Similarweb Top Websites Scraper — Category Rankings

Turn Similarweb's public Top Websites tables into data. One row per ranked website: rank, domain, the site's own category, month-over-month rank change, isNewToRanking, avgVisitDuration, pagesPerVisit and bounceRatePercent, across 209 categories and 67 countries — every slug read live off Similarweb's own taxonomy, so nothing here can be mistyped. No login, no API key, no CAPTCHA solving.

Read the cap before you plan a run

Similarweb does not publish the same number of rows for every URL shape. All eight shapes were measured on 2026-08-15, through Apify RESIDENTIAL:

What you ask for	Rows Similarweb publishes
Worldwide, no category (/top-websites/)	50
Worldwide + category (/top-websites/finance/)	50
Worldwide + subcategory (/top-websites/finance/investing/)	50
One country, no category (/top-websites/united-states/)	50
One country + category (/top-websites/united-states/finance/)	5
One country + subcategory	5

Combining a country with a category gives you 5 rows, not 50. That is a product decision on Similarweb's side, not a paging limit — nothing in the URL, the query string or the page's embedded state pages past it. This Actor names every affected pair in the run log before it fetches anything, and you are billed per row delivered, so a 5-row slice costs you a tenth of a 50-row one. If you want 50 rows, use Worldwide + a category, or a country with no category.

Two more limits, stated up front rather than discovered in the CSV:

• There is no visit count on this table. Similarweb publishes rank and the three engagement metrics on the ranking page and nothing else for logged-out visitors. This Actor does not invent one. Each row carries websiteAnalysisUrl — the site's own Similarweb page — if you need to go deeper by hand.
• These are Similarweb's modelled estimates, from its panel and clickstream data. They are not any site's own analytics, and Similarweb itself labels them "estimated data" everywhere on the page. Treat rank and bounce rate as directional market signal, not as measured truth.

---

The shipped defaults

{
  "categories": [
    "finance",
    "computers-electronics-and-technology/programming-and-developer-software",
    "e-commerce-and-shopping/marketplace"
  ],
  "countries": ["worldwide"],
  "includeAllCategories": false,
  "maxItems": 500
}

Click Try for free and hit Start — that block is literally the input the Actor ships with. A run on those defaults returned 150 rows in 11 seconds, from 3 slices, on 3 HTTP calls, and costs $0.15.

A larger validation run on 2026-08-15 — 22 categories × 2 countries (worldwide + United States), including both overall rankings — returned 1,255 rows from 44 of 44 slices in 3 minutes 3 seconds on 45 HTTP calls (one transient proxy retry), with zero duplicates.

---

What a row looks like

{
  "rankingScopeCountry": "Worldwide",
  "rankingScopeCountryCode": null,
  "rankingScopeCategory": "Finance",
  "rankingScopeCategorySlug": "finance",
  "dataMonth": "July 2026",
  "snapshotDate": "2026-07-01",
  "rank": 1,
  "domain": "paypal.com",
  "siteCategory": "Finance > Banking Credit and Lending",
  "siteCategoryId": "finance/banking_credit_and_lending",
  "rankChange": 0,
  "isNewToRanking": false,
  "avgVisitDuration": "00:02:23",
  "avgVisitDurationSeconds": 143,
  "pagesPerVisit": 4.96,
  "bounceRatePercent": 21.2,
  "websiteAnalysisUrl": "https://www.similarweb.com/website/paypal.com/",
  "rankingUrl": "https://www.similarweb.com/top-websites/finance/",
  "faviconUrl": "https://site-images.similarcdn.com/image?url=paypal.com&t=2&s=1&h=…",
  "dataSource": "Similarweb public estimates",
  "scrapedAt": "2026-08-15T16:37:20.191Z"
}

Field fill — measured, not estimated

Measured on the full 1,255-row validation run above (22 categories × 2 countries), sorted descending. Sparse fields are bold.

