AI Web Scraper: 7 Tools Priced by What a Credit Actually Buys

If you want every competitor's price in one sheet by Monday morning, the useful question is not which scraper is smartest. It is who does the work, and what each page costs.

An AI web scraper turns a web page into rows. You point at the data or describe it in plain English, and the AI works out where it sits on the page. Apify's description of its own AI scraper puts the selling point in one line: "No hard-coded logic that breaks when a website changes."

That is the promise of AI web scraping. The bill is where the tools split. Each vendor counts in its own unit: a page, ten rows, one row, a page load, half a cent. So I priced every tool in the unit it actually sells, and counted each free tier the same way.

I read every pricing page, product page and help page cited here on 17 October 2026. Bright Data shows up in the research scripts behind this site and Firecrawl in Gravity's catalog; each one's section says how.

Tools 1 to 5 need no code. Tools 6 and 7 are for builders. The legal question has its own section, answered from the primary sources. And if scraping is one chore among many, the guide to the AI tools a small business actually uses covers the rest.

How I compared them

Every tool got the same five questions, answered from its own pages:

• Who does the work? You click, you describe the job, a developer writes code, or the vendor's own team runs it for you.
• What does one credit buy? The word means something different at each vendor.
• What does free give you, counted? In the vendor's own unit.
• What happens when a site blocks you? One charges more for pages on protected sites, one hands the browser back to you, and one bills the error page.
• Where does the data land? Browse AI and Thunderbit send rows to Google Sheets; Firecrawl returns Markdown or JSON through its API.

Prices are in US dollars as each vendor's own page showed them on 17 October 2026, with the billing state noted, because four of the seven pages open on annual billing. No scraper was run, trialled or signed up for, so every feature claim below is the vendor's own.

Quick comparison

Prices in USD from each vendor's own page on 17 October 2026, with each page's default billing state shown first, except Thunderbit (monthly first). A credit is a different unit at each vendor; see each section.

Read the unit column before the price column. The same list of 50 products is 5 credits at Browse AI, where ten rows make a credit, and 50 credits at Thunderbit, where each row is one. Neither number is a bill until you multiply it by what that vendor charges for a credit.

1. Browse AI: best for watching competitor pages

Browse AI calls itself an AI web scraper and describes the method in one line: "point it at any website, the AI builds your columns, then you review and approve." You can add a field in plain language. After that, a robot runs on a schedule and checks the page for changes.

The watching is why you would pick it. Monitoring runs as often as every five minutes, and every plan, the free one included, carries the REST API, webhooks, residential proxies and integrations such as Google Sheets, Airtable and Zapier. Plans count websites as domains: 2 on Free, 5 on Personal, 10 on Professional.

One page or ten rows costs a credit, whichever is more.

That is Browse AI's own rule, and its FAQ works the examples. A list page with 50 products costs 5 credits; a page that returns two fields still costs 1. Then the warning: "Pulling a list is cheap. Opening each item is not." Taking 100 products off a listing page costs 10 credits, and opening each of those 100 product pages costs 100 more. Monitoring spends credits on every check:

• A page checked daily runs about 30 credits a month.
• Hourly is closer to 720.

Amazon, LinkedIn and eBay pages run through residential proxies, so they start at 2 to 10 credits each. At one credit a check, the free 50 credits cover roughly one page checked daily. Run out, and, in Browse AI's words, "your robots pause rather than charging you by surprise." If you want nobody on your side to build anything, Premium adds "our team of data engineers and AI experts", starting at $500 a month billed annually.

Best for

A small business watching a handful of competitor pages for price or listing changes.

Pricing

Free; Personal $19 a month billed annually ($228, 12,000 credits a year) or $48 month to month (2,000 credits a month); Professional $69 billed annually or $87 monthly. As of October 2026.

Free tier

50 credits a month (up to 50 pages and/or 500 rows), 2 domains, no card.

Verdict

Personal on annual billing is the cheapest step up if 1,000 credits a month (12,000 a year, paid upfront) will do.

