RAG Web Search: Google Results as Clean Markdown

Type a search query and get the top Google result pages as clean Markdown, plain text or HTML, one Dataset row per page. It is a RAG web search tool for LLM apps and AI agents: no login and no API key, and a typical run with 3 pages takes a few seconds and costs about $0.009.

Start with the prefilled query apify actors and 3 pages. Pages that block robots, ask for a login or sit behind a paywall are not bypassed: they come back as free status rows, and the Actor reads the next Google result instead.

Table of contents

• What it does
• Quick start
• Use cases
• Input
• Output
• Pricing and limits
• API, export and integrations
• Troubleshooting
• FAQ
• Related Actors
• Support

What it does

1. Searches Google for your query in the country you choose.
2. Takes the organic results in order. Ads and Google's own pages are skipped. URLs are the real page addresses, not Google redirect links.
3. Opens each page with a normal HTTP request and keeps only the main content: no menus, footers, cookie banners or scripts.
4. Saves each readable page as one row with the content in one field: markdown, text or html.
5. Stops when it has the number of readable pages you asked for, or when the time budget ends.

If you put a URL instead of a query, the Actor skips Google and reads only that page.

Good fit

• Give an LLM or agent fresh web context for a question (retrieval-augmented generation).
• Collect the readable text of the top results for a topic before chunking and embedding.
• Read one known URL as Markdown inside the same workflow.

Not a fit / limits

• No JavaScript rendering. Pages that build their text in the browser (many web apps, YouTube) come back as empty.
• No login, cookies or paywall bypass. Such pages come back as login_required or paywall.
• Up to 10 pages per run and English-language Google results. For more queries, start more runs.
• PDF and other files are not converted (not_html row).

Quick start

1. Open the Input tab.
2. Keep the prefilled Search query or URL (query) apify actors, or type your own question.
3. Keep Pages to return (maxResults) at the prefilled 3.
4. Click Start.
5. Open the Output tab. The Results view shows rank, URL, title, word count and status; the Markdown view shows the page content.

Expected result: 3 rows with fetchStatus: "ok" and Markdown content, usually in under 10 seconds, plus free status rows for any pages that could not be read.

Use cases

Buyer job	Input path	Output entity	Key fields	Next integration
Web context for an LLM answer	query = the user's question, maxResults 3–5	one row per page	url, title, markdown	pass markdown to the model with url as the source
Topic corpus for a vector database	query per topic, maxResults 10, outputFormat text	one row per page	text, wordCount, url	chunk text, keep url as metadata
Read one known page	query = a single https:// URL	one row	markdown, httpStatus, fetchStatus	same as above

Input

Input fields

Field	Plain name	Type	Required	Default / prefill	Allowed values	Effect	Example	Recommendation
query	Search query or URL	string	yes	prefill apify actors	1–500 characters; Google operators work (site:, quotes, -word); or one http(s) URL	what Google searches, or the single page to read	rag chunking strategies	write it like a normal Google search
maxResults	Pages to return	integer	no	default 3, prefill 3	1–10	how many readable pages you get; status rows do not count	5	3–5 for chat answers, 10 for a corpus
outputFormat	Content format	string	no	markdown	markdown, text, html	which content field is filled	text	Markdown keeps headings and links; text is smallest
countryCode	Google country	string	no	us	two letters (gl), e.g. us, gb, de	which country's results Google shows; language stays English	gb	use the country of your users
timeoutSecs	Time budget (seconds)	integer	no	40	10–120	the Actor stops reading pages after this time and returns what it has	60	raise it if you often get TIME_LIMIT

Interacting settings

When	Input A	Input B	Effective behavior	Example	Cost/result effect
Some top pages are blocked	maxResults	—	the Actor reads further Google results (up to 2 result pages) to fill the number	maxResults: 3, 3 blocked → reads results 4–6	status rows are free
Slow sites	maxResults	timeoutSecs	when time runs out the run ends as partial with the rows it already has	maxResults: 10, timeoutSecs: 10 → 6 rows	you pay only for saved rows
query is a URL	query	countryCode	Google is not used; countryCode is ignored; maxResults is 1	https://apify.com/actors	one search event + one page event

Input examples

Answer a question with web context

{
  "query": "rag chunking strategies",
  "maxResults": 3,
  "outputFormat": "markdown"
}

Expected behavior: one Google search, then the top readable pages. Expected output: 3 rows with rank, url, title, markdown. Limit or cost note: 1 search + 3 page events.

