Operation

v32.17.0

Last updated Oct 6, 2026

AI Agent-ready n8n community node for DuckDuckGo search. Search the web, images, news, and videos with no API key required and no outbound telemetry. Optionally fetch and extract the main text of result pages or any URL, and get DuckDuckGo Instant Answers

290 Weekly Downloads

2,759 Monthly Downloads

View on NPM GitHub Repository

Included Nodes

Operation

Description

DuckDuckGo Search Node for n8n

![npm version](https://www.npmjs.com/package/n8n-nodes-duckduckgo-search)
![npm downloads](https://www.npmjs.com/package/n8n-nodes-duckduckgo-search)
![License: MIT](https://opensource.org/licenses/MIT)
![GitHub release](https://github.com/samnodehi/n8n-nodes-duckduckgo/releases/latest)
![Build](https://github.com/samnodehi/n8n-nodes-duckduckgo/actions)

An n8n community node for DuckDuckGo search. Search the web, find images, discover news, and explore videos — no API key required, no outbound telemetry.

> ### If a search comes back with a rate-limit error, that is the node working
>
> DuckDuckGo rate-limits by IP address and answers a client it considers automated
> with a human-verification page — served as HTTP 202, which ordinary HTTP clients
> read as success. Nodes that do not check for it parse that page, find nothing, and
> hand back an empty array, so a block looks exactly like a query with no results.
>
> This node recognises that page and says so, then **backs off locally for about a
> minute** instead of sending requests that would only extend the block. Datacenter and
> shared IPs — n8n Cloud, most VPS hosts, CI — hit it soonest. An empty array from this
> node means no results, and nothing else.
>
> Details, and how to stay under the limit: Bot-detection challenge.

✨ Features

Four search types: Web, Image, News, Video
AI Agent compatible: Works as a tool in n8n AI Agent workflows (usableAsTool: true)
No API key required: Completely free, no registration
No outbound telemetry: No analytics or telemetry data sent anywhere — search requests go only to DuckDuckGo
Structured output: Predictable, consistent field names per operation type
Fallback resilience: News and Video searches fall back to an alternative result path on primary failure, with results clearly labeled
Specific error messages: Named errors for VQD token failures, 403 blocks, and web page parser regressions
Search operators: Advanced query syntax for Web Search (site:, filetype:, intitle:, etc.)
Region/locale support: Configure DuckDuckGo locale codes per operation
Safe search: Configurable per operation
Optional page content extraction: Fetch and extract the main text of result pages (Web and News, opt-in), with optional page metadata
Extract Page Content operation: Give any URL → clean main text + metadata (a “read any page” tool for AI Agents)
Instant Answer operation: Direct answers, abstracts, and definitions from DuckDuckGo’s free Instant Answer API
Search Suggestions operation: Query autocomplete from DuckDuckGo’s suggestion endpoint — the surface least affected by rate limiting
Ranking rules: Boost, downrank or discard results by domain or URL, applied locally to results already fetched — no extra requests, no service

—

🤔 Why this node?

No API key or paid search API required — uses DuckDuckGo’s public search endpoints directly
Works as an n8n AI Agent tool — attach it to any Agent node; no extra setup needed
Four search types in one node — Web, Image, News, and Video from a single, consistent interface
Clean JSON output designed for automation — predictable field names, no noise, easy to wire into downstream nodes
A short result set says why — if the node returns fewer results than you asked for because something stopped it, it warns on the canvas with how many it asked for, how many it got and why, rather than handing back a quietly short list. News and Video fetch further pages when you ask for more than the first page holds (for News, 26–30 results a page in tests), following DuckDuckGo’s own next-page pointer, dropping results DuckDuckGo repeats across pages, and stopping at five pages to spare your IP’s rate limit
Fallback labels for News and Video — when results come from the fallback path, isFallback: true tells you so

—

📦 Installation

Via n8n Interface (recommended)

1. Go to Settings → Community Nodes
2. Enter: n8n-nodes-duckduckgo-search
3. Click Install
4. Restart n8n if required

Via npm

npm install n8n-nodes-duckduckgo-search

—

🚀 Quick Start

1. Add Node: Drag the DuckDuckGo Search node into your workflow
2. Choose Operation: Select Web, Image, News, or Video search
3. Enter Query: Type your search terms
4. Set Limits: Adjust maxResults and safeSearch as needed
5. Execute: Run your workflow

—

🔍 Operations

Web Search

Searches DuckDuckGo and returns organic web results.