Field	Fill	Note
rank	100%	position inside the slice you asked for
domain	100%
siteCategory / siteCategoryId	100%	the site's OWN main category, which is often a different leaf from the slice it was listed under
rankChange	100%	positions gained vs last month; 0 on 29.7% of rows, non-zero on 70.3%
avgVisitDuration / avgVisitDurationSeconds	100%
pagesPerVisit	100%
bounceRatePercent	100%
dataMonth / snapshotDate	100%	Similarweb's monthly snapshot, not the scrape date
websiteAnalysisUrl / rankingUrl / faviconUrl	100%
rankingScopeCountry / rankingScopeCategory	100%	echoes back what you asked for, so a merged CSV stays sortable
rankingScopeCategorySlug	92%	null on the "overall ranking" slices, which have no category
rankingScopeCountryCode	12.4%	null for the Worldwide scope, which has no country
isNewToRanking	1.0% true	Similarweb's own "new entry" flag — 12 of 1,255 rows. A genuinely rare signal, and the most interesting column in the file if you are hunting risers
monthly visits	absent	not published on this table at all — see the cap section above

---

Who buys this

• Competitive intel and agency pitch decks — "here are the top 50 sites in your category, with engagement benchmarks, and here is where you sit." bounceRatePercent, pagesPerVisit and avgVisitDurationSeconds across a whole category give you the benchmark line that makes a category chart mean something.
• M&A and investor screening — 209 categories × 50 rows is a screenable universe of category leaders with a month-over-month rank delta on every one.
• Ad-tech and B2B prospecting — a ranked, category-tagged list of the biggest sites in a vertical is a target list; websiteAnalysisUrl is the drill-down.
• Rank tracking over time — run it monthly on a schedule and diff rank and rankChange. The isNewToRanking flag surfaces entrants without you having to diff anything.

rankChange and isNewToRanking are the columns you cannot rebuild yourself from a single pull: they are Similarweb's own month-over-month comparison, already computed.

---

How it gets the data

The ranking pages server-render their whole table into window.__APP_DATA__ before React runs, so this Actor reads the app's own state object rather than scraping styled DOM nodes. That is both more precise — the rendered table prints 23.85% and 8.43, the state carries 0.23845442795516247 and 8.429506649675165 — and more stable, because the table's class names are build artefacts of a versioned bundle while the state keys are the app's contract.

Transport, measured 2026-08-15 through Apify:

• This route carries no bot wall: plain HTTP, no x-amzn-waf-action header, no token, no browser. 44 of 44 slices succeeded on the validation run.
• Datacenter proxies are refused. 13 of 30 datacenter calls came back a bare HTTP 403 from CloudFront (56.7% success), so Apify RESIDENTIAL is the default and you should leave it alone.
• The Actor retries a slice up to 3 times with a fresh residential session before giving up, and a slice it could not read is never billed.
• It runs in 512 MB and needs no Chromium.

(For contrast, and so you are not surprised: Similarweb's per-domain pages under /website/ are a different story — those sit behind an AWS WAF that escalates to a visual CAPTCHA. This Actor does not touch them. It only reads the ranking tables, which are open.)

robots.txt

https://www.similarweb.com/robots.txt, fetched 2026-08-15, contains these Disallow lines for User-agent: *, verbatim:

Disallow: */search/*
Disallow: */adult/*
Disallow: /corp/*.pdf$
Disallow: /corp/solution/
Disallow: /corp/lps/
Disallow: /corp/get-data/
Disallow: /corp/unlock-growth/
Disallow: /silent-login/
Disallow: /signin-oidc/
Disallow: /signout-oidc/

The /top-websites/ tree is not disallowed, and this Actor requests nothing else. Because */adult/* matches /top-websites/adult/, the adult category has been removed from the picker even though Similarweb publishes it — and the Actor refuses it with an error if you send it through the API anyway. 209 categories ship, not the 210 Similarweb lists.

---

Pricing

$0.001 per ranked website delivered. 150 rows (the shipped default) = $0.15. The whole worldwide taxonomy — 209 categories × 50 — is about 10,450 rows for $10.45.

You are charged per row you actually receive. Slices that fail to load are not billed, and a domain repeated inside one slice is dropped before billing. maxItems × $0.001 is a hard ceiling on what a run can cost you.

---

Notes and gotchas, all measured

• A subcategory is not a subset of its parent. finance and finance/investing are two separate 50-row rankings. Taking both is how you go deep in a vertical, and the overlap is small.
• siteCategory ≠ the slice you asked for. A site can rank inside finance while its own main category is finance/banking-credit-and-lending. Both are in the row so you can group either way.
• The same domain appears once per slice. Ask for 3 categories and a domain that ranks in two of them gives you 2 rows, with different rankingScopeCategory values. Deduplication is per country|category|domain, which is the honest key — you are not billed twice inside one slice.
• dataMonth is Similarweb's snapshot month, not today. It rolls over once a month; every row in a run carries the same one.
• Similarweb's own numbers move. Rank, bounce rate and pages-per-visit change with each monthly snapshot. Schedule the Actor monthly if you want a series.