Plain text from the UK results

{
  "query": "how to make cold brew",
  "maxResults": 5,
  "outputFormat": "text",
  "countryCode": "gb"
}

Expected behavior: Google UK results; blocked pages are skipped and the next results are read. Expected output: up to 5 ok rows with text, plus free status rows for pages that could not be read.

Read one page

{
  "query": "https://apify.com/actors",
  "outputFormat": "html"
}

Expected behavior: no Google search; the Actor reads this page only. Expected output: 1 row with cleaned html.

Output

Output fields

Field	Meaning	Type	Can be empty?	Example / enum	Source / notes
query	your search query (or the URL)	string	no	apify actors	input
rank	position among Google organic results, 1 = top	integer	no	1	Google; ads are not counted
url	real page address	string	no	https://apify.com/actors	Google redirect and tracking parameters removed
title	page title, or the Google title if the page was not read	string	yes	Automate anything with Apify Actors · Apify	page <title>
description	Google snippet	string	yes	Actors are web data automations…	Google; null in URL mode
markdown	main content as Markdown	string	yes	# Automate anything…	only when outputFormat is markdown; null on status rows
text	main content as plain text	string	yes	Automate anything…	only when outputFormat is text
html	cleaned main-content HTML	string	yes	<main><h1>…	only when outputFormat is html
wordCount	words in the content	integer	no	427	0 on status rows
fetchStatus	result of reading the page	string	no	ok, blocked, login_required, paywall, not_html, empty, timeout, error	only ok rows are charged
httpStatus	HTTP status of the page	integer	yes	200	null when no response
error	why there is no content	string	yes	HTTP 403	null on ok rows
fetchedAt	when the page was read	string (ISO 8601)	no	2026-10-15T04:12:15.881Z	—

Only one of markdown, text or html is present in a row, the one you chose, so only the matching view (Markdown, Text or HTML) shows content. Content is cut at 200,000 characters.

Output entities and surfaces

Situation	Dataset/API/KVS/live view	Result shape	User action
Normal item	Dataset, views Results, Markdown, Text, HTML	ok rows first, ordered by rank	use the content field
Domain-empty result	Dataset empty; RUN_SUMMARY status not_found	no rows	change the query
Partial result	Dataset ok rows + status rows; RUN_SUMMARY status partial with partialReason (TIME_LIMIT, CHARGE_LIMIT, NOT_ENOUGH_READABLE_PAGES)	fewer ok rows than asked	raise timeoutSecs or your spending limit, or rephrase
Technical failure	run FAILED, status message and RUN_SUMMARY status error	no rows, nothing charged	start the run again later

RUN_SUMMARY in the default key-value store also has requested, returned, serpRoute, statusCounts and chargedEvents.

Output example

From a real run with the prefilled input (content trimmed):

{
  "query": "apify actors",
  "rank": 1,
  "url": "https://apify.com/actors",
  "title": "Automate anything with Apify Actors · Apify",
  "description": "Actors are web data automations that power your AI and operations. Run them in the cloud or locally to scrape websites, process data, or connect APIs, all ...",
  "markdown": "# Automate anything with Apify Actors\n\nActors are web data automations that power your AI and operations. …",
  "wordCount": 427,
  "fetchStatus": "ok",
  "httpStatus": 200,
  "error": null,
  "fetchedAt": "2026-10-15T04:12:15.881Z"
}

A status row has the same fields with no content field, for example "fetchStatus": "blocked", "httpStatus": 403, "wordCount": 0.

Pricing and limits

Pricing unit

The Actor uses pay-per-event pricing. You pay only for these events; platform usage is included.

