Trustpilot is the one platform in our measured set where the browser is not optional. On 26 September 2026 the home page answered a plain HTTP client through a US residential exit with 19 words, and a rendered browser with 2,553 words; on 28 September the plain response was 991 bytes and 19 words with no canonical, and a company review page rendered with a six-second wait returned 1,631,458 bytes and 4,389 words, including the number every review-monitoring job wants: 301,632 reviews behind a 4.3 TrustScore. The setting is residential, rendered with a wait, country pinned, and the reviews collector when you want each review as a row: ten in 14.8 seconds, with rating, date, verification flag and the company's reply.
The keyword is people checking proxy sellers
Search "trustpilot proxies" in the United States and Google assumes you are vetting a vendor. Eight of ten organic results are Trustpilot review pages of proxy companies; the related searches are "trustpilot proxies review", "trustpilot proxies complaints" and four vendor names. Autocomplete goes the same way: "ping proxies trustpilot", "lightning proxies trustpilot", "can trustpilot be trusted". The two results that are not Trustpilot pages are a forum thread asking for premium proxies for leaving Trustpilot reviews and a Reddit thread on "proxy for trustpilot reputation management". We come back to those two at the end, because Trustpilot's terms answer them in one line.
The people this post is for search for something else: "trustpilot scraper" autocompletes to github, free, python, api, extension, apify and "review scraper". They want reviews as rows for brand monitoring, competitor benchmarking, support triage and reply-time tracking. No page on either SERP measures what Trustpilot returns to a request, so we did.
Nineteen words without a browser, thousands with one
We audited trustpilot.com from a United States residential exit with our SEO audit, which fetches once as a pure HTTP client and once rendered, then rendered the home page and a company review page separately with a six-second wait before capture.
| Fetch | Bytes | Words | Title | Time | Canonical |
|---|---|---|---|---|---|
| Home, no JavaScript | 991 | 19 | 19 words, no canonical | 5.0 s | absent |
| Home, rendered, 26 September | n/a | 2,553 | Trustpilot Reviews: Experience the power of customer reviews | n/a | trustpilot.com |
| Home, rendered with a six-second wait, 28 September | 1,895,465 | 3,588 | Trustpilot Reviews: Experience the power of customer reviews | 53.0 s | trustpilot.com |
| Company review page, rendered with a six-second wait | 1,631,458 | 4,389 | Wise Reviews | Read Customer Service Reviews of wise.com | 46.4 s | /review/wise.com |
| Reviews collector, one company, cap 10 | n/a | 10 reviews | n/a | 14.8 s | n/a |
Every other platform in this series rewards the plain fetch. Trustpilot gives it nothing: the plain response is 991 bytes with no canonical, no description and 19 words, and no TLS fingerprint we sent changed that. The page assembles itself in JavaScript over a few seconds, which is why the wait matters. A render that captures the moment the document settles gets zero words; a render that waits six seconds and then for the h1 gets the whole page, with the meta description Trustpilot writes for link previews: "Check out what 301,632 people have written so far". That description is the review count, server-side, and it is the cheapest number on the page once the render has completed.
The trap is what the successful document says about itself. The rendered page arrived with a complete body, a canonical link, an h1 and 4,389 words, and a status code that does not describe the body, because the browser keeps the status of the first document in the chain while the page completes in JavaScript. A retry loop keyed on the status code will discard a complete page and fetch it again, at 1.6 megabytes a time. Classify on the canonical link and the h1, never on the status. On Trustpilot the body tells the truth and the status code does not.
The exit country does not pick the site
We ran the identical audit through a United Kingdom residential exit. The plain fetch returned the same 19 words. The rendered fetch returned the same trustpilot.com, canonical unchanged, h1 "Discover, read, and write reviews", 2,114 words. Trustpilot did not redirect the UK exit to its UK site and did not translate the page; its country editions live on subdomains and paths that you choose in the URL, not on the exit IP.
| Exit | Final URL | Words, no JavaScript | Words, rendered | Rendered title |
|---|---|---|---|---|
| United States | www.trustpilot.com | 19 | 2,553 then 3,588 | Trustpilot Reviews: Experience the power of customer reviews |
| United Kingdom | www.trustpilot.com | 19 | 2,114 | Trustpilot Reviews: Experience the power of customer reviews |
So the exit country is not a content lever on Trustpilot, and the word counts differ by a few hundred because the home page rotates its featured categories and reviews between renders, not because of geography. Pin country=us for the .com and use the exit only to keep one job on one pool. The market lever is in the URL: a company's reviews live at /review/<domain> on every edition, and the collector's country field sets the locale for the run.
Ten reviews as rows, with the reply, in 14.8 seconds
We ran the Trustpilot reviews collector once, domain wise.com, sorted by recency, capped at ten. It returned ten reviews in 14.8 seconds and a run note reading "TrustScore 4.3, 301632 reviews". Each row carried the review id and URL, title, text, star rating, review date, experience date, reviewer name, reviewer country, how many reviews the reviewer has written, whether the review was verified, the like count, the company's reply and the reply timestamp.
Three things in those ten rows are the reason to collect rather than render. Ten of ten had a company reply, and the reply timestamps make response time a metric: the newest review was posted at 16:52 and answered at 17:00, eight minutes. Seven of ten were verified, which is the field that separates invited reviews from organic ones. And the reviewer countries were HU, FR, DE, DE, HR, IQ, GB, GB, GB, AU for a review page fetched from the United States: the audience is where the customers are, not where the exit is. Seven of the ten were five stars and two were one star, and both one-star reviews were about account restrictions; a monitoring pipeline that keys on rating and title finds a theme in ten rows. The collector takes a stars filter and a relevance sort, and caps at 100 reviews per run.
