What is web data extraction?

Web data extraction is the second half of scraping. Fetching gets you a page. Extraction turns that page into records: a product with a name, a price and a stock status, rather than four hundred kilobytes of markup that happens to contain them.

The two halves fail in different ways and are worth keeping separate in your head. Fetching fails loudly, with a status code or a block page. Extraction fails quietly, returning empty fields that flow downstream and get noticed weeks later.

The approaches, and when each fits

Selectors. You write a CSS selector or XPath pointing at where the value sits in the document. Exact, fast, free to run, and tied to a structure that will change.

Regular expressions. Pattern matching over raw text. Genuinely useful for well-formed values inside a string, a price, a date, an identifier. A poor fit for navigating HTML structure, which nests in ways regular expressions handle badly.

Reading the underlying data. Many pages fetch their content from an internal JSON endpoint, or embed it in a script tag as a state blob. Finding that is often the highest-leverage move available: the data arrives already structured, and it changes far less often than the markup around it.

Schema-based and model-driven. You describe the fields you want and let the extractor locate them. Survives redesigns and unfamiliar sites, costs more per page, and can be wrong in ways a selector cannot.

The failure mode worth designing against

A selector that no longer matches returns nothing. It does not raise. So a scraper whose target site was redesigned keeps running, keeps reporting success, and writes nulls until somebody looks at the data.

The defence is asserting on what you expect rather than only on what you got: a product record without a price is a failure, not a product. Checking the shape of the output, not just the status of the request, is what turns a silent problem into a loud one.

Where Zenrows fits

Extraction happens at fetch time rather than as a separate parsing stage. Extract returns structured JSON from a page in a single request, building a tailored extraction per site and keeping it working when the layout changes. CSS Extractor takes your own CSS or XPath and returns JSON when you want exact control, and Autoparse handles common page types such as products and articles without any selectors at all.

Returning fields rather than pages also keeps responses small, which matters when a page is large enough to hit a 413 or a model's context limit.