Extract Amazon customer reviews into CSV

Overview

Review analysis needs a dataset that preserves text and context. This demo attempts collection across three Amazon products and exports the reviews accessible in a signed-out session. The actual recorded yield is smaller than the target described on YouTube.

What happens in the video

The agent checks three Amazon products, corrects product-identifier mismatches, and encounters sign-in requirements on review pages. Its final message reports 37 review rows collected from visible product-page reviews, and a CSV is opened on screen. The published description mentions approximately 300 rows, but the recording reports 37.

Before you start

• Product URLs and a field schema.
• A defined public-review collection scope.

Step-by-step guide

Practical steps for adapting the recorded task. The video summary above describes the demonstrated result; these checks guide your own run.

1. 1

   Identify the products

   Check each product identifier against the loaded product page. The recording finds mismatched input identifiers and switches to pages matching the intended products. Keep the actual product URL attached to the extracted reviews.

2. 2

   Extract review fields

   Collect ratings, titles, dates, text, and available metadata from accessible review content. The recording encounters sign-in requirements on full review pages. Record that limitation and use only the reviews actually available in the session.

3. 3

   Validate and export

   Use one review per row, check duplicates, and preserve CSV punctuation. The final recorded message reports 37 rows, not approximately 300. Check the exported count and record access failures alongside the dataset.

Try this prompt

Results

• A CSV containing 37 review rows according to the final recorded message.
• A run log recording sign-in requirements on the attempted full-review pages.

Questions about this workflow