The exact prompts, review-collection method and pain-point frequency results from mining Amazon competitor reviews, plus copyable steps you can reuse.
Project background
Cross-border e-commerce product selection is shifting from "picking winners off the bestseller list" to "reading reviews to find the gaps." A usable Amazon competitor review analysis has to align the rating distributions of several ASINs side by side, group 1–3 star reviews into common problems and count how often each one appears, separate the problems the product side can fix from those that belong to logistics and customer service, and finish with actionable selling-point recommendations. Done by hand, this usually takes more than half a day. This time, with Talos, I produced a product-selection pain-point frequency report across 6 competitors in 3 minutes.
Challenge
The difficulty comes down to two things. Review information is scattered across multiple product pages, and each ASIN differs in rating base, star distribution and negative themes, so aligning them side by side means paging back and forth by hand. Frequency counting is slower still: anyone can say "battery problems are the biggest," but actually counting the negative mentions for each theme by hand is time-consuming and easy to get wrong.
Method
Step 1: State the sample, the actions and the deliverable in one prompt
The core of the prompt is to specify the data scope, the analysis actions and the deliverable format at the same time. The full prompt used here:
Help me run a competitor negative-review analysis for portable blenders.
Competitors (Amazon US):
- https://www.amazon.com/dp/B0C2FF74J6
- https://www.amazon.com/dp/B0CN3PZX5W
- https://www.amazon.com/dp/B0H58VC12M
- If you want to switch categories, pick 3-5 from this list: https://www.amazon.com/gp/bestsellers/kitchen/
Open the review sections on these pages, look only at 1-3 star reviews, and do three things:
1. Group them into a list of common problems, marking the frequency of each;
2. Separate the ones the product can solve from logistics or customer service issues;
3. Give me 3 selling-point recommendations and state which pain point each one targets.
Output a table + a conclusion, and save it as a file.
Deliverable: product-selection pain-point report (competitor comparison + pain-point frequency + selling-point recommendations)This prompt contains four key elements:
- An explicit sample: 3 specific ASINs, plus a fallback instruction to "pick 3-5 from this list."
- Actions split into three: a common-problem list with frequencies, attribution between product-side and logistics/customer service, and selling-point recommendations, each mapped one to one.
- A narrowed data scope: only 1–3 star reviews, so the analysis effort stays on negative signals.
- A fixed output format: one table plus one conclusion, saved to a file.
Step 2: Talos executes automatically
After receiving the prompt, Talos worked through the following stages in order:
- Identity and scale check: opened each product page one by one, read the brand, rating, number of ratings and the 5/4/3/2/1 star distribution, and established the negative-review volume for each competitor.
- Review collection: went into the review section to locate negative reviews; the full review list page requires a login, so it switched to Amazon's official topic summary (Customers say), which provides the mention count, positive count and negative count for each topic.
- Sample expansion and cleaning: following the prompt's fallback instruction, added 3 wireless models from the category list, and identified that one of the specified ASINs had only 13 ratings, all of them free-product reviews, so it was removed from the negative sample.
- Structured extraction: used a script to extract the star distributions of 6 pages, the negative frequencies of 12 topics and the visible 1–3 star review text into 40 rows of detail data.
- Aggregation and attribution: merged semantically equivalent topics under fixed rules, ranked the four major problems, and assigned each problem to the product side or the logistics/customer service side.
- File output: one report and one detail table, written directly to the session working directory.
The whole run took about 3 minutes.
Step 3: Output structure
Competitor comparison
| Competitor | Rating / ratings | Share of 1–3 star | Topic with the most negative mentions |
|---|---|---|---|
| Ninja Blast BC151NV | 4.3 / 14,109 | 18% | Battery life 669, durability 360 |
| Ninja Blast Max BC251GY | 4.2 / 3,171 | 20% | Battery life 108, durability 92 |
| Hamilton Beach 51171G | 4.3 / 1,082 | 19% | Durability 45, lid closure 41 |
| nutribullet Flex | 3.7 / 758 | 34% | Battery life 55, value for money 41 |
| NexaEase 20oz USB-C | 4.1 / 1,099 | 24% | Reliability 35, value for money 31 |
| Kourice wireless 17oz | 5.0 / 13 | 0% | None; all 13 ratings are free-product reviews |
Frequency of the four common problems
| Problem | Total negative mentions | Negative rate range |
|---|---|---|
| Short battery life, charging faults | 855 | 55%–82% |
| Cannot crush frozen fruit or ice, weak power | 635 | 27%–68% |
| Breaks after little use | 613 | 62%–90% |
| Not worth the price | 543 | 47%–84% |
The report also produced a counterintuitive conclusion: portability and easy cleaning are the two best-rated aspects of the category (425 of the 506 mentions of portability for the Ninja Blast are positive), and almost all negative reviews come down to "whether it lasts."