Event	When it is charged	Free plan	Bronze	Silver	Gold+
search	once per run that gets an answer: one Google search (also when Google has no results) or one URL read	$0.003	$0.00285	$0.0027	$0.00255
page	per saved row with fetchStatus: "ok"	$0.002	$0.0019	$0.0018	$0.0017

Status rows (blocked, login_required, paywall, not_html, empty, timeout, error) are free. If Google cannot be reached, the run fails and nothing is charged. If your spending limit is lower than the next row, the run stops as partial and keeps only the paid rows.

Cost examples

Free-plan prices:

Scenario	Input scope	Expected events	Base charge	Optional charge	Stop/cap behavior
Typical	maxResults 3	1 search + 3 pages	$0.009	none	stops at 3 ok rows
One page	maxResults 1, or one URL	1 search + 1 page	$0.005	none	—
Largest	maxResults 10	1 search + 10 pages	$0.023	none	stops at 10 ok rows
No Google results	any	1 search	$0.003	none	run ends not_found
All pages blocked	any	1 search	$0.003	none	run ends partial, status rows only
URL could not be read	one URL	1 search	$0.003	none	run ends partial, one status row

Limits

• 1–10 readable pages per run; Google results from up to 2 result pages are tried.
• One page request waits up to 15 seconds; the whole run follows timeoutSecs (10–120).
• Page size up to 5 MB; content up to 200,000 characters.
• 256 MB memory is enough; measured runs used under 60 MB.

API, export and integrations

• API: start the Actor with the Apify API and read the Dataset items; each item is one page.
• Export: the Dataset downloads as JSON, CSV, Excel or HTML from the Output tab.
• SDK: call it from the Apify Python or JavaScript client and pass the markdown field to your RAG pipeline (for example LangChain or LlamaIndex loaders).

Use with AI agents (MCP)

• MCP server URL with only this Actor: https://mcp.apify.com?tools=cliqtomedia/rag-web-search (see Apify MCP server).
• Smallest useful input:

{ "query": "apify actors", "maxResults": 3 }

• Key output fields: url (source to cite), title, markdown (page content), fetchStatus (ok = usable), rank.
• A typical call returns 3 pages in about 1–7 seconds and costs about $0.009 on the Free plan.
• Limits: 10 pages per call, no JavaScript rendering, logins and paywalls are not opened.

Troubleshooting

Fewer pages than I asked for

Symptom: the run ends as partial with fewer ok rows. Likely reason: many top results blocked the request or needed JavaScript, or the time budget ended. Check: partialReason and statusCounts in RUN_SUMMARY. Action: for TIME_LIMIT raise timeoutSecs; for NOT_ENOUGH_READABLE_PAGES rephrase the query or add -site: for sites that block.

A known site is always blocked

Symptom: the same domain comes back as blocked with HTTP 403. Likely reason: the site refuses automated requests. The Actor does not solve CAPTCHAs or bypass anti-bot checks. Action: use other results, or a source that allows automated access.

The run failed

Symptom: run status FAILED, message about Google. Likely reason: Google did not return a readable results page on any route. Action: start the run again in a few minutes. Nothing was charged.

FAQ

Do I need a Google API key or a login?

No. The Actor reads public Google results and public pages. It does not open pages that need a login.

What happens when Google finds nothing?

The run succeeds with status not_found, no rows, and one search event ($0.003 on the Free plan).

Is the content complete and up to date?

It is the page as served at fetchedAt, main content only. Text that loads with JavaScript, comments and hidden blocks are not included. Some sites repeat a lead paragraph in their HTML; it then appears twice.

Can I search in other languages?

Google results are requested in English (hl=en); countryCode changes the country. Pages in other languages are returned as they are.

Related Actors

• PDF & Image to Text OCR — for results that come back as not_html PDF files.
• Website Metadata & Open Graph Scraper — titles, descriptions and preview images for a list of URLs.

Support

For a bug, question or feature request, open an issue on the Issues tab of this Actor. Include the run ID, your input (without private data) and the RUN_SUMMARY message. Never send API tokens or cookies.