Parameters:

Example:

{
  "operation": "search",
  "query": "open source AI models 2025",
  "webSearchOptions": {
    "maxResults": 10,
    "safeSearch": "moderate",
    "region": "us-en"
  }
}

Sample output:

[
  {
    "position": 1,
    "title": "Top Open Source AI Models in 2025",
    "description": "A comprehensive overview of the leading open source AI models...",
    "url": "https://example.com/ai-models-2025",
    "hostname": "example.com",
    "sourceType": "web"
  }
]

> Note: Web Search output does not include snippet (removed; use description instead) or favicon (removed; was always empty).

> Result URLs have advertising click identifiers removed. utm*, fbclid, gclid, msclkid, mccid and the rest of that family are stripped from url on Web, News and Video results, so the same page compares equal between runs and an AI Agent does not paste a tracking string into whatever it writes. Nothing else is touched: the path, the fragment, the order of the remaining parameters and every unrecognised parameter are left exactly as DuckDuckGo returned them, and a URL with no tracking in it is passed through byte-for-byte. Image results are left alone entirely, because a query string on an asset URL is often a CDN signature. See Result URL handling.

—

Image Search

Searches DuckDuckGo images and returns image metadata.

Parameters:

Example:

{
  "operation": "searchImages",
  "imageQuery": "mountain landscape sunset",
  "imageSearchOptions": {
    "maxResults": 10,
    "safeSearch": "moderate"
  }
}

Sample output:

[
  {
    "title": "Mountain Sunset Over the Rockies",
    "imageUrl": "https://example.com/images/mountain-sunset.jpg",
    "thumbnailUrl": "https://tse1.mm.bing.net/th?id=...",
    "url": "https://example.com/photography/mountain-sunset",
    "width": 1920,
    "height": 1080,
    "source": "https://example.com/photography/mountain-sunset",
    "sourceType": "image"
  }
]

> Field note: imageUrl is the direct image URL. url is the source page URL where the image was found. thumbnailUrl is the DuckDuckGo-hosted thumbnail.

—

News Search

Searches DuckDuckGo news results.

Parameters:

Example:

{
  "operation": "searchNews",
  "newsQuery": "renewable energy breakthrough",
  "newsSearchOptions": {
    "maxResults": 10,
    "safeSearch": "moderate",
    "region": "us-en",
    "timePeriod": "w"
  }
}

Sample output:

[
  {
    "title": "New Solar Panel Achieves Record Efficiency",
    "description": "Researchers have announced a breakthrough in photovoltaic efficiency...",
    "url": "https://news.example.com/solar-efficiency",
    "imageUrl": "https://news.example.com/images/solar.jpg",
    "date": "2025-05-15T09:00:00.000Z",
    "relativeTime": "2 hours ago",
    "syndicate": "Tech News Daily",
    "isOld": false,
    "isFallback": false,
    "sourceType": "news"
  }
]

> Field note: syndicate is the news source/publisher name. When the fallback path is used, syndicate will be "DuckDuckGo Fallback" and isFallback will be true. See Fallback Behavior.

—

Video Search

Searches DuckDuckGo video results.

Parameters:

Example:

{
  "operation": "searchVideos",
  "videoQuery": "machine learning tutorial beginners",
  "videoSearchOptions": {
    "maxResults": 10,
    "safeSearch": "moderate"
  }
}

Sample output:

[
  {
    "title": "Machine Learning Full Course for Beginners",
    "description": "A complete introduction to machine learning concepts...",
    "url": "https://www.youtube.com/watch?v=...",
    "imageUrl": "https://i.ytimg.com/vi/.../hqdefault.jpg",
    "duration": "2:14:30",
    "published": "2024-11-10T00:00:00.000Z",
    "publishedOn": "YouTube",
    "publisher": "FreeCodeCamp",
    "viewCount": "2100000",
    "isFallback": false,
    "sourceType": "video"
  }
]

> Field note: When the fallback path is used, publisher will be "DuckDuckGo Fallback" and isFallback will be true. See Fallback Behavior.