Three selling-point recommendations
| Selling point | Pain point targeted | Key evidence |
|---|---|---|
| One charge, a whole week without worry | Battery life and charging | Battery life is the largest negative topic, 855 mentions in total |
| Frozen fruit and ice cubes, no blade jams | Power and blending | Power and blending total 635 negative mentions |
| First-year replacement only, no repair; we pay shipping | Short lifespan and warranty hassles | Lifespan 613 negative mentions; negative rate above 62% for all five products |
The final output was a report of more than 9,000 characters (about 2,500 words) and a 40-row detail table of negative frequency by topic. Data current as of October 2026.
Results
| Metric | Manual | Talos | Change |
|---|---|---|---|
| Total time | Half a day or more (estimated at a 4-hour minimum) | About 3 minutes | About −99% |
| Competitor coverage | Paging through review pages ASIN by ASIN | 6 ASINs in one run | More complete |
| Frequency counting | Manual count, review by review | Official topic counts, 40 detail rows across 12 topics | Reproducible |
| Review scope | 1–3 star only, requires login and paging | Public topic counts plus low-star review text as evidence | Traceable scope |
| Output format | Notes or a hand-built spreadsheet | Markdown report plus CSV detail | Editable and reusable |
Talos completed negative-review mining for 6 competitors in about 3 minutes, producing a product-selection pain-point report of more than 2,500 words and a 40-row frequency breakdown. The same work by hand takes more than half a day.
Business impact
- Single run: product-selection research drops from half a day to a few minutes, so research can happen before the decision instead of after the fact.
- Repeat runs: the product links and category list are the easiest parts of the prompt to swap; the same workflow carries over to a new category without rewriting the analysis structure.
- Information quality: frequency turns from a feeling into traceable data, and every conclusion can be traced back to a specific topic label and review text, so the team does not have to re-read the reviews to agree on the numbers.
- Format compatibility: output is a Markdown report and a CSV detail file that go straight into a planning document, an online spreadsheet or a product database, with no extra format conversion.
Reusable takeaways
- Give the sample in batches: name 3 definite competitors first, then add "if that's not enough, pick from this list," so a single ASIN without negative reviews does not force the whole run to start over.
- Write the analysis actions as a numbered list: frequency grouping, attribution and selling-point recommendations as three separate items, and the deliverable will address them in that order.
- State explicitly that only negative reviews matter: put "look only at 1–3 star reviews" in the prompt so the analysis effort is not diluted by positive reviews.
- Ask for "save it as a file": once the report and detail table are on disk they can be reused and edited directly, and the next category only requires swapping the sample.
FAQ
Q: Can all Amazon negative reviews be captured? A: The full review list page requires a login. What is publicly available is the star distribution, the official topic summary (including mention counts and positive/negative splits for each topic) and some pinned review text. The frequencies in this report use the negative counts from the official topic labels, so the scope is traceable; if you need every 1–3 star review captured one by one, you can log in through the browser panel and run it again.
Q: How are the frequencies calculated, and could items be counted twice? A: The raw frequencies come from the negative mention counts for each topic as provided by the platform. During aggregation, semantically equivalent topics are merged under fixed rules, and each rule is documented; the platform's topics overlap by nature, and the report flags this.
Q: Can the report be converted to Excel or PowerPoint? A: Yes. This run produced a Markdown report plus a CSV detail file, and the CSV opens directly in Excel; if you need a deck, feed the report content back in and generate one.
Q: Can this method be reused for other categories? A: Yes. It works for pet supplies, small appliances and outdoor gear; just replace the product links and category list in the prompt, and the analysis structure stays the same.