Can a saved Chrome Gemini Skill compare three product pages and return details worth trusting? I tested one reusable prompt with three public Logitech pages, then checked each name, price, feature, and source link against the originals.
The useful part of this test is verification. A neat AI table is a starting point; it is not proof that every detail is right. Watch the lesson, then use the checks below when you try a similar workflow.
How the Chrome Gemini Skill works
A Skill saves instructions for Ask Gemini in Chrome. I can choose the saved instruction from the slash menu instead of typing the same request again. I still choose which tabs to share on each run.
For this test, I shared public product pages for Lift, MX Anywhere 3S, and MX Master 4. The recorded run used Flash-Lite, without Auto Browse or an added paid feature.
The saved prompt: Compare only the product tabs I share with you. Make a table with product name, price shown on its page, two stated features, and a source link for each product. Write “Not stated” for anything the pages do not say. Do not use Auto Browse, visit other sites, or make purchases.
The recorded table gave Lift at $79.99 with a 57-degree vertical angle and soft rubber grip; MX Anywhere 3S at $89.99 with 8K DPI tracking and quiet clicks; and MX Master 4 at $119.99 with haptic feedback and Actions Ring shortcuts. These are the page prices observed on October 5, 2026, not current offers.
How I verified the comparison
I checked every product name and displayed price against its original page, then checked the two stated features and matching source reference in each row. This one run matched three names, three recorded prices, six features, and three source links. No requested field was missing.
The saved prompt worked as a reusable shortcut in this test. It saved typing and produced a useful first draft, but the source check made the comparison trustworthy. Afterward, I deleted the temporary Skill and confirmed it was absent from the slash menu. Product pages change, so verify the next result before relying on it.