2. Thunderbit: best for one table from the page you are on

Thunderbit is a web scraper Chrome extension, with an Edge version too, "Built for sales & ops teams". You open the page you want and click. In Thunderbit's words: "Let AI think of how to organize this website, and extract the table for you." When you need something specific, you describe the data types you want.

Rows go to Google Sheets, Airtable or Notion, and saving there costs no credits. It reads PDFs, images and documents too; if most of your sources are files, AI document processing tools are priced for that job.

The free plan stops at six pages a month.

Each of those pages can use at most 30 credits. One row of data is one credit, and the FAQ adds: "If subpage scraping is enabled, 1 row of data = 2 credits." Starter moves you to 500 credits a month and, per the FAQ, "advanced features like pre-built scrapers, subpage scraping, bulk scraping, and pagination".

For popular sites like Amazon, eBay and Google Maps, a template gives you the same fields every time, "with no AI credits used". Use one where it exists; pre-built scrapers start on Starter. Over the limit, you upgrade or "wait until your credits reset".

Read two terms before paying. Thunderbit "does not offer refunds for unused time or credits", and free-plan data is kept for 14 days (60 on paid plans). The 7-day trial includes all Pro features, 8 pages and unlimited credits, so run your real job inside it first.

Best for

Pulling one table from the page in front of you, today.

Pricing

Starter $15 a month (500 credits) and Pro $38 a month (3,000 credits), billed monthly; Starter is $9 a month billed yearly (5,000 credits a year). As of October 2026.

3. Octoparse: best for big recurring jobs on popular sites

Octoparse sells "Agentic Web Scraping for Anyone", and its homepage promises "Three ways in, none of them code":

1. Run a ready-made template.
2. Let AI Assistant browse the site and build the task for you.
3. Start from Auto-detect in the desktop app, then refine the task with drag-and-drop.

The pricing page does not say which plans include AI Assistant. Ask before you count on it.

Everything on the free plan runs on your own computer.

Free is the desktop app with 10 tasks, one device and 50,000 rows of export a month, up to 10,000 per export, with "no credit card required". Data leaves as Excel, CSV, JSON, HTML or XML. The catch sits in the plan table: "Run tasks on local devices only". Your computer does the work, so it has to stay on while the task runs.

Standard moves the work to Octoparse Cloud, with 100 tasks, 500+ preset templates, residential proxies, automatic CAPTCHA solving and task scheduling. The template examples include Amazon products and reviews, Google Maps businesses, X posts and TikTok. Startups get 30% off for a year, and every paid plan comes with a 5-day money-back guarantee.

Pricing

Free; Standard $69 a month billed yearly or $83 monthly; Professional $249 or $299 (250 tasks, up to 20 concurrent cloud processes); Enterprise custom. As of October 2026.

Verdict

Start free on a machine you can leave running, and pay for Standard when the job needs a schedule.

4. Gumloop: best when scraping is one step of a bigger AI job

Gumloop is an agent platform whose agents can browse the web and fetch pages. The agent browser runs Chromium, signs in with saved logins, fills forms and downloads files. "When it hits a login, 2FA code, CAPTCHA, or purchase confirmation, the agent pauses and hands the browser to you." To read a public page, an agent can use Web Fetch and skip the browser entirely.

Traffic can go through a residential proxy, and "The proxy has no extra charge." The docs are candid about the ceiling: "Some sites block every automated browser, whatever the proxy, and retrying won't help."

Every task carries an 8% fee on top of what the agent spends.

Gumloop bills what the agent spends, in credits, then adds a fee. Four rules from its pricing FAQ and docs, as of October 2026:

• One credit is $0.005, so $1 buys 200.
• Pro, from $37 a month, includes 20,000 credits (7,400 plus 12,600 bonus) that reset each billing period.
• An 8% orchestration fee applies to the full list cost of each task, tokens, compute and tool calls included. Bring your own model keys and it rises to 16%.
• At the limit, agents stop until you switch on pay as you go, which is off by default.

The docs add a warning: "The same agent might cost 2 credits for a quick question and 200 for a deep research task."

