AI Web Data Extractor

Give it a web page and say what you want in plain English. You get clean JSON back: one row per item on a list page, one record for a single page, or a direct answer to a question. No code, no selectors.

From $0.007 per page. Failed pages are free.

What is AI Web Data Extractor?

It reads a public web page the way a person would, and returns the fields you name as columns. Ask for name, price and rating of every product and you get a table with exactly those three columns, a row for each product, across the next pages too. Prices and ratings arrive as numbers, ready for a spreadsheet, a database or an AI agent.

It works on shops, job boards, directories, news sites, tables and JavaScript-built pages, in any language.

What can you use it for?

• Price monitoring: products and prices from a shop's category or search pages, on a schedule.
• Product catalogues: open every product page behind a list and collect brand, SKU, description and images.
• Lead lists: turn a directory into company names, websites, phones and addresses.
• Job data: titles, companies, locations and salaries from job boards and careers pages.
• News monitoring: headlines, authors, dates and links from news sites and blogs.
• Research datasets: tables, rankings and statistics as rows.
• Facts for AI agents: an agent asks a site a question and gets a short, sourced answer instead of a whole page.

Sites it reads well

Tested on 115 public websites it had never seen, in eight languages (15 October 2026): 83% returned usable data, 68% with every field correct. Pages that failed were free.

Kind of site	Example ask that works
Shops and catalogues	name, price and rating of every product
News front pages	headline, author, date and link of every article
Job boards and careers pages	title, team and location of every job
Tables and rankings (Wikipedia, sports, finance)	country, population and area of every country in the table
A single product, film, recipe, package or repo page	name, latest version, downloads and license
Directories, forums and event listings	name, date and group of every event
Pricing and plan questions	what does the Team plan cost per user per month?
Government and science data	date, location and magnitude of every earthquake

How to use it

1. Paste one or more URLs.
2. Under What do you want?, name the fields, for example name, price and rating of every product.
3. Click Start. Your data appears in the Output tab.

Want each item's own page too? Put the list page in URLs and write every product link in Follow links. You get one full record per item page.

Tip: start with one URL and Max pages per URL set to 1, check the fields, then scale up.

Examples to copy

You want	URLs	Follow links	What do you want?
Prices from a shop	A category page	—	name, price, currency and stock status of every product
A full product catalogue	Category pages	every product link	name, brand, price, SKU, description and image URL
Job listings	Job search results	every job posting	title, company, location, salary and full description
A lead list	A directory page	every company profile	company name, website, phone and address
Headlines	A news home page	—	headline, author, date and link of every article
An answer about a company	Its home page	—	does this company offer a free plan?
Reviews	A product's review page	—	reviewer, rating as a number, date and text of every review

What you get

Three shapes, picked from your wording. Every example below is a real run, from 15 October 2026.

Rows: title, price and star rating of every book on https://books.toscrape.com/

sourceUrl	title	price	star_rating	currency
https://books.toscrape.com/	A Light in the Attic	51.77	3	GBP
https://books.toscrape.com/	Tipping the Velvet	53.74	1	GBP
https://books.toscrape.com/	Soumission	50.10	1	GBP

20 rows from 1 page, for $0.011.

Rows from a table: city, country and population of every city in the table on Wikipedia's list of largest cities

city	country	population
Jakarta	Indonesia	41,913,860
Dhaka	Bangladesh	36,585,479
Tokyo	Japan	33,412,512

Record: crate name, latest version, downloads and license on https://crates.io/crates/serde

{
  "sourceUrl": "https://crates.io/crates/serde",
  "crate_name": "serde",
  "latest_version": "1.0.229",
  "downloads": 1531313118,
  "license": "MIT OR Apache-2.0"
}

Reply: what does the Team plan cost per user per month? on https://github.com/pricing. It reads the pages it needs and says which ones.

{
  "sourceUrl": "https://github.com/pricing",
  "plan_name": "Team",
  "price": 4,
  "currency": "USD",
  "billing_period": "month",
  "sources": ["https://github.com/pricing", "https://github.com/team"]
}

A value the page doesn't show comes back empty. Download as JSON, CSV or Excel.

How much does it cost?

You pay per page read, plus $0.001 per run. Nothing else.

Your Apify plan	Per page	Per 1,000 pages
Free, Starter	$0.010	$10
Scale	$0.008	$8
Business and above	$0.007	$7

• Failed pages are free.
• Set Max total charge on a run and it never spends more.
• Apify's free plan includes $5 of credit a month, about 500 pages.

For example, 50 shop category pages with 40 products each is 2,000 products for about $0.50 on the Free plan.

Use it with AI agents and your tools

• AI agents: add it to Claude, Cursor or any MCP client through Apify MCP. The agent writes the request and gets JSON back.
• LangChain and LlamaIndex: call it through Apify's integration for each.
• No-code: Make, Zapier and n8n through their Apify apps; send results on to Google Sheets, Slack or a webhook.
• Schedules: run it daily or hourly with Apify Schedules to watch prices or listings.

Or call it from code:

curl -X POST "https://api.apify.com/v2/acts/datacortex~ai-web-data-extractor/run-sync-get-dataset-items?token=YOUR_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{"urls":["https://books.toscrape.com/"],"ask":"title, price and star rating of every book","maxPagesPerUrl":1}'

from apify_client import ApifyClient

client = ApifyClient("YOUR_TOKEN")
run = client.actor("datacortex/ai-web-data-extractor").call(run_input={
    "urls": ["https://books.toscrape.com/"],
    "ask": "title, price and star rating of every book",
})
for row in client.dataset(run["defaultDatasetId"]).iterate_items():
    print(row)

Get the best results

• Name the fields and say "every": name, price and rating of every product, not get the products.
• Use the words the page uses: "SKU", "Availability", "Points". Describe content, not looks: the price after discount, not the red number.
• Point it at the right page: a list or search page for rows, an item's own page for a record, a home or pricing page for a question.
• For tables, say which one: every row in the main table or every city in the table.
• Limits and filters work: the top 10 stories, every book under £20.
• Ask questions as questions: does it have a free plan, and what does it include? The answer cites the pages it used.
• The first run on a new site takes longer (often 30–120 seconds) while it learns the page. Later runs on the same site and fields are faster.
• Say the format if it matters: date as YYYY-MM-DD, price as a number.

FAQ

Do I need CSS selectors or code? No. Describe the fields in plain English and it finds them on the page.

Does it work on JavaScript-heavy pages? Yes. Pages are fully loaded before they are read.

Does it follow pagination? Yes, for rows. It reads next pages up to Max pages per URL (10 by default).

How is this different from a website crawler? A crawler gives you each page's full text. This gives you the specific fields you name, as columns. Use a crawler to feed whole pages to a model, and this when you want the data in a table.

What if a site turns visitors away? Turn on Residential network and run it again. It costs the same per page.

Can it read pages behind a login? No, public pages only. Sites whose terms forbid scraping (such as LinkedIn, Yelp and Zillow) are declined.

Is it legal? Extracting public, non-personal data is generally legal. Personal data falls under laws like the GDPR, so only collect it with a legitimate reason. See Apify's guide to the legality of web scraping.

If something goes wrong

A URL that fails writes one row explaining what happened, and is not charged.

• Missing fields: use the words the page uses, like "SKU" or "Availability".
• Too few rows: raise Max pages per URL.
• Follow links found nothing: describe the links the way the page shows them, like every "View details" link.

Help

Something not working, or a site you need? Open an issue on the Issues tab. We read every one.