—

Extract Page Content

Fetches any URL (not a search) and extracts its main readable text — useful when an AI Agent already has a URL and needs to read the page.

Parameters:

The node reads at most the first 2 MB of a page’s HTML. A longer page is read up to there – where the article text normally starts – and the result has pageContentTruncated: true. If those 2 MB hold no readable text, content is empty and the error field says No readable text in the first 2 MB of the page.

Sample output:

{
  "url": "https://example.com/article",
  "content": "Extracted main text of the page…",
  "sourceType": "pageContent",
  "title": "Article Title",
  "siteName": "Example"
}

> ⚠️ This operation fetches a third-party URL directly (not DuckDuckGo). See Privacy & Security.

—

Instant Answer

Returns DuckDuckGo’s Instant Answer for a query — a direct answer, a Wikipedia-style abstract, a definition, and related topics — via the official free, no-key API.

Parameters:

Sample output:

{
  "query": "DuckDuckGo",
  "heading": "DuckDuckGo",
  "abstract": "DuckDuckGo is an internet search engine that emphasizes protecting searchers' privacy…",
  "abstractSource": "Wikipedia",
  "abstractURL": "https://en.wikipedia.org/wiki/DuckDuckGo",
  "relatedTopics": [
    { "text": "…", "url": "https://duckduckgo.com/…" }
  ],
  "type": "A",
  "sourceType": "instantAnswer"
}

> Note: Instant Answers exist only for some queries (entities, definitions, calculations). For general queries the fields may be empty — use Web Search instead.

—

Search Suggestions

Returns DuckDuckGo’s autocomplete suggestions for a partial query — what the search box would offer as you type. Useful for query expansion, keyword research, and giving an AI Agent alternative phrasings before it commits to a search.

This uses DuckDuckGo’s suggestion endpoint, which needs no VQD token and returns a very small payload. In endpoint testing it was the surface least affected by rate limiting, so it often still answers when Web Search is temporarily blocked.

Parameters:

Example:

{
  "operation": "autocomplete",
  "autocompleteQuery": "how to build",
  "autocompleteOptions": {
    "maxResults": 5,
    "region": "us-en"
  }
}

Sample output (default):

{
  "query": "how to build",
  "suggestions": [
    "how to build a pc",
    "how to build credit fast",
    "how to build a website",
    "how to build an ai agent"
  ],
  "count": 4,
  "sourceType": "autocomplete"
}

Sample output (splitIntoItems: true) — one item per suggestion, ready to fan out into a loop:

[
  { "query": "how to build", "suggestion": "how to build a pc", "position": 1, "sourceType": "autocomplete" },
  { "query": "how to build", "suggestion": "how to build a website", "position": 2, "sourceType": "autocomplete" }
]

—

⚙️ Configuration Reference

Common Parameters (Web, Image, News and Video search)

Operation-specific parameters

Cache Settings

Available on all operations via the Cache Settings collection:

Cache is in-memory only and is not shared across n8n worker processes or restarts. It is scoped to the running n8n process.

—

📋 Output Field Reference

Web Search

Image Search

News Search

Video Search

—

🎯 Ranking Rules

Search engines rank for everybody. A workflow usually wants something narrower —
only documentation sites, never the content farms that copy them, this vendor’s
own pages before the aggregators.

Ranking Rules does that locally, on results the node has already fetched.
Available on Web, News and Video Search under Options → Ranking Rules.

How it behaves

Rules are checked in order and the first match wins. Put specific rules

above broad ones: a Boost on docs.example.com written above a Discard on
example.com keeps the documentation and drops the rest of the site. No fixed
precedence between effects could express that.

Ordering is three buckets, not a score. Boosted results keep their order

among themselves and move above everything else, downranked ones likewise
below, and anything unmatched stays exactly where DuckDuckGo put it. There is
no weight to tune.