Every new account starts with a 14-day Pro trial that rolls into a paid subscription unless you cancel first, so put a reminder in your calendar on day one (future you will be grateful). If you are really shopping for a workflow builder, start with this comparison of n8n alternatives.

5. Apify: best when someone has already built the scraper

Apify is a store of ready-made scrapers called Actors. Its site menu read "Browse 89,427 Actors" on 17 October 2026, and builders can publish their own (where to publish AI agents covers that side). If someone has already built a scraper for the site you need, start there.

The one that fits this guide is the AI Web Scraper, maintained by Apify. Its README calls it an "AI-first web scraper that extracts structured data from any website using natural-language prompts. No programming knowledge required." You paste URLs and a prompt, then pick Single page, Scout or Agentic mode. AI tokens are included in the Actor's cost, so you need no separate LLM subscription, and Apify's built-in scheduler can run it on repeat.

Your Apify plan sets the price of each page.

The Actor "charges you per extracted page regardless of the extraction mode used". The rate falls as your platform plan rises, from $30 per 1,000 pages on Free to $20 on Business, as of October 2026. Each paid plan includes the same amount to spend: the Starter plan, $19 a month billed monthly, gives you $19 to spend in the store. On Free, $5 a month covers up to 166 page extractions, the figure on Apify's own card.

Cap Agentic mode before you run it. It "finds the right pages through the sitemap and Start URLs, then crawls and extracts structured data", and its maxPagesToVisit setting "applies per Start URL", so a short list of start URLs can still bill a lot of pages.

Best for

A big, popular site somebody has already written a scraper for.

Pricing

AI Web Scraper $30 (Free), $25 (Starter), $22 (Scale) or $20 (Business) per 1,000 pages. Plans $0, $19, $199 and $999 a month billed monthly; 10% less billed annually. As of October 2026.

Free tier

$5 of usage a month, no card: up to 166 AI Web Scraper pages.

6. Firecrawl: best for builders feeding pages to an AI

Firecrawl calls itself "The web data API to search and scrape". A developer sends it a URL and gets back what its homepage calls "llm-ready data": Markdown, JSON, a screenshot and more. Its terms of use put it plainly: "Firecrawl is an API that converts any website into LLM-friendly data."

It is open source under AGPL-3.0, and it connects "to any AI agent or MCP client in minutes". The docs also list a browser Playground for trying it without code, an Agent endpoint, and change tracking that "compares the current content of a page against the last time you scraped it".

Asking for JSON multiplies the cost of a page by five.

• A basic scrape, crawl or map costs 1 credit per page.
• The JSON, Question and Highlight formats add 4 credits a page.
• Search costs 2 credits per 10 results.
• A 403 or 404 page "is still returned to you and costs 1 credit"; a scrape that returns no result is not charged.

So the free 1,000 credits a month buy about 200 pages of JSON. Credits roll over only on Scale and Enterprise. On paid plans you can add pay-as-you-go credits in $5 increments, up to a monthly limit you set.

One disclosure: Gravity's own catalog uses Firecrawl. In October 2026, three of its nine agents, among them Pain point discovery and Job scout, list Firecrawl as the tool they use to search the web and read pages.

Best for

Builders feeding web pages to an LLM or an agent.

Pricing

Free 1,000 credits a month; Hobby $16 a month billed annually or $19 monthly (5,000 credits); Standard $83 or $99 (100,000); Growth $333 or $399 (500,000). As of October 2026.

Free tier

1,000 credits a month, no card, 2 concurrent requests.

7. Bright Data Scraper Studio: best for hard sites at volume

Scraper Studio is the AI scraper builder on Bright Data's scraping network. You choose a website, "describe the data in plain English, and the AI instantly generates a ready-to-run scraper", which appears in an IDE workspace for testing, running and edits. The product page says "one-click Self-Healing" keeps scrapers stable, with "Proxies, browsers, and unblocking included."

The AI writes the scraper, and someone still has to run it.

