AI-Assisted Commerce: Read the Evidence, Improve Product Information

CH.01

What the Published Figures Measure

Growing AI use can support a test. It does not establish that human shoppers or product pages have been replaced.

Shopify reports more than 8x year-over-year AI-referred sessions and nearly 13x orders in Q1 2026. These are referrals to Shopify storefronts, not a count of autonomous agent purchases. The published post does not provide a complete reproducible cohort or raw dataset.

Salesforce’s January 8, 2026 release estimates AI and agents influenced 20% of retail sales, US$262 billion, during November 1–December 31, 2025 through recommendations and customer engagement. McKinsey’s October 17, 2025 report projects US$3–5 trillion of global B2C goods commerce could be orchestrated by 2030, using adoption and merchant-readiness assumptions. A forecast is not observed orders or your future revenue.

CH.02

Different Shopping Tasks, Different Access Paths

AI may help a person compare products, send a visitor to a page or execute an authorized task. Verify the actual product and mode before designing an integration.

Product pages, feeds and APIs can complement each other. McKinsey describes agents interacting with websites as well as APIs. Do not assume every assistant ignores visual pages or that every merchant needs a checkout API.

Shopify’s nearly 50% higher AI-referral conversion applies to Q1 2026 sessions beginning on product detail pages, compared with organic search. It does not describe all visitors, all merchants or autonomous checkout; the page itself remains part of the measured journey.

CH.03

Fix Observable Information Problems

Start with a mismatch you can reproduce, rather than assume an invisible agent rejection caused a revenue change.

Keep important specifications in readable text as well as images. Match units, variants, price and stock across the product page and any feed. Accurate data helps comparison, but completeness is not a published universal ranking weight or a guarantee of selection.

A missing referral does not reveal why an assistant omitted a product. Check access, eligibility, query fit and measurement gaps, and record unknown causes.

CH.04

Compare Conversion with the Right Denominator

A channel comparison must define the visit, completed action, time window and attribution model.

Kaiser and Schulze’s Marketing Science study uses 973 websites. Its main six-month model covers February 2–August 2, 2025 and compares website sessions that result in transactions. Affiliate conversion odds are estimated about 86% higher than organic ChatGPT-referral odds after model adjustments. This is not AI converting 86% below affiliate, a 14-of-100 index or a data-readiness experiment.

Last-click channel labels miss earlier discovery influence. This cohort and Shopify’s Q1 2026 product-page cohort differ in period, platforms, filters and method. Neither establishes your conversion rate or proves that product-data gaps caused the difference. On your site, compare transactions per eligible session alongside sample counts, revenue, margin and attribution limits.

CH.05

A Product-Information Worklist

Use the following checks where the channel and merchant integration actually apply.

CH.06

Choose the Next Test from Your Own Evidence

A market forecast does not dictate a fixed three-year spending plan.

Begin with your actual referral volume, buyer questions and product-information failures. Select a small test with a named owner, cost limit and measurable outcome. These stages are a suggested workflow, not a deadline or a claim about lost revenue.

Retain the human purchase experience and accurate machine-readable information together. Broader integration depends on actual demand, platform support, authorization and operating capacity, not a forecast alone.