> ## Documentation Index
> Fetch the complete documentation index at: https://docs.reader.dev/llms.txt
> Use this file to discover all available pages before exploring further.
# Formats and extraction
> What Reader returns, how to shape it, and how to trim boilerplate.
Reader turns HTML into something an LLM (or your code) can actually read. You control two things: the **format** Reader gives you back, and how aggressively it **extracts** the meaningful part of the page.
## Formats
Pass `formats` on the request to choose what comes back:
| Value | What you get |
| ---------------------- | --------------------------------------------------------------------------------------------------------------------- |
| `"markdown"` (default) | Clean, structured markdown. Headings, lists, links, code blocks preserved. Best for LLMs and RAG. |
| `"html"` | The cleaned HTML Reader used as the source for markdown conversion. Useful when you need to run your own DOM parsing. |
| `"screenshot"` | A full-page screenshot as a base64-encoded PNG. Useful when AI agents need visual context. |
Every response also includes `rawHtml` - the unprocessed HTML exactly as the browser rendered it, before any cleaning or content extraction. This is always returned regardless of `formats`.
You can request any combination:
```json theme={null}
{
"url": "https://example.com",
"formats": ["markdown", "screenshot"]
}
```
The response includes whichever fields you asked for:
```json theme={null}
{
"data": {
"url": "https://example.com",
"rawHtml": "......",
"markdown": "# Example Domain\n\n...",
"screenshot": "iVBORw0KGgoAAAANSUhEUgAA...",
"metadata": { /* ... */ }
}
}
```
Screenshots are full-page captures. They bypass the cache since each capture is unique. The `screenshot` field is a base64-encoded PNG string that you decode on the client side.
## Main content extraction
By default Reader strips away navigation, footers, sidebars, cookie banners, newsletter pop-ups, and other boilerplate, keeping just the article body. This is the `onlyMainContent: true` default.
Turn it off when you need the whole page:
```json theme={null}
{ "url": "...", "onlyMainContent": false }
```
When to turn it off:
* You want the nav bar's links (e.g., to find related pages)
* You're scraping a landing page where there is no "article"
* You're debugging why something was stripped
When to leave it on (most of the time):
* LLM pipelines: boilerplate is noise and tokens
* RAG indexing: you don't want "Cookie Settings" matching your user's query
* Clean markdown output for humans
## Include and exclude selectors
For finer control, give Reader a list of CSS selectors to keep or drop. These compose with `onlyMainContent`.
```json theme={null}
{
"url": "https://example.com",
"includeTags": ["article", "main.content"],
"excludeTags": [".ads", "#newsletter-modal", "aside"]
}
```
* `includeTags`: keep only content matching these selectors. Everything else is dropped.
* `excludeTags`: drop anything matching these, keep the rest.
If you pass both, `includeTags` runs first, then `excludeTags` trims what remains.
## Wait for a selector
Dynamic pages sometimes render the real content a moment after the initial load. For example, a product grid that's hydrated from JSON. Tell Reader to wait for a specific selector before capturing:
```json theme={null}
{
"url": "https://shop.example.com/search?q=phone",
"waitForSelector": ".product-card"
}
```
Reader returns once that selector appears or the per-request `timeoutMs` is hit, whichever comes first.
## What's in `metadata`
Every scrape result includes metadata about the page and the request:
```json theme={null}
{
"metadata": {
"title": "Example Domain",
"description": "This domain is for illustrative examples.",
"statusCode": 200,
"duration": 487,
"cached": false,
"proxyMode": "standard",
"scrapedAt": "2026-04-04T12:00:00Z"
}
}
```
* `title`, `description`: extracted from ``, `<meta>` tags, or Open Graph data
* `statusCode`: what the target site returned
* `duration`: how long Reader spent on the request, in ms
* `cached`: whether the content was served from Reader's cache
* `proxyMode`: which mode ran (`"standard"` or `"premium"`). See [Proxy modes](/home/concepts/proxy-modes)
* `scrapedAt`: when the content was captured
## Next
* [Proxy modes](/home/concepts/proxy-modes): how Reader decides to fetch a page
* [Caching](/home/concepts/caching): when to reuse a previous result