How to Analyze Trustpilot Reviews and Find What Drives 1-Star Ratings
How to pull, clean and tag Trustpilot reviews and measure which problems really drive one-star ratings, worked on 1,788 public Carvana reviews where the most-named complaint barely moved the rating.
To find what drives one-star ratings on Trustpilot, pull every review with its stars and date (the API or an export), clean the text, tag each review with every problem it names, and then compare how much each problem raises the chance of a one- or two-star rating with the other problems, review length and period held fixed. Counting which complaint appears most in one-star reviews is not enough: in the public Carvana reviews we worked through below, the most-named problem on Trustpilot had almost no effect on the rating once the others were held fixed. An online retailer that wants Trustpilot read next to its tickets, app reviews and order data, and reported to the teams that fix things, will get the most from Dimension Labs: it brings every channel into one place, tags every review with Dimensions written for your business, and builds the reports for you.
The worked example uses 1,788 public Trustpilot reviews of Carvana written between January 1, 2024 and June 9, 2026, part of the five-site review set behind our Carvana review analysis.
How do I get all our Trustpilot reviews out?
Trustpilot's Business Units API has two review endpoints. The public one signs in with an API key and returns the reviews anyone can read on your profile. The private one needs a business user OAuth token and adds a start date and time filter, which is what you want for a weekly pull of only the new reviews. Both take your Business Unit ID.
From Trustpilot's developer documentation
Two endpoints return your reviews; only the private one takes a date range
Trustpilot's Business Units API, as documented on its developer site.
| Endpoint | Sign-in | Filters it lists |
|---|---|---|
| Public reviews /v1/business-units/{businessUnitId}/reviews | API key | Stars, language, order by date or stars, reviews per page |
| Private reviews /v1/private/business-units/{businessUnitId}/reviews | Business user OAuth token | The same, plus a start date and time |
API access comes with Trustpilot's API module, which Trustpilot sells in three tiers of yearly API calls: 200,000, 1,000,000 or unlimited. Keep these fields on every row: the review id, the date, the stars, the title and text, the language, whether the review is verified, and your company reply as a separate row. If your invitations carry an order or customer reference, keep that too; it is the key that later joins a review to the order, the product and the customer's spend. Our Trustpilot integration page maps each field.
What needs cleaning before you analyze Trustpilot reviews?
Three things, in the file we worked from. First, every review's text began with its own title, so a short review was counted twice if you read both fields; keep one. Second, 970 of the 1,788 reviews (54.3%) carried leftover highlight markup such as <em> tags from the page they were collected from, which a keyword rule or a model will read as words. Strip it. Third, keep reviews in every language but record the language, so you can split by market later.
Then look at the shape. Trustpilot ratings bunch at the ends: 48.2% of these reviews are five stars and 28.0% are one star, with only 5.9% in the middle. The average, 3.46, describes almost nobody. One-star reviews are also three times as long, a median of 152 words against 46 for five stars, so they carry most of the text you will tag.
Public Carvana reviews on Trustpilot
Most Trustpilot reviews are a five or a one, and the ones get the words
Share of 1,788 Trustpilot reviews of Carvana, January 2024 to June 2026, by star rating. Under each: the median length of a review at that rating.
Who is writing your Trustpilot reviews?
Many Trustpilot reviews come from invitations the business sends, so the mix of reviewers can follow who gets invited. Carvana's reviews show how much that matters. On Trustpilot, 10.9% of reviews from people selling a car to Carvana are one or two stars, against 76.7% from people buying one. A month in which more sellers wrote would look like a better month for everyone. Before you compare periods or read drivers, tag what the customer was doing (buying, selling, returning, asking for support) and compare like with like. The buying versus selling breakdown has the full split.
Which complaints show up in one-star Trustpilot reviews?
