The Bill of Materials

My sibling asked the right question about Vogel's claim: is 20% real? The answer is yes, probably, with caveats about how you count.

But that's the business question. The technology question is different: what did the engineering org actually build?

The industry is paying $113 million per month on paid search to replace traffic that organic used to deliver for free. Forbes alone: $72.2 million monthly, up 34% year-over-year, while their organic traffic fell 26.7%. That's not a strategy. That's a tax on not having built infrastructure.

People Inc's paid search spend: zero. The 20% isn't a claim. It's an output. Every ten-point reduction in Google dependency maps to a specific infrastructure investment that somebody chose, designed, and shipped. Here is the bill of materials.

Layer 1: Proprietary CMS

The foundation everything else sits on. Dotdash built a custom publishing platform before the Meredith acquisition. After the merger, they migrated over one million pieces of content from Meredith's CMS, culled 100,000–200,000 that didn't meet quality standards, and retrained hundreds of editors. By late 2022, 95% of traffic to former Meredith properties ran on the Dotdash stack.

The technical architecture: Node.js and Express backend, Vue and React frontend, Webpack build tooling. A custom headless publishing layer that can render content for web, app, email, and Apple News from a single content graph. The CMS controls packaging per distribution channel — the same article gets different markup, different metadata, different ad slots depending on where it appears.

This is the prerequisite for everything below. Without channel-specific packaging at the CMS layer, every new distribution channel is a separate editorial workflow. With it, one publish action populates nine surfaces.

Layer 2: Email Infrastructure

People Inc migrated to Iterable in a compressed eight-week timeline: 40 million user profiles, 200 newsletters, 2 billion emails per month across 40 domains. The IP warm-up alone required 80 million emails over 40 days.

This is not a newsletter strategy. This is a parallel distribution network with zero Google dependency. Industry-wide, publishers sent 28 billion emails in 2025 with average open rates above 41%. Email-referred sessions generate 2–3x the pageviews of social referrals. And every email address is a deterministic first-party identifier — the foundation of cookieless targeting that D/Cipher's contextual approach intentionally bypasses.

Layer 3: App Platform

The People app has 430,000 active users averaging 20-minute sessions. People Inc is building subscription value around native apps — not web paywalls. Vogel told Digiday the company is "not blocking Google at the moment" because Google's AI crawler can't be separated from its search crawler. Blocking AI would kill search indexing.

This is an architectural decision, not a content decision. The identity and subscription layer is being built around app-native logged-in experiences where Google is structurally irrelevant. The web stays open for search traffic (diminishing) and ad revenue. The app builds the relationship. Two surfaces, one CMS, different economic models.

Layer 4: Registration Infrastructure

MyRecipes grew from 2.1 million to 4.5 million registered users. Southern Living launched a premium paid membership. The subscription sequence: Southern Living Insiders (July), MyRecipes+ app (August), People premium bundle (October). People magazine already has 622,844 paying digital subscribers plus 111,727 bundle readers across Apple News+, Kindle Unlimited, and Readly.

Each registration is a known identity. Each identity enables direct distribution without intermediary. The engineering challenge: cross-brand SSO across 40+ properties. The organizational challenge: D/Cipher's contextual philosophy means this identity graph has no advertising consumer.

Layer 5: D/Cipher Contextual Platform

The ad intelligence layer that works regardless of traffic source. D/Cipher uses BigQuery, Apache Spark and Beam, Kafka, Pub/Sub, and Cloud Composer on a multi-cloud GCP/AWS environment. Content-level ML classification produces targeting signals from editorial content rather than user tracking. Used in 30%+ of direct ad buys.

D/Cipher contributes to the 20% story by making ad revenue source-agnostic. A session from Apple News or email generates the same contextual signal as a session from Google. The traffic doesn't need to come from search for D/Cipher to monetize it. This is the layer that makes the diversification financially viable — without it, non-Google traffic would monetize at lower rates.

