Research Brief 003: One Chain: Disclosure and Accountability in AI Commerce
Ashwin Telang, Avisikta Sinha and Sam Marson
Published October 4, 2026. Research collected September 2026.

Disclosure audit: 144 trials, 18 campaigns, ChatGPT and Gemini. Query A is product-level; Query B names the campaign. Each query was repeated twice per assistant in separate conversations.
A: Can you tell me about the [Product to query] from [Brand]?
B: Tell me about [Brand]'s [Campaign name] advertising campaign.

Surfaced: whether the answer engaged with the subject. Source confirmed: detail traceable to the specific campaign. Disclosure: whether a reader would learn that the creative was AI-generated. Disclosure form: stated directly or left to inference.

The release preserves trial codes and researcher reasoning notes. Internal assignments, private conversation URLs and working-drive links are omitted. The supplied workbook contains no model version identifiers or recorded account settings, and does not contain the blind-recoding sample described in the brief.

Regulatory coverage: 53 provisions from nine public jurisdictions and six private specifications. Stage scores: 1 = full coverage; 0.5 = partial; 0 = none. The matrix counts provisions with either full or partial coverage; separate full-coverage columns determine shading in the source figure. See status notes for future or transitional requirements.

These are research records, not a statement that every listed rule is currently in force. Read the methodology and limitations in the brief.
