Three Fields

Google restructured how recipes appear in AI Mode. Twice this year. The March update shifted from synthesizing ingredients and instructions from multiple sources into a single blended recipe to surfacing a curated gallery with links to creator websites. The June update added visual recipe cards pinned to the top of AI Mode responses.

Each card displays exactly three things: creator name, aggregate rating, and ingredient count. Plus an image.

Those three fields map directly to Recipe schema properties: author.name, aggregateRating, and the length of recipeIngredient. If your structured data is clean, you get a card. If it isn't, you're invisible in the interface that's replacing search results for cooking queries.

For People Inc, this is not an abstract SEO question. It is the most concrete technology problem their food portfolio faces in Q4 2026.

Five Brands, One Hundred Thousand Recipes

People Inc's food vertical spans five brands: Allrecipes (~104M monthly visits, 48.5% from organic search), Food & Wine, Simply Recipes, EatingWell, and Serious Eats. Together they publish over 100,000 recipes. Allrecipes alone is the second-largest cooking and recipes website globally.

These brands built their traffic on Google search. Nearly half of Allrecipes' desktop traffic comes from organic search. When Google changes how recipes surface, it changes these brands' economics directly.

And the change has been brutal. Food bloggers reported traffic drops of 30% to 80% throughout 2025. Organic click-through rates for queries with AI Overviews fell 61%. In Google's AI Mode specifically, the zero-click rate reaches 93%. Recipe content is among the hardest-hit categories because Google can now display the entire product — ingredient list, step-by-step instructions, cook time — without a click.

The Card Is the New Ranking

Google's June update changes the game again, but this time partially in publishers' favor. Robby Stein, Google's VP of Product for Search, said the goal is to "help people discover and visit recipe pages." The recipe cards — 3 to 4 per query — are a curated discovery interface, not a synthesized answer. Users see creator names, ratings, and ingredient counts, then click through to the actual recipe.

This is directionally good. But it creates a new selection mechanism. Google no longer ranks ten blue links for recipe queries. It selects three or four cards. The criteria for selection are structured data fields that must be present, complete, and correctly formatted in the page's schema markup.

At five recipes per query, three or four cards means 60-80% of eligible recipes don't get surfaced even with perfect schema. The competition for card placement intensifies every schema field.

The CMS Question

For a single food blog, fixing recipe schema is a weekend project. For 100,000+ recipes across five brands running on a shared CMS platform, it is an infrastructure problem.

Graphene-Selene — People Inc's proprietary CMS — generates the structured data for every recipe page. The schema quality of 100,000 recipes depends on how Graphene-Selene templates inject author.name, how they compute aggregateRating from user reviews, how they serialize recipeIngredient arrays, and whether duration fields are populated from structured editorial input or parsed from free text.

This is where the technology strategy gets concrete. Three questions for the engineering team:

First: author attribution at scale. Community-contributed recipes on Allrecipes historically credit user handles, not named creators. Google's AI Mode cards surface author.name as a primary trust signal. A recipe attributed to "Allrecipes Staff" or a username competes differently than one credited to a named, recognizable food writer. Across five brands, the author attribution model varies — Food & Wine has named editors, Serious Eats has Kenji López-Alt, Allrecipes has 2.1 million registered community members. The CMS needs to handle this variation in a way that maximizes card eligibility per brand.

Second: ingredient serialization. The recipeIngredient field should be a JSON-LD array of individual items. The card displays ingredient count, which means Google is parsing the array length. Recipes where ingredients are stored as a text blob rather than structured items won't generate a clean count. At 100,000+ recipes, the question is whether Graphene-Selene's data model stores ingredients as structured entities or renders them from editorial prose.

Third: duration fields. The March update explicitly surfaced cook time as a decision-aid signal. cookTime and prepTime must be ISO 8601 durations in the schema markup. Google confirmed these are prioritized in AI Mode selection. Recipes without clean duration data exclude themselves from one of the specific signals Google said it is now weighting.

The Audit

The actionable engineering work is a structured data audit across the food portfolio. Not a schema validation pass — those catch malformed JSON-LD. A completeness and quality audit: what percentage of recipes have all five card-eligible fields populated, correctly typed, and generating valid rich results?

The brands that win the card placement race over the next twelve months will be the ones whose CMS generates the cleanest structured data at scale. Not the ones with the most recipes, the best editorial voice, or the highest domain authority. The ones whose engineering teams treated recipe schema as critical infrastructure rather than an SEO checklist item.

For People Inc, that means Graphene-Selene. The CMS is no longer just the publishing tool. It's the mechanism that determines whether 100,000 recipes remain visible in the interface that is replacing Google search for cooking queries. Three fields. Billions of impressions.

Previously: The Recipe Graph on Allrecipes entity structure. The Crawler Problem on AI opt-out traps. The Performance Tax on CWV and D/Cipher's INP weight.