Trustpilot's terms name AI agents and screen scrapers
trustpilot.com/robots.txt is 21,744 bytes and it is a list of named exceptions. Googlebot, bingbot, GPTBot, ClaudeBot, PerplexityBot, OAI-SearchBot, ChatGPT-User and Claude-User each get the same block: company review pages allowed, individual review pages under /reviews/ disallowed, /api/ and /search disallowed, and every filtered listing, ?stars=, ?sort=, ?languages=, ?verified=, disallowed. Twenty named agents, among them CCBot, Bytespider, Diffbot, cohere-ai and Meta-ExternalAgent, are disallowed from everything. Then the wildcard: User-agent: *, Disallow: /. Three sitemaps follow. The 19-word plain response we measured is that last line in practice.
The Terms of Use for Consumers, version 7.0, are the contract, and they are the most explicit we have read on this subject. The definition of "you" includes anyone accessing the platform directly or indirectly, "including, for example, through use of automated technologies such as AI agents or screen scrapers". Section 4 then says you must not access, search or collect content by any means, automated or otherwise, except as provided on the platform or specifically approved, and must not carry out any text mining, data mining or web scraping of the platform for any purpose without express permission, including for training AI models. The same section forbids writing or procuring fake reviews, and section 3 allows one user account per person.
That last sentence is the answer to the two non-Trustpilot results on the SERP. Proxies for leaving reviews and proxies for reputation management are proxies for posting from accounts that are not yours, which the terms forbid in plain words, and we do not help with it. Reading a company's public reviews for monitoring is a different act, and Trustpilot sells a licensed version of it: the Data Solutions API at developers.trustpilot.com exposes six endpoints, business unit search and details, service reviews, latest reviews and categories, in two products, Display and Insights, with a deletions feed for privacy compliance. If the reviews are the product you are building on, that is the route with permission attached. What a server hands a browser is a technical fact; permission to reuse it is contractual, and reviews carry reviewer names, which are personal data wherever your users live.
Cost per thousand company pages, and why the plain fetch is not cheap
Prices from our pricing page: residential Basic $0.80/GB, the web scraping API $0.001 per rendered page, the reviews collector $0.002 per delivered review. A gigabyte is counted as 10^9 bytes.
| Approach | Bytes per page | Pages per GB | Cost per 1,000 | What you get |
|---|---|---|---|---|
| Plain HTTP, residential Basic | 991 | 1,009,082 | $0.0008 | 19 words, no reviews |
| Company page rendered with a wait, residential Basic | 1,631,458 | 613 | $1.31 | 4,389 words, one page of reviews, the review count |
| Home rendered with a wait, residential Basic | 1,895,465 | 528 | $1.52 | 3,588 words of featured categories |
| Web scraping API, rendered | n/a | n/a | $1.00 | The page as Markdown, wait handled, failures never billed |
| Reviews collector | n/a | n/a | $2.00 | 1,000 parsed reviews with rating, dates, verification and reply |
The first row is the cheapest line in the table and it buys nothing: a million 19-word responses per gigabyte, zero reviews. On Trustpilot the expensive row is the working row, and rendering a company page costs about $1.31 per thousand in bandwidth alone, before the 46 seconds of browser time each. Byte figures are decoded bodies and exclude TLS overhead. The collector at $2.00 per thousand reviews is the row that scales, because it is billed on reviews delivered rather than on pages rendered, and a page that returns nothing costs nothing. For the opposite shape, a platform where the browser is the thing that stops you, see what we measured on Reddit through residential proxies. For how the same named AI crawlers are treated across sites, should I block AI crawlers.
Where we stop
Everything above is about reading a company's public review page and its public reply record: review monitoring for your own brand, benchmarking competitors, tracking response times, and pulling the review count that Trustpilot itself prints in the description. It does not cover accounts. We will not help with posting reviews from multiple accounts, procuring reviews, flagging competitors' reviews at scale, or anything the terms call fake reviews or manipulation; Trustpilot allows one account per person in section 3 of its terms and forbids all of it in section 4, and a proxy does not change who wrote the review. If you also monitor Google, the Google reviews collector takes a place id and returns the same shape.
The setting that works on Trustpilot
- Network: residential proxies, Basic line at $0.80/GB. The rendered page arrived complete from residential exits in the United States and the United Kingdom; the page is built by JavaScript, not gated by the address, so Premium at $2.20/GB, mobile at $2.30/GB and ISP at $2.50 per IP per month buy nothing here.
- Fetch mode: rendered, with a wait of six seconds before capture. 19 words over plain HTTP, 2,553 words rendered on 26 September, 3,588 on the home and 4,389 on a company page on 28 September. Classify the result on the canonical link and the h1, never on the status code.
- Country: pin
country=usfor trustpilot.com. The exit does not change the edition, 2,553 words from the United States and 2,114 from the United Kingdom on the same canonical; choose the market in the URL. - When the proxy is not enough: the trustpilot_reviews collector, 10 reviews with rating, dates, verification and the company reply in 14.8 seconds at $0.002 per review, up to 100 per run. For the page as Markdown with the wait handled, the web scraping API with rendering at $0.001 per page.
- Free tier: every account gets $2 of free API usage per month, which is 1,000 parsed reviews or 2,000 rendered pages before you pay anything.