ToolScore
Data & Files 4 min read

Where Scraping Money Quietly Leaks: Six Traps in Vendor Fine Print

6 verified fine-print traps the number that matters here

By Stan K.

404s billed as success. One search page billed as 50 records. Proxy bandwidth burned on blocks. No-refund captcha deposits. The fine-print traps that turn a $99 plan into a surprise, verified against official policy pages.

Scraping vendors advertise one price and bill another — not dishonestly, mostly, but through fine print nobody reads until the first weird invoice. Here are six traps we verified against official policy pages, so you know exactly where the leaks are.

1. 404s and 410s are “successful requests”

ZenRows and ScrapingBee both bill HTTP 404 (Not Found) and 410 (Gone) as successful deliveries — the server responded validly, so the credit is spent. Server errors and timeouts get auto-retried free, but a dead URL is a charge. In a project that probes URLs you discovered yesterday, a 10% link-rot rate is a 10% surcharge you never see itemized.

2. A CAPTCHA wall inside an HTTP 200 is a “delivered page”

If the target returns status 200 with a challenge page or an empty shell, several APIs count it as delivered and bill it. The defense is content validation — a wait_for selector that fails the response if the real content never appears. Most teams discover this trap two invoices late.

3. Raw proxies bill bandwidth on blocked requests

Oxylabs states it in plain words: billing is based on bandwidth consumed, regardless of the HTTP status code returned. A block page is still bytes through the tunnel, and retries re-burn the same bytes. Managed unlockers exist exactly to flip this — Bright Data’s Web Unlocker bills only successful responses — which is why per-request unblockers look expensive per unit and often aren’t.

4. One request can be fifty records

Bright Data’s Scraper APIs bill per delivered record, not per request. One search-results URL that yields 50 product listings = 50 records billed. Even a failed scrape caused by an invalid input URL is billed, because the resources were spent. If you’re comparing a per-request API against a per-record API, you are comparing different units — convert both to cost per working page first.

5. Captcha solver deposits that never come back

CapSolver’s terms: no refunds in any form. 2Captcha’s explicit refund policy covers reCAPTCHA v3 rejects only, and its reportIncorrect mechanism reviews your accuracy stats — refunds dry up if your acceptance rate collapses. CapMonster Cloud refunds deposits within 3 days; SolveCaptcha charges only for solved. Deposit small, test the real target, then scale. The verified per-1K prices and refund policies for all five major solvers are in the captcha table of our hub.

6. Credits with an expiry date

ScraperAPI and ScrapingBee credits don’t roll over between months. SOAX credits expire in 60 days (annual-plan credits last 12 months). Bright Data subscription credits stop hard at cycle end. The only rollover champions in the space are IPRoyal — residential traffic never expires — and DataImpulse, same policy. If your workload is bursty, “use it or lose it” pricing quietly reprices your effective $/GB upward.

The pattern behind all six

Every trap converts a unit you can easily count (requests, GB, solves) into a unit you actually pay for (working data). The vendor isn’t lying; the denominator is just different from yours. Always re-price to cost per 1,000 working records before comparing vendors.

The checklist before you commit

  • Does the API bill 404/410? (ZenRows, ScrapingBee: yes)
  • Do you have a wait_for/content check for 200-with-captcha responses?
  • Raw proxies: is your block rate included in the bandwidth budget?
  • Per-record API: how many records does your typical target page yield?
  • Solver: what’s the refund policy, and did you test on the real target?
  • Credits: monthly expiry or genuine rollover?

The architecture-level math — how these traps compound across a million pages — is in The Real Cost of Scraping, and the full vendor tables with prices live in the Data & scraping hub.

Numbers change — the math doesn't
Token prices and plan limits shift monthly; the structural arguments in this guide age slowly. For current per-tool numbers see the live rankings.