AI Web Scraper - Verified Structured Data from Any Page

Extract structured data from any web page with AI. Name the fields you need, paste the URLs, and get clean JSON. Every returned value is checked against the page text, so the scraper does not hand you values that are not on the page. No selectors, no code, no prompt engineering.

Why "verified"?

AI scrapers sometimes return values that look plausible but are not on the page: a guessed brand, a rounded price, a placeholder like "Your product's name". This Actor runs a check after extraction: a text or number value is kept only if it actually appears on the page. Anything else is removed from data and listed in removedValues, so you can see exactly what was dropped and why.

Measured on 224 real product and job pages from 224 different websites (October 2026):

Pipeline	Exact-match accuracy	Returned values not found on the page
This Actor	87.8%	0.8%
ScrapeGraphAI (gpt-4o-mini, default pipeline)	85.0%	5.0%
Crawl4AI LLM extraction (gpt-4o-mini, default pipeline)	78.9%	4.3%

Ground truth came from each page's own schema.org data, kept only where the value is visible on the page; all three pipelines were scored with the same independent check. Verification did not cost accuracy: the same model without verification scored 87.4%. Results on your pages will vary.

What you can extract

Anything a person can read on the page: product name, price, currency, SKU, brand, ratings, job title, company, salary, location, event date, contact details, article author, specifications and more. You define the fields:

[
  {"name": "name", "type": "string", "description": "Product name as shown on the page"},
  {"name": "price", "type": "number", "description": "Current selling price as a plain number"},
  {"name": "currency", "type": "string", "description": "ISO 4217 currency code, e.g. USD"},
  {"name": "in_stock", "type": "boolean"},
  {"name": "condition", "enum": ["new", "used", "refurbished"]}
]

Supported types: string, number, integer, boolean, and fixed choices with enum. A plain list of names such as ["title", "company", "salary"] also works. Descriptions are optional but improve accuracy.

How to use it

1. Paste one or more page URLs.
2. Define the fields to extract (or keep the product example).
3. Optional: add extra instructions, for example "Salaries are monthly" or "Use the first listed location".
4. Click Start. Results appear in the dataset as each page finishes, ready to export as JSON, CSV or Excel.

Output example

{
  "url": "https://shop.example.com/products/trail-shoe-x",
  "status": "extracted",
  "pageTitle": "Trail Shoe X - Example Shop",
  "data": {"name": "Trail Shoe X", "price": 89.95, "currency": "USD", "in_stock": true, "condition": null},
  "verification": {"name": "verified", "price": "verified", "currency": "verified", "in_stock": "not_checked", "condition": "empty"},
  "removedValues": null,
  "charged": true
}

verification shows, per field, whether the value was verified on the page, removed, empty, or not_checked (yes/no and fixed-choice fields are interpretations, so they are not checked). Turn verification off with Verify values against the page if you prefer raw AI output.

Pricing

Pay per event, no subscription:

Event	Price
Page extracted	$0.01

1,000 pages cost $10. A page is billed when it was downloaded and processed, even if none of your fields were on it. Not charged: invalid URLs, pages that return errors, pages that block access, and pages with no readable text.

Set a maximum charge for the run and the Actor never exceeds it: once the budget is reached, remaining pages are returned as skipped without being processed.

Use it from the API, integrations and AI agents

from apify_client import ApifyClient

client = ApifyClient("<YOUR_APIFY_TOKEN>")
run = client.actor("<username>/ai-web-scraper-verified").call(run_input={
    "urls": ["https://shop.example.com/products/trail-shoe-x"],
    "fields": ["name", {"name": "price", "type": "number"}, "currency"],
})
for item in client.dataset(run["defaultDatasetId"]).iterate_items():
    print(item["status"], item["data"])

Works with schedules, webhooks, Make, Zapier and n8n. AI agents connected to the Apify MCP server can find and call it.

Limits

• Pages are fetched over plain HTTP. Sites that render their content only with JavaScript return EMPTY_PAGE and are not charged.
• One page at a time per URL: it extracts the main item of each page, not lists of many items.
• Login-only content is not supported.
• Sites that block automated access may fail even through the proxy; failed pages are not charged.

FAQ

Do I need an OpenAI or other AI account?

No. AI usage is included in the price.

What languages work?

Pages in most languages; values are returned as written on the page.

Why is a value null?

Either the page does not contain it, or the AI returned something that was not on the page and verification removed it. Check removedValues to tell the two apart.

Is it legal to scrape these pages?

The Actor only reads publicly accessible pages. You are responsible for complying with each site's terms and the laws that apply to you.