# AI search as an eCommerce channel — measurement sheet

Companion to **AI search is becoming a new eCommerce sales channel**.

Use this to **instrument** AI-assisted discovery. Do not treat it as a ranking playbook. GEO is not a guaranteed ranking technique.

Industry benchmarks below are **Adobe Analytics** as reported by [Digital Commerce 360 (19 Aug 2026)](https://www.digitalcommerce360.com/2026/08/19/adobe-ai-referral-traffic-data-july-2026/) on **1T+ visits** to U.S. retail sites. They are not FactualMinds measurements and not a KPI you inherit.

## Current vs emerging (do not collapse)

| Layer | What you can measure now (2026) | What is still emerging |
| ----- | -------------------------------- | ---------------------- |
| AI referral | Sessions whose referrer / UTMs are ChatGPT, Perplexity, Gemini, Copilot, Google AI Overviews click-through | Fully agent-completed checkout at scale |
| Conversational shopping | On-site copilot or chat that calls catalog tools | Third-party agents that buy without a human in the loop |
| Merchant visibility | Whether PDPs and homepages are machine-readable; whether your SKUs appear in sampled answers | A stable "AI rank" you can optimize like page-1 SEO |
| Recommendations | Session + stock + policy via your recs / search | External agents substituting your merchandising rules |

## Adobe July 2026 — cite, do not copy as a target

| Signal | Value | How to use it |
| ------ | ----- | ------------- |
| AI-referral traffic to U.S. retail | **+62% YoY** (July 2026) | Direction of a **channel**, not your store's growth rate |
| vs October 2024 | **+1,219%** | Long baseline from when generative platforms widened; not a 2025–2026 sprint target |
| Conversion vs non-AI traffic | **60% higher** | Mix/intent bias is likely; do not promise this lift |
| Revenue per visit vs non-AI | **53% more** | Same caveat |
| Engagement vs non-AI (Adobe, same article) | **+14%** engage; **+59%** time on site; **−33%** bounce; **+28%** add-to-cart | Useful for hypothesizing quality of referred sessions — still not yours |
| Homepage machine-readability (expanded cohort, July) | **39%** of homepages **not** machine-readable (**61%** visible) | Audit **your** homepage + PDPs; do not average in Adobe's cohort |
| Apparel LLM visibility | **76%** (24% not machine-readable) | Category context only. Electronics 70%, cosmetics 68%, sporting goods 67%, furniture/home 64%, general merchandise 63%, grocery 59% |

Adobe also noted an April figure (~25% of homepage content not optimized) on a **narrower** set. July's **39%** is the **expanded** cohort. Do not mix the two denominators.

## First-party instrumentation (your store)

Fill this for **one** month. Empty cells are honest.

| Metric | Source | Last month | Notes |
| ------ | ------ | ---------- | ----- |
| Sessions with AI referrer / documented UTM | Analytics | | ChatGPT, Perplexity, Gemini, Copilot, other — list hosts |
| Revenue from those sessions | Analytics + orders | | Do not blend with on-site copilot |
| Conversion rate vs rest of site | Analytics | | Expect mix bias |
| Revenue per visit vs rest of site | Analytics | | Same |
| Share of landing pages that are PDPs vs homepage | Analytics | | AI answers often deep-link |
| Sampled answer inclusion (10 queries × 2 engines) | Manual / eval | | Pass/fail, not a rank |
| Homepage machine-readable (yes/no) | Your audit | | JSON-LD / SSR facts |
| One category: % SKUs with required attributes | PIM | | Agent contract, not SEO score |

## When NOT to call it a sales channel yet

- You cannot separate AI referrers from generic "direct"
- You have fewer than a few dozen attributed sessions (noise)
- Leadership wants to replace SEO reporting with "GEO rank"
- You are attributing on-site copilot chats as AI-referral revenue

## Reproduce

Copy this file. Fill **your** analytics. Cite Adobe only as industry context. Platform cost/latency (Gateway ~180→95 ms, ~$791/mo at 50K sessions) is **not** a channel KPI — keep it on the AgentCore TCO sheet.