Blog

The Recipe That Failed You Was Never Cooked

PUBLISHED Oct 1, 2026
A flat grey sign with a red ring around two recipe-card puzzle pieces whose shapes do not fit together
0:00 / 0:00

The recipe said forty-five minutes. It is one in the morning, there are parchment-covered baking sheets across your kitchen, and what was supposed to be a big tray of chocolate acorns for an event is a collection of misshapen, chocolate-dunked cookies that look nothing like the photograph. You have reread the instructions more times than you can count, and you have reached the only verdict left: you must have bought the wrong size cookies.

That night happened to Brenda Goodman, an experienced baker writing for CNN, and the verdict is hers. The recipe came from a smiling pastry chef called Anna Kelly, who said she had made it at least twelve times. Neither Kelly nor her recipes, Goodman found, appears to be real.

The Recipe Nobody Made

Goodman's recipe came from a fake food blog, one of a great many. The more common version now arrives without any blog at all. Search for a dish and the answer often sits at the top of the page, written by Google's AI: an ingredient list, a method, sometimes a real blogger's name attached for reassurance. Food writers have a name for what it frequently is. "We call them Frankenstein recipes," Adam Gallagher of the blog Inspired Taste told CNN: ingredients from one source, steps from another, presented as a single dish. As the Gallaghers put it on NBC News, "recipes don't work like that, which is why they are riddled with errors."

The stitching is the norm, not the exception. A 2025 Pew Research Center study of 900 Americans' Google searches found that 88% of AI summaries cited three or more sources, and that people clicked a link inside the summary on just 1% of visits. This is now simply how people search: in Pew's 2026 survey, six in ten American adults said they read AI search summaries.

Read those numbers from the cook's side of the counter. The recipe arrives with its authors folded away, and ninety-nine times in a hundred nobody unfolds them. That is not carelessness. The page is designed to be the answer. So when the dish fails, there is nobody in view to doubt except you.

Where the Blame Lands

Cooks are remarkably reliable about blaming themselves. "I still didn't question the recipe," Goodman wrote. One Inspired Taste reader, after following a version of their recipe that was missing instructions, wrote to them: "I consider myself a decent cook. I was left discouraged and puzzled." When the cook does not take the blame, it goes to the blogger whose name was on the summary. Of one stitched version of their pho, the Gallaghers wrote that readers "could make the recipe, not like it, and then blame Inspired Taste for the results." The one party that never gets blamed is the system that assembled the recipe in the first place.

It is a familiar kind of unfairness. Recipes have always leaned on knowledge they never state, the way season to taste assumes you already know the destination, and a format that fails the cook tends to get read as a cook who failed, a pattern ADHD cooks feel first. The Frankenstein recipe is the extreme case: a recipe nobody tested, delivered with the confidence of one that was.

A Recipe Is a Promise

Underneath all of this is something food writers have always understood and rarely needed to say out loud. Bloomberg, reporting on independent food creators last year, put it as "the simple promise of a recipe: that someone has actually cooked it before you have." In the same piece, Eb Gargano of Easy Peasy Foodie described an AI-assembled version of one of her cake recipes that would have a six-inch cake baking for three to four hours at 320°F. "You'd end up with charcoal!" she said. "No matter how clever the AI is, it can never actually test a recipe in a real kitchen and see how it works."

Google says it understands this. "People still want to go and read original recipes from creators," the company told CNN. They may well want to. On Pew's numbers, one visit in a hundred gets there. The changes Google has made this year move links to recipe sites higher up the page, but by Inspired Taste's account the stitched recipe is still the default answer.

So here is the reframe, and it is a kind one. A tested recipe makes you a promise, and if it fails, it is fair to wonder what went wrong in your kitchen. An untested one made no promise at all. If the dish collapsed, the first question is not what you did wrong. It is whether anyone, anywhere, ever cooked the thing you were following.

We build AI for the kitchen, and we said a year ago that we cannot trust AI to create recipes. Nothing since has changed our minds. A recipe with a person behind it, someone who cooked it, got it wrong and fixed it, is still the only kind worth your evening. And when you find a cook like that, you have found something worth going back to.

Want every recipe you cook to come from a real person? Learn more at ChefTalk.

At ChefTalk, we're building a voice-first cooking companion that transforms any recipe into hands-free, step-by-step guidance. Learn more.

Sources