Overview

Amazon has upgraded the Manage Your Experiments console, allowing brand‑registered sellers to run A/B tests on product bullet points and full‑length descriptions for the first time. The same update adds a “run to significance” timing option that ends a test automatically once the data reaches a 66 % confidence level, eliminating guesswork about test length. Sellers who rely on data‑driven listing tweaks can now experiment on two of the most visible content blocks and let Amazon’s engine determine when results are reliable.

Key Points

• Bullet‑point testing launched — Sellers can now compare two versions of the five‑point summary, such as swapping “water‑resistant” for “rain‑proof” to see which drives more clicks.
• Description‑level testing added — Full product narratives can be split‑tested, for example a technical spec‑heavy copy versus a benefits‑focused story.
• Run‑to‑significance timing — Experiments automatically stop when one variant hits at least 66 % statistical confidence, preventing premature or overly long tests.
• Brand‑registry requirement — Only sellers with an Amazon Brand Registry account can access these new experiment types.
• No code changes needed — All variations are entered through the existing Manage Your Experiments UI, keeping the workflow familiar.

How Manage Your Experiments Works

1. Create a new experiment — In the console, click “Add experiment,” choose the ASIN, and select either “Bullet points” or “Product description” as the test type. Example: A seller selects ASIN B07XYZ123 and opts to test bullet points.
2. Define control and variant content – Upload the current listing as the control and craft a distinct version for the variant, such as reordering the third and fourth bullet or adding a new keyword phrase. Example: Control bullet 3 reads “Durable stainless steel”; variant bullet 3 reads “Premium stainless steel construction.”
3. Select a duration mode – Choose “Run to significance” or set a fixed date range. When “Run to significance” is chosen, Amazon monitors traffic and sales, ending the test once one version reaches ≥66 % confidence.