A new framework from Search Engine Land treats AI brand recommendation as a six-stage funnel, and most retail sites fail before reaching stage two. The piece, written by contributor Benu Aggarwal, argues that AI engines and shopping agents now filter, rank and transact with brands before a shopper ever opens a traditional results page. That reordering already carries measurable revenue: Salesforce data put agent-influenced purchases at a fifth of all global online orders during Cyber Week 2025, worth an estimated $67 billion.
Aggarwal frames the funnel as sequential: get found, get understood, get retrieved, get trusted, get chosen, then enable a transaction that never touches the brand’s own website. Each stage has a technical gate attached to it. Crawl access for Google, OpenAI, Anthropic and Bing bots, clean XML sitemaps, working robots.txt rules and server-side rendering handle discovery. Structured data, a coherent entity graph and consistent naming across the web determine whether an AI system understands what the brand actually sells.
The retrieval stage is where GEO, generative engine optimization, the practice of optimizing content for citation inside AI-generated answers, becomes concrete. Aggarwal recommends front-loading the essential takeaway and the key numbers into the opening sentence of each section, before a model’s context window forces a truncation. That is a real break from classic on-page SEO, where comprehensiveness and length could substitute for concision.
Trust is where AEO, answer engine optimization, the discipline of optimizing for direct-answer surfaces, intersects with reputation management well beyond the brand’s own site. The article argues that AI systems cross-check review sentiment, pricing consistency and location data across the web, and that any conflict between those sources lowers a model’s confidence in citing the brand at all. Google’s E-E-A-T criteria, per the source, remain among the strongest signals for whether a brand gets referenced.
The final stage, agentic commerce, carries the clearest business case. Adobe data cited in the piece shows traffic from AI referrals to American retail sites grew 4,700 percent year over year through mid-2025. Aggarwal ties that growth to a stack of emerging protocols: Google’s Universal Commerce Protocol for chat-based bookings, OpenAI and Stripe’s Agentic Commerce Protocol for pushing inventory into AI systems, and the Agent Payments Protocol for letting an agent complete checkout, all running on top of the Model Context Protocol.
The piece does not address what happens to attribution when an agent completes a purchase without a click-through. Traffic can fall even as AI-influenced revenue rises. Search Engine Land’s framework offers no method for separating the two inside analytics tools still built around sessions and pageviews. That measurement gap is a real signal, not a solved problem.
Search teams should treat the next ninety days as an audit, not a rebuild. Confirm that Google’s, OpenAI’s and Anthropic’s crawlers can reach every product page unblocked, publish an llms.txt file, and run a query fan-out test to see which competitor gets cited when the brand does not. Crawl access and structured data are cheap fixes. A fractured entity graph across reviews, maps and directories takes months to repair, and it is the gate most sites will fail first.
Search Engine Land published Benu Aggarwal’s six-stage AI decision layer framework on July 9, 2026.
