Summary

Genesis 03 examines what ChatGPT and Gemini return after recommending ecommerce brands: brand-owned product pages, third-party destinations, or no usable product-page path.

Key takeaways

Download PDF
Cite this
@techreport{peace2026genesis03,
  title = {AI Shopping Visibility and the Buy Path Problem},
  author = {Peace, Kalan},
  institution = {Caeliai},
  year = {2026},
  url = {https://research.caeliai.com/research/genesis-03}
}
Fig. 1

Only 36.8% of observations had a path to the brand's own product page.

Owns the path36.8%
Leaks the path17.6%
No path45.5%

Share of 380 brand observations by returned destination. Genesis 03, May 6, 2026. Percentages as published in the report.

Source: Genesis 03 (this page).

Back to research

Genesis 03 · Technical report

AI shopping visibility and the buy path problem.

Kalan Peace · Caeliai · Published May 6, 2026

N = 380 brand observations · 66 public conversations · ChatGPT + Gemini

Published May 6, 2026. Also published as “AI Shopping Visibility: The Buy Path Problem”.

Genesis 03 examines what ChatGPT and Gemini return after recommending ecommerce brands.

This field study records whether the returned destination is a brand-owned product page, a third-party merchant, or no usable product-page path.

What the study measures.

The study treats AI recommendation as a routing system. Once a model names a brand, the recorded outcome is whether the returned destination is a brand-owned product page, a retailer or marketplace, or no usable product-page path.

Method
  • 66 public ChatGPT and Gemini shopping conversations.
  • Each conversation is linked to a Markdown record for an evidence trail.
  • Temperature-zero extraction for brand, merchant, and routing outcome.
  • Deterministic aggregation into Owns the path, Leaks the path, No path, and site-health layers.
Outcome taxonomy

Owns the path records a brand-owned product page. Leaks the path records a retailer or marketplace destination. No path records a brand name without a usable product-page path.

Fig. 03:1 · Routing outcome distribution N = 380 brand observations · May 2026
Owns the path · 36.8% Leaks the path · 17.6% No path · 45.5%

Source: Genesis 03 source matrix · 380 brand observations from 66 conversations. Percentages may not total 100% because of rounding. A destination does not establish a click or purchase. This historical cohort is separate from later share archives.

36.8% Owns the path
17.6% Leaks the path
45.5% No path
Core claim
A recommendation can name a brand without returning a usable product-page path.

For this study, visibility and path quality are separate observations: whether a brand appears, and where the returned link resolves.

Site health layer

Genesis 03 also scans canonical brand domains as a separate diagnostic layer. The paper finds site health is weakly correlated with AI routing, so it should be treated as a related but distinct signal.

257 Domains scanned
231 Accessible sites
41/100 Median health score
Source trail

The full PDF includes the 66-link source matrix so each result can be traced back to a public AI conversation and the extraction evidence used for scoring.

Caeliai · measured May 2026 · contact@caeliai.com