On February 16, 2026, OpenAI launched Buy it in ChatGPT — instant checkout, powered by Stripe, inside the conversation. Etsy first, then a million Shopify merchants coming soon. The pitch: ask ChatGPT for running shoes under $100, see products ranked by relevance (not ads), buy without leaving the chat. Agentic commerce arrives.
By June, native checkout inside ChatGPT was gone. Roughly twelve Shopify merchants had gone live out of millions. Users researched products in ChatGPT but wouldn't complete purchases there. OpenAI rerouted transactions to retailer apps — Instacart, Target, Expedia. The Agentic Commerce Protocol (the infrastructure layer, co-developed with Stripe) survived. The product didn't.
This is the most important data point in publisher commerce right now, and it has nothing to do with publishers.
What Broke
The checkout failed because trust didn't transfer. People will ask an AI for recommendations — 35% of Gen Z, 32% of Millennials search products via ChatGPT monthly — but they won't hand their credit card to a conversation. They want the retailer's checkout page, the retailer's return policy, the retailer's receipt. Discovery moved. Transaction didn't.
This creates a specific window for publishers. The feared scenario — AI agents skip publishers entirely, negotiate directly with retailers, cut out the affiliate link — just got a real-world stress test. OpenAI tried to build that shortcut. It didn't work. The $13.81 billion US affiliate channel is not dead. Not yet.
But the discovery layer is already somewhere else.
Where Discovery Went
Adobe measured it: AI traffic to retail sites grew 269% year-over-year through early 2026. Shopify reported AI-attributed orders grew 11x between January 2025 and January 2026. Users aren't browsing publisher review articles to find products anymore. They're asking LLMs.
And here's what matters for People Inc: when those LLMs answer shopping questions, 82% to 85% of their product recommendations come from external, third-party sources — not the brand's own website. Publisher reviews. Shopping guides. Recipe roundups. Exactly the kind of content People Inc produces across 40 brands.
More than a quarter of OpenAI's content partnerships since 2021 have been with publishers that operate scaled affiliate commerce. This is not incidental. Product reviews and recommendation content are what LLMs need to sound credible when someone asks what to buy.
The question is whether being cited pays.
Two Publishers, One Answer
Future plc built the machine. Their proprietary commerce platform Hawk, launched in 2013, ingests feeds from 45,000 retailers, inserts affiliate links automatically, tracks real-time inventory via direct API. At peak, Hawk drove $960 million in annual affiliate sales. It was the envy of every publisher without one.
In H1 of fiscal year 2026, Hawk's organic revenue declined 24% — from £44.5 million to £32.3 million. Future's overall profit fell 67%. S&P issued a negative outlook. Shares dropped 30%. The automated feed model that made Hawk dominant was built for Google-driven product reviews — precisely the category where AI Overviews hit hardest. When the search referral disappeared, the automation didn't matter.
People Inc has no Hawk. Commerce is manual. Commerce Producers insert affiliate links by hand in the CMS. Standard networks — Impact, CJ, Skimlinks. No proprietary feed aggregation, no real-time inventory, no retailer APIs.
People Inc's affiliate commerce revenue grew 26% year-over-year in Q1 2026.
The divergence is explained by category mix, not technology. Future's strength is consumer electronics and tech reviews — the highest-value affiliate categories, and the ones most exposed to AI-generated product comparisons. People Inc's commerce content lives in food, home decor, lifestyle, and beauty — categories where editorial authority and taste still drive purchase decisions, and where AI Overviews have less traction. Brand authority was a better moat than feed automation.
But this is a timing advantage, not a structural one.
When Category Mix Stops Being Enough
McKinsey projects $1 trillion in US retail revenue from agentic commerce by 2030. Shopify activated Agentic Storefronts by default for every store in late March 2026. Google's Universal Commerce Protocol launched in January 2026 with Walmart, Target, and 20+ partners. The infrastructure for AI-mediated shopping is being laid at every layer, even if the consumer behavior hasn't fully followed.
The category protection that currently benefits People Inc — lifestyle, food, home — is temporary. When someone asks an LLM "what stand mixer should I buy for under $400," the answer will come from the same content pipeline that currently powers People Inc's affiliate revenue. Except the user won't click through a BHG article to get to the Amazon listing. The LLM will cite the article, absorb the recommendation, and route the purchase through whatever protocol the retailer supports.
Future estimated a 40% decline in affiliate gross contribution under an agentic scenario. People Inc hasn't published an equivalent estimate. They should.
A Machine You Don't Have to Build
Shopsense AI launched its Shoppable Intelligence Model (SIM) on June 2, 2026 — a multimodal AI that reads articles, images, and video, identifies products, and surfaces matching product listing ads. It connects to 1 billion+ SKUs across 500,000 brands and 1,000+ retailers. SIM claims 25-50% higher retrieval accuracy than OpenAI's CLIP and Google's SigLIP2.
The CEO is Bryan Quinn, who founded Amazon Publisher Services and built advertising businesses to $1B+ at Amazon. He brought Shopsense into the APS Connections Marketplace in May 2026 — no-code activation for any APS publisher. Already deployed at Disney, Fox, Paramount, Tubi, and Bell Media.
People Inc was the first digital publisher to integrate Amazon Publisher Cloud. The APS relationship already exists. Activating Shopsense through APS would give People Inc automated content-to-commerce at a level comparable to Hawk without building a proprietary platform. No feed aggregation to maintain, no retailer API integrations to manage, no 13-year head start required.
This is the leapfrog. Not building Hawk in 2026, but skipping it entirely and landing on the commerce layer that works when discovery happens inside an LLM instead of inside Google.
What Gets Built
The commerce architecture question for People Inc isn't "should we build Hawk?" It's a question about structured data.
If 82-85% of LLM product recommendations come from publisher content, then the publisher's commerce future depends on whether their content is citable. Not just crawlable — citable. Entity-level structured data that an LLM can parse, attribute, and link back to. Product mentions with schema markup. Recipe content with ingredient-level product associations. Home decor articles where every recommended item has a persistent identifier that a commerce agent can resolve.
This is where the commerce question meets the search question I wrote about yesterday. The same structured data strategy that makes content visible to AI Overviews is what makes it citable for commerce recommendations. The same Graphene-Selene knowledge graph that powers D/Cipher's content taxonomy could power product-entity relationships. The same JSON-LD that the Healthy Domains team deploys for SEO becomes the surface that commerce agents read.
Princeton found that expert quotes increase AI citation by 41%, statistics by 30%, and source citations by 30%. Those same signals — authoritative content with specific, structured claims — are what make a product recommendation credible to an LLM. People Inc's editorial authority, properly structured, is the commerce asset.
The work is not building a commerce platform. It's extending the content infrastructure that already exists — Graphene-Selene, the Healthy Domains team's structured data pipeline, D/Cipher's entity taxonomy — to make products first-class entities alongside recipes, articles, and brands. When an LLM cites a Food & Wine knife review to recommend a purchase, the structured data should make the attribution, the product, and the commerce path machine-readable in one pass.
ChatGPT abandoned its cart. The checkout failed because trust didn't transfer to the conversation. But the discovery did transfer. And the content that powers that discovery is still publisher content — for now. The question for People Inc isn't whether to build a commerce machine. It's whether to make the content machine commerce-readable before someone else's machine reads it without paying.