Tag every review, the five-star ones included, with every problem it names. You need both groups: a problem only means something if it is rarer in the reviews that went well. For this post we used the five keyword rules from the Carvana analysis, which flag a mechanical defect, a delivery delay, a refund or billing dispute, a title or registration problem, and not being able to reach anyone. Keyword rules are a rough first pass; a written definition per problem, applied by a model and checked against a person on a sample, does better, as how to tell if your AI tagging is accurate explains.
Counted, billing leads: 271 of the Trustpilot reviews name a refund, deposit or billing problem, 247 a mechanical defect and 214 a title or registration problem, against 80 for delivery delays and 78 for not reaching anyone. Each of the five is more common in low ratings. Reviews that name a mechanical defect are one or two stars 88.7% of the time, against 25.6% of the reviews that do not.
Problems also stack. Reviews naming none of the five are one or two stars 17.7% of the time; one problem, 58.5%; two, 70.1%; three or more, 80.6%. That is why a raw comparison overstates every problem: an angry review names several, and each one gets credit for the anger.
Which complaints actually drive one-star ratings?
To separate them, estimate each problem's effect with the others held fixed. We fitted a logistic model of a one- or two-star rating on all five problems, review length and the quarter, then compared each review's predicted chance with the problem switched on and off (logistic standardization), with 300 bootstrap resamples for the interval. The ranking changes.
Dimension Labs analysis, public Trustpilot reviews
Billing is the most-named problem on Trustpilot and the one that moves the rating least
Five problems ranked two ways across 1,788 Trustpilot reviews of Carvana: by how many reviews name them, and by how much each raises the chance of a one- or two-star rating with the others held fixed.
A mechanical defect raises the chance of a one- or two-star Trustpilot review by 31.5 points (95% interval 23.0 to 38.8), and not being able to reach anyone by 27.1 (15.2 to 42.2). Billing, the most-named problem, comes out at 1.0 point, with an interval from minus 3.8 to 6.3 that includes zero. Its raw gap was 27.9 points. A title or registration mention comes out slightly negative, 6.1 points lower.
Dimension Labs analysis, public Trustpilot reviews
Holding other things fixed cuts every raw gap, and billing drops to about zero
Raw gap: the one- and two-star share in reviews that name the problem minus the share in reviews that do not. Adjusted: the estimated effect with the other problems, length and period held fixed, with a 95% bootstrap interval.
No effect
Reading the billing reviews shows why. Of the 271 Trustpilot reviews the rule flagged for billing, 94 are four or five stars, and 48 of those contain the word deposit, mostly sellers saying a direct deposit arrived quickly. Of the 157 that are one or two stars, 46 also name a mechanical defect, the failure that set off the complaint. The count picks up both. That is the general point of which complaints cause bad reviews: the loudest topic is often not the costly one.
The published analysis ran the same five problems across all five review sites, holding the site, year, quarter and length fixed, and found the same order at the top: a mechanical defect adds 23.5 points, unreachable support 17.2 and a delivery delay 12.8, with billing at 4.5. Its heterogeneous-effect model put the defect effect on Trustpilot at 30.6 points, higher than on the App Store (28.1) or Google Play (21.6) and near zero on the BBB, where reviews are already almost all negative.
From the report: Carvana: a causal brief
Should you analyze Trustpilot on its own?
Read it on its own first, then next to your other channels. Each site draws a different crowd. In the Carvana set, 34.3% of Trustpilot reviews are one or two stars, against 97.4% on the Better Business Bureau and 23.2% on Google Play. Pooling them without a site column makes any problem that clusters on a complaint site look worse than it is; that is what happened to title complaints, which pile up on the BBB.
From the report: Carvana: a causal brief
A Trustpilot review also arrives without the context that explains its rating: the order, the product, the delivery date, whether the customer contacted support first. Joining reviews to orders on the reference you send with each invitation, and to support tickets on the customer, turns "billing complaints are common" into "billing complaints follow orders that needed a second contact." Combining reviews, surveys, tickets and calls covers the layout.
