Content & review analysis — turning stars and verbatim reviews into copy.

Episode 02 turns to the other dominant research surface on every Amazon detail page: customer reviews. Amazon does not ship reviews as one flat list — it ships three structured layers stacked on top of each other, and learning to read each layer separately is the entire point of this episode. The verbatim reviews are only the bottom of the stack. The two layers above them give you most of the answers without you having to read a single full review.

The three layers Amazon already does for you

On any detail page, clicking the star rating now scrolls you inline into the reviews section (it used to open a separate page — that is gone). From there the three layers are:

1. Overall stars + per-attribute scores.The top block. Beyond the headline 4.3-out-of-5, most categories now expose attribute-level scores Amazon defines per category — for a kitchen towel that might be value for money, softness, transparency, absorbency, packaging; for a shelf it is stability, ease of assembly, quality. Amazon has been refining this attribute set for three years and the choices are deliberate — they tell you which attributes the category cares about before you read a single review.
2. Topic clouds / keyword tiles. The row of clickable phrases Amazon auto-extracts from the review text — "easy to assemble", "looks beautiful", "stable", "nice packaging". The ordering is not strict frequency (Amazon does not publish the ranking logic), but the tiles surface the themes Amazon's own NLP considers representative for the ASIN. Click any tile and the reviews filter to just the ones discussing it.
3. Full verbatim reviews with filters.The classic list, with sort (most helpful, newest), filter by star count, by image/video reviews, by verified-purchase, and — critically — by variant.

Read top-down. The attribute scores tell you what the category measures. The topic cloud tells you which words shoppers actually use for it. The verbatim reviews tell you the why, the edge cases, and the alternative use cases. Skipping the top two layers and starting straight in the verbatim list is the most common mistake.

Read attribute scores against the category, not the SKU

A 4.6 on stability means nothing in isolation. It means a lot when the three highest-ranking competitors score 3.8, 4.0 and 4.1 on the same attribute. That delta tells you two things at once: the category has a stability problem, and your SKU genuinely solves it. That is a bullet, an A+ comparison row, and probably an image overlay. The same logic in reverse — a 3.6 on ease of assembly while competitors sit at 4.5 — tells you to remove every "tool-free assembly" claim from your copy until you fix the product.

The variant filter matters more than it looks

Older detail pages let you skip horizontally between variant reviews. That UI is gone. The replacement path is non-obvious: click on a star count in the rating breakdown to filter, then clear the star filter while leaving the variant filter active. That gives you the full review pool for one specific child SKU. This matters most when a parent groups loosely related products (a laundry basket and a drying rack incorrectly variated together is the canonical example) — without the variant filter you cannot tell which complaints belong to which SKU, and a review saying "breaks often" reads completely differently depending on which child it was written about.

The same screen also separates verified-purchase reviews, image-and-video-only reviews, and global (international) reviews from the local pool. Verified and local-only is the default cut for most analyses; the others are useful as a second pass.

The verbatim search trick: don't read, query

Reading every review of every competitor is not the workflow. The workflow is to take the recurring themes from the topic cloud and the attribute scores and search for them in the verbatim text. Type the word stem ("stable", "breaks", "smell", "loud", "leak") into the review search box. Amazon returns the count of reviews containing that stem, then lets you slice that subset by positive vs critical reviews. The ratio is the signal you act on:

• 30 of 34 mentions positive. The theme is a strength. Feature it prominently in bullets and A+. The few negative mentions almost always describe an edge case or a returns-handled story — useful context, not a reason to hide the claim.
• 5 of 14 mentions positive, 9 negative.The theme is contested. Do not make it a headline claim. If competitors show the same ratio, the entire category has a problem — and solving it on your SKU becomes the single most defensible differentiation angle in the category.
• Almost all mentions negative.Do not mention the attribute in copy at all. Fix the product, then revisit.

The example in the walkthrough: a brand's products showed eight reviews containing "breaks". Reading them revealed nine of fourteen "handle" mentions were actually positive, because the brand offers a 20-year guarantee and ships replacements immediately. The negative-sounding word was, statistically, a positive story — and the guarantee became a confident A+ module rather than a hidden footnote.

Image and video reviews are content briefs

Filter the reviews to "image and video reviews only" and a different kind of intelligence surfaces: how shoppers actually use the product. The classic example from the walkthrough — a tall, wide vase whose reviews are almost entirely photos of shoppers gifting it filled with bottles or sweets, not flowers. The implication is direct: do not write "perfect for a bouquet" in the bullets, because that is not what the category buys it for. The copy should reflect the gifting and decorative use the review images are documenting.

Image reviews also reveal packaging stories — a category where "every glass individually wrapped" keeps appearing in reviews is a category where packaging is a buying objection, and an A+ module showing the packaging is worth more than another product hero shot.

Seller feedback as a leaky review channel

A surprising amount of product-quality language ends up in seller feedback instead of product reviews. Shoppers do not always distinguish the two UIs. Pull up the seller-feedback page for a competitor and scan it the same way: most of it will be shipping and packaging commentary, but the product complaints that leak through are unfiltered and often more specific than what makes it into the moderated review list.

One category-aware caveat: unverified review skew

Verified-purchase reviews are the cleaner signal in most categories. In beauty and personal care, however, a large share of purchase volume happens offline in drugstore chains — but a disproportionate share of complaints about those offline purchases gets written on Amazon as an unverified review. The result: unverified reviews in beauty skew systematically negative versus the verified pool. When analysing a beauty ASIN, look at verified and unverified separately rather than treating the merged number as the truth.

What this episode hands off

Episode 02's output extends the same per-category sheet Episode 01 started. Each row is a theme; each theme now carries an attribute score, a topic-cloud presence flag, a positive/negative verbatim ratio, and one of the four downstream actions from Episode 01 (bullet, image/icon, A+ module, or product change). With Q&A themes and review themes both captured, two of the four input streams for the Episode 04 foundation document are done. Episode 03 turns to the third: researching what Amazon itself requires for the category — the style guide, the attribute pipeline, the title pattern, and the constraints every text field downstream will have to respect.