Reading Trustpilot Properly: Invited Reviews, Replies and the TrustScore Myth

Why the official API does not help you here

Trustpilot does publish APIs, and that fools people into thinking the data is one key away. It is not. The Business API is part of a paid Trustpilot plan and is scoped to the domains your account has verified ownership of — you can read and reply to your own reviews, send invitations, and pull your own TrustScore history. What you cannot do through any official endpoint is read a competitor's reviews, or sweep a whole category of companies in your market. That is exactly the data most teams want, and the only way to it is the public review pages at trustpilot.com/review/<domain>.

So the "Trustpilot API" you end up using for research is a scraper: it loads those public pages, walks the pagination, and returns JSON. The upside is you can point it at any company. The thing to internalise is that the page carries more structure than the star rating, and most of the value is in the fields people ignore.

The review object, field by field

A Trustpilot review is not just stars and text. These are the fields that change what conclusions you can draw.

Invited vs organic: the split that exposes engineered ratings

This is the heart of honest Trustpilot analysis. Companies on a paid plan can send automated invitations after every purchase, which surfaces a self-selected, mostly-happy population. Organic reviews — written by someone who went to Trustpilot of their own accord — skew more negative and more emotional, because people self-mobilise when something goes wrong.

The practical move: compute two scores in your own pipeline — one over invited, one over organic — and look at the gap. A small gap means the rating is broadly representative. A large gap (say invited 4.6, organic 3.1) means the public score is being floated by invitations, which is exactly the thing you want to know before you benchmark yourself against it or cite it in a pitch.

The business reply is half the dataset

Most scrapers and most analysts throw the reply object away, which is a mistake. For reputation work, whether and how a company answers its 1-star reviews is often more telling than the reviews themselves. A brand that replies to 80% of negatives within 48 hours with a real resolution reads completely differently from one that copy-pastes "please email support@" or ignores them. Capture reply.text and reply.repliedAt, derive a reply-rate and a median reply latency on the negative bucket, and you have a service-quality metric no star average gives you.

Pulling a competitor's negatives in one call

Pagination, the cookie wall, and rate limiting are all hidden behind the Actor. A run that grabs recent low-star reviews for a competitor looks like this:

curl -X POST "https://api.apify.com/v2/acts/renzomacar~trustpilot-reviews/run-sync-get-dataset-items?token=<APIFY_TOKEN>" \
  -H "Content-Type: application/json" \
  -d '{
    "companyDomains": ["competitor.com"],
    "stars": [1, 2],
    "sortBy": "recency",
    "maxReviews": 400,
    "language": "all"
  }'

Cap maxReviews on purpose. A well-known brand can carry tens of thousands of reviews, and for a complaint analysis the most recent few hundred per star bucket already show the pattern. Pulling everything just burns credits and earns you a rate-limit response.

Sentiment, but keep experienceDate attached

You can run the deduped review text through any sentiment model, but Trustpilot gives you a field Amazon does not: experienceDate separate from publishedDate. That lets you align complaints to when the bad thing happened, not when someone got around to writing about it. If a shipping change in March produced a spike of 1-star experiences, you will see it cleanly on experienceDate even though the reviews trickled in across April. Trend on experience date for root-cause work, and on published date when you are modelling how the live TrustScore will move.

Why your computed average will not match the badge

Expect a mismatch and do not chase it. The displayed TrustScore is time-weighted (recent reviews dominate), Bayesian (a handful of 5-star reviews is pulled toward the mean until volume justifies the score), and it folds in review frequency. A naive mean of scraped stars ignores all three. Use the raw stars for distribution and trend analysis, and treat the TrustScore as a separate, vendor-computed number you can capture but should not try to reverse-engineer from a single page.

Monitoring your own brand vs mining competitors

The two jobs need different cadences. Monitoring your own listing is an incremental, high-frequency pull — fetch only reviews newer than your last publishedDate high-water mark, alert on any new 1- or 2-star, and watch reply latency. Mining a competitor or a whole category is a deeper, less frequent sweep where you care about the full distribution, the invited/organic split, and language patterns across rivals. Do not run a nightly full re-scrape of a 40,000-review competitor; sync incrementally and you stay cheap, fast, and well under the rate limit.

Disclosure: links to Apify on this page are affiliate links, marked rel="sponsored". If you create a paid account through them we may earn a commission at no extra cost to you. We recommend Apify because we build and ship Actors on it ourselves, including the Trustpilot Reviews Actor linked above.