Something happened in 2024 that nobody fully reckoned with. AI stopped being a tool that businesses used to analyze customers and became a tool that customers used to bypass businesses. The intermediary layer arrived — and it changed the architecture of commerce in ways we're only beginning to understand.
Go back to our MacBook Air searcher from Part One. When they typed that query into Google, the first thing they saw wasn't a list of blue links. It was an AI Overview — a synthesized answer generated by Google's Gemini model, drawing from multiple sources, presented with the quiet authority of a finished conclusion. The user got an answer without clicking a single link. Without visiting apple.com. Without seeing a single ad for a MacBook accessory or a competing laptop.
That AI Overview didn't just summarize information. It intercepted a decision. It absorbed the user's intent, processed it through a language model trained on the open web, and returned a verdict — all before the user reached any business's owned property. From the user's perspective, it felt helpful. From the business's perspective, it was invisible.
AI didn't disrupt the funnel. It built a bypass around it — and charged no toll on the way through.
The New Stack
To understand what's happening, you need to see how the technology stack has changed. Before 2024, the path from question to purchase had a predictable structure: search → results → click → browse → consider → buy. Every step happened on someone's platform, and that platform captured data about the interaction. The stack was leaky, as we saw in Part Two, but at least the layers were visible.
Now there's a new layer in the middle. And it's opaque.
The critical insight is the opacity barrier. The AI layer sits between the user and everything else, but it doesn't report downstream. Google's AI Overview doesn't tell apple.com that a user read a comparison and was almost convinced. ChatGPT doesn't notify Samsung that it recommended a Galaxy over an iPhone. Perplexity doesn't share its shopping recommendations with anyone's CRM.
The AI layer absorbs signal and doesn't emit it. It's a black hole in the decision stack.
Let's make this viscerally concrete. Here's what happens inside the AI layer when our MacBook searcher's query arrives — and what each party actually sees.
// Google AI Overview — Internal // Processing Pipeline (simplified) { "query": "m4 vs m5 macbook air", "intent_classification": { "type": "comparison", "category": "electronics", "commercial_intent": 0.82, "urgency_signal": "medium" }, "sources_consulted": [ "apple.com/macbook-air", "9to5mac.com/m5-review", "youtube.com/watch?v=...", "reddit.com/r/macbookair", "engadget.com/comparison" ], "generated_overview": "Yes, the M5 offers doubled base storage (512GB vs 256GB), faster Wi-Fi 7...", "confidence": 0.91, "ad_auction_triggered": true, "ads_served": 3, "user_clicked_through": false, "time_on_overview": 14.2, //sec "outcome": "satisfied_no_click" } // Google captured EVERYTHING: // intent, confidence, sources, // dwell time, click behavior
// What Apple.com receives // about this interaction: ∅ Nothing. Absolutely nothing. // No referral event. // No impression pixel. // No attribution webhook. // No "your product was // mentioned" notification. // No API. No feed. No log. // // Apple's content (apple.com // /macbook-air) was used as a // SOURCE for the AI Overview. // Apple received 0 visits, // 0 data, 0 signal in return.
The Players in the Middle
This isn't a single company doing this. It's an entire new category of intermediary — and each player has a different relationship to the businesses whose products they're recommending.
Answers queries directly in search results. Leverages the world's largest search dataset. Reduces click-through to source websites. The business whose content trained the answer may never receive a visit.
Users directly ask "should I buy X or Y?" and receive personalized-feeling recommendations. No ads, no sponsorships (yet). No attribution. The conversation is private and untrackable by any business.
Combines search and synthesis with source citations. Has launched shopping features. Sits at the intersection of discovery and transaction — potentially the most disruptive position in the stack.
AI-powered feed algorithms already shape what users see. Next step: AI assistants embedded in social apps that answer product questions using engagement data. Influence masquerading as discovery.
To understand the scale of invisibility, let's look at the actual data that flows — and doesn't flow — when a user asks ChatGPT for a product recommendation.
// ─── User asks ChatGPT: "Should I get the M4 or M5 MacBook Air?" ─── // ChatGPT Internal Processing (approximate): { "conversation_id": "conv_8f7a6b5c4d3e2f1a", "user_id": "user_anon_xyz789", "model": "gpt-4o", "message": "Should I get the M4 or M5 MacBook Air?", "context_from_conversation": [ "User previously mentioned being laid off", "User asked about budget laptops last week", "User is a software developer" ], "response_generated": "For most users, the M4 MacBook Air at $1,099 offers excellent value. The M5's main advantages are doubled base storage and Wi-Fi 7 — but if budget is a concern, the M4 gives you 90% of the performance at $200 less. Given your situation, I'd lean toward the M4.", "products_mentioned": [ { "product": "MacBook Air M4", "sentiment": "recommended", "brand": "Apple" }, { "product": "MacBook Air M5", "sentiment": "acknowledged", "brand": "Apple" } ], "recommendation_direction": "steered_to_lower_price", "revenue_impact_estimate": "-$200 (M5 → M4 downgrade)", "tokens_used": 847 } // ─── Data sent to Apple about this interaction: ─── null // ─── Data sent to ANY business about this interaction: ─── null // ─── Webhook or API available for businesses to query: ─── null // ChatGPT just influenced a $200 revenue delta for Apple. // Apple has NO IDEA this conversation happened. // No notification. No attribution. No feedback loop. // This is repeated millions of times per day, across every product category.
