Take any recipe on Allrecipes. Pick one. Chicken parmesan. Seven ingredients, measured and specified in Schema.org Recipe markup — machine-readable, tagged with quantities, cooking times, nutrition data, dietary restrictions. Now look at that list again. Not as a recipe. As data.
- 4 skinless, boneless chicken breasts
- 2 eggs, beaten
- ¾ cup seasoned bread crumbs
- ½ cup grated Parmesan cheese
- 2 cups marinara sauce
- 8 oz mozzarella, sliced
- ¼ cup fresh basil, chopped
- + Chicken Breast, boneless · Meat & Seafood
- + Large Eggs, 12ct · Dairy & Eggs
- + Italian Bread Crumbs · Pantry
- + Parmesan, grated · Cheese
- + Marinara Sauce, 24oz · Pasta & Sauces
- + Fresh Mozzarella, 8oz · Cheese
- + Fresh Basil, bunch · Produce
The transformation is trivial. Every ingredient maps to a product. Every quantity maps to a package size. Every recipe is already a shopping list. The only question is who captures the transaction.
The Pipeline
On May 13, Amazon retired its Rufus shopping AI and replaced it with Alexa for Shopping — an agentic assistant that creates custom buying guides, manages grocery orders across Amazon Fresh and Whole Foods, and remembers dietary preferences. It runs on Anthropic's Claude, Amazon Nova, and custom models trained on Amazon's product catalog, customer reviews, and licensed publisher content.
The publisher deals came fast. The New York Times signed for $20 to $25 million annually — news, NYT Cooking, The Athletic. The recipe-to-cart vision is explicit: ask Alexa what's for dinner, get a NYT Cooking recipe, add every ingredient to your Amazon Fresh cart. Recipe to cart to doorstep.
Then Condé Nast and Hearst signed multi-year agreements for Rufus — Good Housekeeping, Cosmopolitan, Vogue. Product recommendations, gift guides, kitchen equipment reviews. The content that, as Digiday put it, "naturally maps to consumer intent." Amazon now has 200-plus publisher deals across its Alexa+ platform. And it is licensing the Alexa for Shopping technology to other retailers through AWS — any retailer can launch an AI shopping assistant in 60 days using Amazon's architecture.
People Inc has no announced Amazon deal.
Three Deals and a Hole
People Inc has three AI licensing agreements. OpenAI pays at least $16 million a year for all-you-can-eat content access. Microsoft launched a pay-per-use content marketplace with People Inc as a launch partner. Meta signed a multi-year deal in December. Three tech giants, three licensing arrangements.
All three are the same kind of deal. Archive access. Here is our corpus — train on it, surface summaries, pay a licensing fee. The content flows into language models and comes back as paraphrased text on someone else's surface.
Amazon's recipe deal is structurally different. It is not about training. It is about transaction. The recipe doesn't get summarized — it gets executed. An ingredient list becomes a cart. A cart becomes a delivery. The publisher's content generates a purchase, not a page view. The monetization is commerce, not CPM.
This should be the most natural fit in publishing for the company that owns Allrecipes, Food & Wine, Better Homes & Gardens, Simply Recipes, Serious Eats, and MyRecipes. Lifestyle content that maps directly to products. Recipe structured data already tagged for machine consumption. Fifty-two million monthly visitors asking what to cook — and then needing to buy the ingredients.
What Extraction Looks Like
Allrecipes is 64% dependent on organic search. Google's AI Overviews are actively consuming that traffic — food creators reported 30 to 80 percent drops in 2025. Google's AI Mode now synthesizes what creators have called "Frankenstein recipes" — ingredient lists assembled from multiple sources, attributed to no one. Allrecipes' Schema.org markup, designed to make recipes discoverable, also makes them extractable.
Google takes the data for free. The same crawler that indexes for search also harvests for AI — what Vogel called, at Cannes, "an incredible abuse of market power."
Amazon would pay.
Not a flat licensing fee for archival access, but a commercial relationship where structured recipe data feeds a transaction pipeline. Where the ingredient list generates an order. Where the value chain doesn't end at content — it extends through cart, through delivery, through the purchase event that advertising is supposed to produce but rarely does.
Why It's Missing
Maybe the deal is in negotiation. Maybe People Inc turned Amazon down. Maybe the scarcity strategy Vogel described at Cannes — "Scarcity is key... everybody calls, because it turns out everybody needs us" — means they are holding out for better terms.
But the pattern is visible. People Inc's licensing deals are all content deals — send the archive to a model, receive a fee. Amazon's recipe partnership is a commerce deal — send the ingredient list to a cart, participate in the transaction. The distinction matters because it's the difference between being paid for what you made and being paid for what you sell.
Content licensing has a ceiling. A flat fee or a per-use royalty on training data. Commerce has a multiplier. A recipe that generates a $30 grocery order every time it's surfaced has a different economic profile than a recipe that generates a $0.003 CPM impression.
The ingredient list was always a shopping list. The question was never whether someone would build the pipeline. The question is whether the largest recipe publisher in the world will be in it when the pipeline opens.