audit found, but still had a success rate of 55%.

“I want to give Pinterest a little bit of credit, because they were early in introducing a way for users to filter AI content out of their feeds, which I think is something that all platforms should be doing,” said Alexios Mantzarlis, a former Google executive who is co-founder of the Indicator and did the AI testing for the story.

“So Pinterest, I think, is understanding from its users that the stuff sometimes is very annoying, and it appears to be reacting,” Mantzarlis said, though its tools aren’t great at flagging AI just yet.

The company says it doesn’t want to ban AI completely because some people find that it inspires them.

The Gallaghers think there are more pragmatic reasons some social media sites haven’t been aggressive in tackling AI content.

The more content a platform has and the longer a user continues to scroll, the more ads it can show, Joanne Gallagher said.

Baking a fake

Spam sites aren’t the only way AI makes up recipes. Adam Gallagher points out that if you Google a particular thing you want to cook, the AI-generated summary that appears at the top of the page may be a complete recipe that was created with parts of recipes taken from the Inspired Taste blog and others that are smushed together and ultimately might not make anything worth eating.

“We call them Frankenstein recipes,” he said. Although the summaries link back to their sources, Gallagher said, few people click through to it, relying instead on the summary results.

Google says it cares deeply about the quality of the information on the internet and points to its policies that prohibit producing content at scale for the primary purpose of manipulating search rankings. Last year, the company says, it launched a policy against scaled content abuse that targets the creation of many low-value pages for the primary purpose of manipulating search and news rankings.

As for the AI summaries in its search, the company says users were looking for these kinds of tools, which are meant to highlight and help surface content rather than prevent traffic to it.

“AI Overviews are often a helpful starting point to learn about a dish, but we see that people still want to go and read original recipes from creators. We’re focused on making it easy for people to discover and visit useful sites that have a good user experience.”

The Gallaghers said the AI spam, or slop, doesn’t hurt established sites like theirs as much because they have a loyal audience that comes to them directly for their recipes.

Instead, they worry about food bloggers who are just getting started, who depend on social media and search tools to generate traffic.

“Unfortunately, on all of these platforms, you have high-quality, trusted creators intermingled with all of this AI slop,” Adam Gallagher said.

When it’s hard to tell them apart, it erodes trust.

“When they keep chiseling away that trust, then you’re going to trust us less too, because you’re not going to find us,” Joanne Gallagher said. “You’re not going to be using Pinterest or Facebook or Instagram or whatever to actually find recipes anymore.”

As for me, even though I live and work online and Googling has become my default way to find just about everything, I find myself coming out of this episode with a new appreciation for my cookbooks.

Who knows, maybe AI will end up boosting book sales, which wouldn’t be such a bad thing.

The-CNN-Wire
™ & © 2025 Cable News Network, Inc., a Warner Bros. Discovery Company. All rights reserved.