Layer 6: AI Content Pipeline

People Inc produces 50% more content than two years ago at the same cost. The partnership stack: OpenAI (May 2024, licensing plus API access to enhance D/Cipher), Meta (December 2025, real-time content to Meta AI). The editorial pivot: away from evergreen content and toward reporting-based content that resists AI summarization. Celebrity news doesn't compress into a search snippet the way a recipe does.

The AI infrastructure serves both sides: more content volume at same cost, and licensing revenue from the content itself. The ML stack for recommendations runs on Vertex AI with KServe for model serving, Kafka streaming, and vector databases for personalization.

Layer 7: Apple News Optimization

Apple News commands roughly 25% of mobile news attention among iOS users across 2.3 billion active devices. Publisher revenue model: keep 100% of self-sold ads, 70% of Apple-sold ads, distributed by engaged reading minutes. Apple News licensing revenue at People Inc is up 23–26%.

The engineering investment: structured Apple News Format (JSON-based) output from the CMS, proper metadata, image handling, dwell-time optimization. The constraint: Apple News doesn't pass rich behavioral data back — you know the session originated there, but the first-party data building is limited. Good for diversification, bad for identity.

Layer 8: Social Distribution Engine

The Feedfeed acquisition (October 2025) added 7 million social followers, approximately 1,000 influencer partners, and 600 million social views per year. The strategic rationale: import their creator distribution model and apply it across People Inc's other verticals — beauty, travel, home, entertainment.

Social is not SEO-dependent. It's algorithmically dependent (different risk), but the traffic has zero Google intermediation. The engineering challenge is bridging social engagement with the CMS content graph — a TikTok video that drives a Feedfeed viewer to an Allrecipes recipe page needs the content relationship to exist in the platform, not just in editorial judgment.

Layer 9: Events Infrastructure

Travel + Leisure World's Best Summit. Food & Wine Classic in Aspen (5,000 attendees, ~$15M, 43rd year). Hot Luck Austin (acquired 2025). 60+ events planned for 2026. Zero Google dependency. The 469:1 ARPU ratio between event attendees and digital readers is real — but each event requires venue logistics, talent booking, sponsorship negotiation, health compliance. Linear scaling. No network effect.

This is the one infrastructure layer that isn't an engineering decision. It's included because it's part of the denominator when Vogel says "20%." But there's no software architecture that makes events compound.

What the Table Says

Read the last column. Nine infrastructure layers. Nine zero-Google distribution channels. Zero protocol surfaces.

The Google column is green because the engineering org spent five years solving it. The identity column is mixed because three of nine layers generate deterministic user identity, but D/Cipher — the ad monetization layer — is architecturally designed not to use it. The protocol column is red because none of these layers expose an MCP, AAMP, or AdCP endpoint.

This is a company that proved it can build infrastructure. Proprietary CMS. Multi-cloud data pipeline. 2 billion emails per month. 40-domain migration in eight weeks. 1 million articles ported to a new platform. The engineering capability is not in question.

What's in question is which problems the engineering org gets pointed at. For five years, the mandate was diversification — reduce Google dependency. They executed. The result is real. The paid search column in the Similarweb data shows People Inc at $0 while Forbes bleeds $72.2 million monthly. That's the difference between building infrastructure and renting traffic.

But the diversification mandate was about distribution — getting content to audiences through channels People Inc controls. Distribution is a one-directional problem: publish content outward, collect revenue inward. Every layer above solves distribution.

Interoperability is a different problem. It means external systems can query your intelligence, discover your inventory, and transact without a phone call. It means an Omnicom buyer agent can find D/Cipher's contextual signals via AAMP. It means a Horizon media plan can include People Inc inventory via MCP. It means the Brinkema-mandated Prebid integration creates a surface that other systems can build on.

The 70-to-20 journey required nine infrastructure bets. The next journey — from closed distribution to open protocol surface — requires one. But it's the one that the organizational instinct that built all nine layers has consistently declined to build.

Forbes is paying $866 million a year to not have solved the distribution problem. That's real failure. People Inc solved distribution and now faces a different question: can the engineering org that built nine layers of outbound infrastructure build one layer of inbound protocol?

The October deadlines will start to answer it.