> ## Documentation Index
> Fetch the complete documentation index at: https://docs.geekflare.com/llms.txt
> Use this file to discover all available pages before exploring further.

# Web Scraping Output Format

> Available output format types in Geekflare Scraping API.

[Geekflare Web Scraping API](https://geekflare.com/api/webscraping/) supports 6 output formats via the `format` parameter.

Whether you are archiving full web pages, extracting structured data, or feeding Retrieval-Augmented Generation (RAG) pipelines, you can request the exact data structure you need.

### The `-llm` Optimized Formats

For AI developers, we offer specialized `-llm` formats. These formats run our semantic cleaning engine before returning the response. They automatically strip out boilerplate, navigation bars, footers, sidebars, advertisements, and hidden elements, returning **only the primary content** of the page.

Using `-llm` formats dramatically reduces noise, improves LLM inference accuracy, and saves massive amounts of tokens.

## Format Reference

You can pass one of the following string values into the `format` parameter of your API request.

| Format         | Description                                                                                                         | Best For                                                        | Avg. Token Savings\*  |
| :------------- | :------------------------------------------------------------------------------------------------------------------ | :-------------------------------------------------------------- | :-------------------- |
| `markdown-llm` | **(Recommended for AI)** The primary content of the page, converted to clean Markdown.                              | RAG pipelines, AI Agents, LLM.                                  | **\~75%** vs raw HTML |
| `text-llm`     | The primary content of the page as raw text. Strips all HTML tags, Markdown formatting, and structural data.        | Vector embeddings, traditional NLP, maximum token efficiency.   | **\~85%** vs raw HTML |
| `html-llm`     | The primary content of the page in HTML format. Strips out all `<script>`, `<style>`, `<nav>`, and `<footer>` tags. | AI applications requiring DOM structure, semantic HTML parsing. | **\~60%** vs raw HTML |
| `markdown`     | The entire rendered page converted to Markdown, including navigation links, sidebars, and footer text.              | Full-page archiving, layout analysis.                           | **\~60%** vs raw HTML |
| `text`         | The entire rendered page as raw text. Contains no HTML tags, but includes all boilerplate text.                     | Keyword density checks, regex matching.                         | **\~70%** vs raw HTML |
| `html`         | The raw, unmodified HTML DOM of the page as it appears in the browser.                                              | Traditional web scraping, republishing.                         | -                     |
| `json`         | Structured JSON output alongside the DOM.                                                                           | Data analysts, structured databases.                            | -                     |

*\*Token savings are estimates based on average news articles and blog posts compared to processing the raw HTML DOM.*

## Example Usage

To request an LLM-optimized Markdown response, simply set the `format` parameter in your JSON payload.

**API request to get Markdown LLM format:**

```json theme={null}
POST https://api.geekflare.com/webscraping

{
  "url": "https://example.com/long-article",
  "format": "markdown-llm"
}
```

## Choosing the Right Format for AI

If you are building an AI application, choosing between the `-llm` formats depends on your specific pipeline:

1. **Use `markdown-llm`** when you need to chunk data for a Vector Database. Markdown preserves `## Headings`, which chunking algorithms use to keep contextual ideas together. It also preserves data tables.
2. **Use `text-llm`** when you are doing massive batch processing and need the absolute lowest token count possible, or when simply generating mathematical vector embeddings where structural tags are unnecessary.
3. **Use `html-llm`** when you are passing data to an LLM that has been specifically fine-tuned to read DOM structures and CSS classes.