Bright Data's own FAQ says so. "Coding is not required to generate a scraper," it starts, then adds: "You will still need a working knowledge of web-scraping concepts to configure and use the scraper." For teams with nobody technical, Bright Data pitches a managed option, "Ideal for non-technical teams": "We build, maintain, and monitor the scraper for you."

Billing is per page load. The first render of a page is one result; clicking through to a new page, or scrolling to load more data (a "lazy load"), is a second. Results come out as JSON, NDJSON, CSV or Microsoft Excel. Scraper Studio's prices on 17 October 2026:

• Free tier: 5,000 page loads a month, no card.
• Pay as you go: $1.50 per 1,000 page loads, "Pay only for success", with monthly spend limits you set.
• Scale: $499 a month for 383,000 page loads, then $1.3 per 1,000 extra.
• Enterprise: custom.

Two numbers on the site are easy to misread. The pricing hub says Scraper Studio "Starts from $1/1k", which is the rate at the $1,999-a-month step of the Scale slider; the entry rate is $1.50. The page also advertised a first-deposit match of up to $500 that day. Treat that as a promotion.

Disclosure: since October 2026 the research scripts behind this site's data trackers and blog research call two Bright Data products, the SERP API for Google results and the Web Unlocker for pages that block an ordinary request, with a cap of 4,000 requests a month in our own code. I have not used Scraper Studio, the AI scraper builder priced here.

If you start on pay as you go, set the monthly spend limit before the first run.

Is AI web scraping legal?

This is not legal advice. Three primary documents each answer a different part of the question:

• RFC 9309, the robots.txt standard (IETF, September 2022). Robots.txt is a site's published request to crawlers. The standard says its rules "are not a form of access authorization", and that a crawler which downloads the file "MUST follow the parseable rules".
• hiQ Labs v. LinkedIn (Ninth Circuit, April 18, 2022). In a preliminary-injunction ruling, the court held that the federal Computer Fraud and Abuse Act's "without authorization" concept "does not apply to public websites". The same opinion says other claims, "such as copyright infringement, misappropriation, unjust enrichment, conversion, breach of contract, or breach of privacy, may also lie." The later history of the case was not reviewed for this guide.
• GDPR, Articles 4(1) and 14. Any information about an identifiable person is personal data, and the identifiers named include "a name" and "an online identifier". Article 14 lists what a controller must tell people "Where personal data have not been obtained from the data subject".

A sheet of competitor prices is not about a natural person, and the Regulation's Recital 14 leaves out data about companies as legal persons. A sheet of named people with emails or profile links is personal data. Before that kind of list goes anywhere, strip the personal data you do not need.

Apify's terms put the duty on you as well: section 6.2 allows processing only data "you are authorized to access". Some sites say no in writing: one research tool's pricing page, quoted in this guide to AI market research tools, forbids bots and scrapers from extracting its results. Sites have their reasons to block bots. Imperva, which sells bot protection, says in its 2026 Bad Bot Report that automated traffic made up "more than 53% of all web traffic in 2025".

Which one should you choose?

I would pick by the job first, then by who on your side will do the work.

• Watch competitor prices or listings for changes: Browse AI. Its free 50 credits cover about one page checked daily, and an agent that tracks competitors explains the setup around it.
• A table from the page in front of you, today: Thunderbit.
• Big recurring jobs on Amazon or Google Maps: Octoparse. Its templates and cloud runs start at $69 a month billed yearly as of October 2026, and the free plan runs tasks on a computer you leave on.
• Scraping is one step in a team's AI workflow: Gumloop, once you have budgeted for the 8% fee.
• Somebody already built the scraper for that site: Apify. Search the store before anyone builds anything.
• You are building an app or an agent: Firecrawl.
• Heavily protected sites at volume: Bright Data Scraper Studio, if someone technical will run it or you can pay for the managed option.

Lead lists deserve their own warning, for the personal-data reasons in the legal section. Read lead generation without bought or scraped lists first, and clean the list before you email anyone.

A weekly competitive briefing shows what that note should contain, and how to tell a broken fetch from a quiet week.

FAQ

Sources