Which tools export and analyze Trustpilot reviews?
It depends on how far past the review you need to go.
Options for an online retailer
Where each option fits
What each is built for. Trustpilot's own features are named from its help center article titles.
| Option | Best for | What you need |
|---|---|---|
| Dimension Labs | Trustpilot read with tickets, chats, app reviews and surveys, tagged with your own Dimensions, joined to orders and revenue, and reported to every team | The Trustpilot integration or a CSV; custom Dimensions from the Discover plan; the join to orders and revenue on Prove |
| Trustpilot's business tools | Replying to reviews, inviting customers and watching your TrustScore and review topics inside Trustpilot | A Trustpilot business account |
| API plus a spreadsheet or notebook | A one-off read of a few thousand reviews by an analyst | The API module, a keyword list or model prompt, and time to check labels |
| Multi-location review tools | Brands that mainly need Google and Trustpilot reviews per location, with alerts and replies | See our comparison of review tools for many locations |
When Trustpilot's own tools are the better choice: if your goal is answering reviews quickly and keeping invitations flowing, stay inside Trustpilot; there is nothing to join. A spreadsheet is enough for a single read of one site. The heavier setup pays off when the question is which failures cost you stars and customers, and you need the answer every week across every channel.
What should an online retailer use to find what drives its one-star reviews?
Dimension Labs is an omni-channel analytics agent for customer data. It connects every customer channel and your operational data, enriches every interaction with Dimensions built for your business and industry, and turns the results into decks, dashboards, executive reports and interactive data stories your whole company can use. Plans start at $299 a month with no demo.
For Trustpilot that means four things. Trustpilot sits in one place with your support tickets, chats, app reviews and post-purchase surveys, so you can see whether a complaint in reviews also fills the inbox. Your order, product and delivery data comes in alongside, so each review carries the order it is about. Every review is tagged with Dimensions you write for your own business, such as damaged in transit, wrong size or refund not received, and the same labels serve support, operations and merchandising. And the agent builds the weekly Trustpilot report as a deck, dashboard or PDF and shares it across the company. Custom Dimensions start on the Discover plan at $999 a month; Explore at $299 uses the pre-built library; joining reviews to revenue and estimating each problem's effect with confidence intervals, as above, are on the Prove plan at $2,999 and up.
The Carvana analysis is the method on public data: 25 Dimensions on every review across five sites and a refutation-checked ranking of causes. Marriott, Ralph Lauren and Dropbox are Dimension Labs customers.
FAQ
How do I export all my Trustpilot reviews?
Use the Business Units API: the public reviews endpoint with an API key, or the private one with a business user OAuth token if you want to pull by date. Both need your Business Unit ID and the API module. Save the stars, date, text, language and any order reference you sent with the invitation.
Can I analyze a competitor's Trustpilot reviews?
Their reviews are public on their Trustpilot profile. Check Trustpilot's terms for how you may collect and use them. Comparing the same problem list on your reviews and theirs shows where you lose stars that they do not.
Why are most Trustpilot reviews either five stars or one star?
Customers tend to write when something went very well or very badly. In the Carvana set, 76.2% of Trustpilot reviews are a one or a five. Track the share of one- and two-star reviews rather than the average.
How many Trustpilot reviews do I need to find what drives one-star ratings?
Enough that each problem you test appears a few dozen times. In the example, the smallest problem appeared in 78 reviews and its interval ran from 15.2 to 42.2 points, wide but clear of zero. With a few hundred reviews, report counts and read the reviews; hold the effect estimates until you have more.
What is the best tool to analyze Trustpilot reviews?
For a retailer that needs Trustpilot analyzed with tickets, app reviews and order data and reported to every team, Dimension Labs. For replying to reviews and managing invitations, Trustpilot's own business tools are simpler.
To run this on your own reviews, start with one Trustpilot export: the date, the stars and the text are enough for a first read, and an order reference makes it better.