Plan for our first store test
What Tenuvio's first test covers, how its products are chosen, and what it can and can't show until a store agrees to a test with its own data.
Published Updated
Protocol v0.1Kind: observationalStatus: planned, not yet run
The first test is a small pilot. Its job is to try the protocol on real product pages and find where the method breaks before any result is published. It uses public pages only, so it can show what assistants say about products but not why.
Design
| Catalog | One store, 10 products with visible prices and stock, most with a GTIN or MPN, at least one with variants and one sold elsewhere. |
|---|---|
| Prompts | 3 per product, one per prompt family: 30 in all. |
| Repeats | 2 runs per prompt per assistant per run day (the full protocol uses 3), on 2 run days at least 7 days apart. |
| Assistants | ChatGPT, Microsoft Copilot, Gemini, Google AI Overviews and Google AI Mode, asked signed out through a third-party service. |
| Country | The US, where most of these shopping features are offered. |
| Store named? | Only if it agrees. Otherwise described by category, country and size. |
Every prompt, capture and run record is published with the result, including empty and failed runs. If the store hasn’t agreed, its name and web addresses are replaced by its description. Other shops that the answers name or link to are shown by type of site.
What it can’t show
Whether fixing a feed or a page changes the answers. That needs an authorized test, where Tenuvio can see the merchant’s feed and note when it changes. In an authorized test the merchant decides whether its name appears and sees the results before publication. It can correct facts about its own catalog but can’t change a score.
Changes to this page
- : Corrected. Ten prompts name the store, and the page said every prompt is published. It now says how an unnamed store and other shops are kept out.