Now contrast that with what should exist. If the AI layer were part of a connected Signal Chain rather than a black hole, it could emit structured attribution events — without exposing the user's private conversation.
// ─── AI_Recommendation_Attribution_Event ─── // A hypothetical webhook that SHOULD be emitted when AI // makes a product recommendation. This does NOT exist today. POST https://api.apple.com/webhooks/ai-attribution Content-Type: application/json X-AI-Provider: openai X-Attribution-Version: 1.0 { "event_type": "ai_product_recommendation", "timestamp": "2026-04-02T11:28:33Z", "ai_provider": "openai_chatgpt", "interaction_id": "attr_anon_9f8e7d6c", // anonymized, no user PII "products_discussed": [ { "product_id": "MBA-M5-15-512", "brand": "Apple", "mentioned": true, "recommended": false, "reason_category": "price_sensitivity" }, { "product_id": "MBA-M4-15-256", "brand": "Apple", "mentioned": true, "recommended": true, "reason_category": "value_proposition" } ], "decision_context": { "comparison_type": "within_brand", "primary_factor": "price", "user_segment": "budget_conscious", // anonymized segment, not PII "confidence": 0.78 }, "attribution": { "revenue_direction": "downgrade", "estimated_delta": -200.00, "currency": "USD" }, "transparency": { "sources_cited": ["apple.com", "9to5mac.com"], "paid_placement": false, "disclosure_shown_to_user": true } } // THIS WEBHOOK DOES NOT EXIST TODAY. // No AI provider emits structured attribution events. // No standard protocol exists for AI recommendation reporting. // The entire concept of "AI attribution" is unbuilt infrastructure.
The Zero-Click Economy
The most measurable consequence of the AI intermediary is the rise of what search analysts call "zero-click searches" — queries where the user gets their answer directly on the search results page and never clicks through to any website. AI Overviews have accelerated this trend dramatically.
Think about what this means for businesses. Three-quarters of the time someone searches for your product, your category, your brand — they may get an answer and leave without ever seeing your website, your pricing page, your value proposition as you frame it. The AI layer is not just intercepting traffic. It's intercepting narrative control.
To understand the economics of zero-click, you need to see the data packet. Here's exactly what Google logs for a zero-click AI Overview interaction — and what it sends to the publisher whose content powered the answer.
// ─── Google's Internal Log for this AI Overview Impression ─── { "event": "ai_overview_impression", "query": "m4 vs m5 macbook air", "user_id": "gaia_12345", // Google account ID "overview_rendered": true, "overview_expanded": true, // user read the full overview "time_on_overview_ms": 14200, "sources_shown": [ { "url": "apple.com/macbook-air", "clicked": false }, { "url": "9to5mac.com/m5-review", "clicked": false }, { "url": "engadget.com/m5-vs-m4", "clicked": false } ], "organic_results_clicked": 0, // ← zero clicks to any website "ads_shown": 3, "ads_clicked": 0, "next_action": "new_search", // user refined query or left "ad_revenue_generated": 0.00, // no clicks = no CPC revenue "user_satisfied": true // inferred from no follow-up query } // ─── What Google sends to 9to5mac.com about this interaction: ─── // Via Google Search Console (delayed 2-3 days): { "query": "m4 vs m5 macbook air", "impressions": 1, // yes, counted as an "impression" "clicks": 0, "ctr": 0.0, "position": 3.2, // That's it. The publisher sees: "someone searched, we appeared, nobody clicked." // They do NOT see: "Your content was synthesized into an AI Overview // that answered the user's question, making a click to your site unnecessary." // The publisher's content was CONSUMED. The publisher was not VISITED. }
The Before and After
Let's make the architectural shift concrete. Here's the same user journey — our MacBook Air searcher — mapped across the old stack and the new stack.
Both scenarios are devastating in different ways. In the first, the business loses the user entirely — the journey ends before it begins. In the second, the business gets the sale but loses control of which product was recommended and why. In neither case does the business have any visibility into the AI's role in the outcome.
The Paradox of the Helpful Machine
Here's the uncomfortable paradox: the AI intermediary is genuinely helpful to users. It saves time. It synthesizes complexity. It reduces the seven-tab spiral to a single, coherent answer. For the user, this is progress.
But for the ecosystem, it's an extraction event. The AI consumes content, data, reviews, and specifications created by businesses and publishers, synthesizes them into an answer, and delivers that answer in a context where the original creators receive no traffic, no attribution, and no commercial benefit. The user is better served. The system that served them is being hollowed out.
This is not a problem that gets solved by blocking AI crawlers or suing language model companies. The genie is out of the bottle. Users now expect AI-mediated answers. The question is not whether AI will sit in the middle of the commerce stack — it already does. The question is whether the rest of the stack will adapt to work with it, around it, or despite it.
The AI intermediary is the most powerful uninvited guest in the history of commerce — and it's already rearranging the furniture.
In the next chapter, we'll propose what adaptation looks like. Not a single product or platform, but an architectural pattern — a way to reconnect the signal that's been fragmenting for two decades and finally build the decision infrastructure that users and businesses both need. We call it The Signal Chain.