Sebastian JamesKey takeaways
- Shared problems get endorsed; personal incidents don't. Search and filter complaints are 10% of all complaints and 34% of the most-endorsed ones.
- Fixes change minds; apologies barely register. Of reviewers who raised their rating, 90% said the problem was fixed or the feature came back. Customer service came up in 5%.
- The core customer groups hold steady for years: customer service, Prime value, account and gift card, delivery and refunds.
- Complaint spikes follow releases. Seven of 12 spikes were dominated by one new app version.
- Each finding is a way to court a competitor's customers: strike on their bad release, reach the people still waiting for a fix, and answer their most-endorsed complaints.
Reviews are the one place customers answer back. Nobody prompts them, nobody pays them, and other customers vote on whether they're right. That makes a pile of reviews something rare: a record of what people believe, in their own words, with a built-in measure of who agrees.
I ran four experiments on 88,222 English-language Google Play reviews of the Amazon Shopping app, September 2018 to October 2026, using Niles, my review analysis engine. Each review carries its text, stars, thumbs-up count, app version and date. The question behind all four: what does real customer language tell you about persuasion, and how do you turn it into sales?
1. Customers endorse the problems they share
Among 49,108 one- and two-star reviews, I compared each complaint type's share of the most-endorsed 1% (244 or more thumbs-up) with its share of all complaints, then checked the result with two models that control for stars, review length and calendar quarter.
| Complaint type | Share of all complaints | Share of most-endorsed 1% | Thumbs-up vs. similar reviews |
|---|---|---|---|
| Search and filters | 10% | 34% | +70% |
| Layout and added steps | 4% | 12% | +38% |
| Nagging: notifications, Prime prompts | 3% | 7% | +28% |
| Crashes, freezes, slowness | 16% | 28% | +24% |
| "I love Amazon, but…" | 6% | 16% | +23% |
| Lost features ("used to," "bring back") | 11% | 23% | +18% |
| Delivery and refund failures | 21% | 19% | −13% |
| Account closed or locked | 2% | 0.2% | −21% |
Search, layout and broken features happen to everyone, so readers vote for the review that says it best. A lost package happens to one person. It draws sympathy, not endorsement. Threats to leave ("I'm switching to…") had no reliable effect at all.
2. Fixes change minds. Apologies don't.
681 reviews carry an explicit edit: "EDIT:," "Update:," "changed my rating." In 278 the direction is clear. Forty reviewers raised their rating. Of those, 90% said the problem they'd complained about was fixed or the feature came back. Customer service came up in 5%; refunds in 2%.
The other 238 lowered their rating or said the problem persisted, outnumbering the converts about 6 to 1. Sixty-two percent of them used the same word: still.
Minds change when the specific thing gets fixed. They harden when time passes without a fix.
3. The core customer groups last for years
I grouped complaints separately for 2018–2022 and 2023–2026, then matched the groups across periods by vocabulary. Four are nearly identical across eight years: customer service, Prime value and shipping, account and gift card, and delivery and refunds. Groups tied to one bug (a language-switching error, a crash on open) came and went. New since 2023: people who feel forced into the AI shopping assistant, and people asking for dark mode.
4. Complaint spikes follow releases
I flagged days with at least triple the usual review volume and found 12 spike episodes. Seven were dominated, over 70%, by a single new app version: crashes in October 2019 and November 2023, a search bar moved to the bottom of the screen in June 2023, a navigation overhaul in December 2025.
The AI assistant shows a different shape. Its share of reviews jumped from about 3% a month to 16% in June 2026 and has stayed between about 7% and 14% a month since. That's not a bad day. It's a standing grievance.
How to use this to win new customers
Read your own reviews and these findings tell you what to fix. Read a competitor's reviews and they tell you who to go after, when and with what message.
- Strike the week a rival ships a bad release. Their complaint spike is your sales alert. Run a switch campaign within days, aimed at exactly what broke.
- Reach the people who are still waiting. A competitor's "still broken after six months" reviewers have given up on a fix. Offer the thing they're waiting for.
- Answer their most-endorsed complaints. Thousands of strangers have already pretested those headlines. "Search that finds what you typed" writes itself.
- Court the frustrated fans. "I love it, but…" reviewers like the category and want a reason to switch. The "but" is your ad.
- Run standing campaigns for the groups that last. Value doubters get a value guarantee. People burned by customer service get a promise of a person who answers.
- Feature what they ignore. The single most-endorsed review in this dataset, with 5,170 thumbs-up, asks for dark mode.
How far to trust it
Complaint types were labeled with transparent keyword rules and spot-checked by hand, so read them as strong direction, not exact counts. Thumbs-up reflects visibility as well as agreement, since app stores show popular reviews first. The rating-edit sample is small. And all of this describes customers who chose to write a review, not every Amazon shopper.
The Niles demo runs on this same review set. Start at "who can be saved."
Give me your reviews.
In a nutshell, I'll hand you a plan to recover lost sales and win new customers: who you can still win back, what they say would do it, and where your competitors are leaving customers on the table.
Email me about a review read