Rules run before Maximum Results, which is the reason to do this here

rather than in a Filter node afterwards — there, discarding two of ten leaves
you with eight. How much that buys depends on the operation: Web Search
fetches everything the request returned and cuts afterwards, so a discard is
backfilled from further down and you still get the number you asked for.
News and Video fetch pages until they hold about as many results as you
asked for, so a discard can leave you short; raise Maximum Results to
compensate, bearing in mind that each extra page is one more request.

position is renumbered to match the order you actually receive, on the operations that carry one (Web Search).
A result the rules cannot judge is kept. A result with no URL, or one that

does not parse, is never discarded by a rule it had no chance to match.

Rules apply to fallback results too, since which site a result came from does

not depend on which path fetched it.

Nothing here contacts anything. No rule causes a request, and none of this is
sent anywhere — it is applied to the list already in memory.

> Two things rules do not apply to: Image Search, which has no ranking
> option, and Return Raw Results, which returns DuckDuckGo’s response before
> the node processes it — for News and Video, the pages collected with repeated
> results removed.

Example

Research that should prefer primary sources and never return aggregators:

—

🔗 Result URL Handling

Two things happen to a result URL before you see it.

DuckDuckGo’s redirect is unwrapped. Links come back as
duckduckgo.com/l/?uddg=; the node extracts and validates the real
destination, so url is the page itself rather than a redirect through
DuckDuckGo.

Advertising click identifiers are removed from url on Web, News and Video
results:

Why: deduplicating results across runs compares URLs, and the same article
carrying a different utm_campaign each time compares as a different page. It
also keeps tracking strings out of whatever an AI Agent writes downstream.

The rule is deliberately narrow, and **every byte that survives is the byte
DuckDuckGo returned**. Only the matching name=value segments are cut out of
the query; the rest is not rebuilt. Rebuilding would re-encode what remains —
?q=a%20b would come back as ?q=a+b, and ~ as %7E — which is a different
URL to a server that signs its query. A URL with nothing to strip is likewise
returned unchanged, so a bare origin does not gain a trailing slash and a host
does not get lowercased. A URL that does not parse, or is not http(s), is
passed through untouched rather than dropped.

Image results are not normalised at all. Their URLs are asset URLs, where a
query string is often a CDN signature or a resize instruction, and stripping
anything risks breaking a working link for no gain.

—

🔄 Fallback Behavior

News Search and Video Search request DuckDuckGo’s news.js and v.js directly, the way DuckDuckGo’s own results page does.

If the primary path fails (e.g. DuckDuckGo returns a server error or no parseable results), the node automatically attempts an alternative HTML-based fallback search.

When fallback results are returned:

isFallback is true on every result item
For News: syndicate is set to "DuckDuckGo Fallback"
For Video: publisher is set to "DuckDuckGo Fallback"

When primary results are returned normally:

isFallback is false on every result item

This allows downstream nodes and AI Agents to detect and handle fallback results differently if needed (e.g. flag lower confidence, re-rank, or discard).

Web Search and Image Search do not use a two-tier fallback; they use a single direct HTML-based search path.

—

⚠️ Error Behavior

The node surfaces specific, actionable error messages rather than generic failures. Where this section says the node throws, the Web, News, Image and Video searches catch that error themselves and return it as an item with success: false and the message in error, rather than failing the execution — see Ready-made workflows for branching on it.

Bot-detection challenge (rate limiting)

DuckDuckGo answers a client it considers automated with a human-verification page instead of results. That page is served with HTTP 202, which ordinary HTTP clients treat as success — so earlier versions parsed it, found no result blocks, and returned an empty array with no error at all.

The node now detects that page and throws:

> DuckDuckGo served a bot-detection challenge instead of results. This blocks the IP address your n8n instance sends requests from, typically for tens of minutes. Wait before retrying, and reduce how frequently this node runs — shared or cloud IP addresses are affected sooner.

What triggers it: roughly fifteen or more requests from the same IP within a few minutes. The block is tied to the IP address — not to your account, query or region — lasts tens of minutes, and is not cleared by retrying or by changing User-Agent. The node therefore marks this error non-retryable: an immediate retry only prolongs the block.

Automatic back-off. Once a challenge is seen, the node refuses further DuckDuckGo requests locally for about a minute instead of sending them, and says how many seconds remain. Every request made during a block is wasted and prolongs it, so this protects the shared IP rather than your workflow’s throughput — requests resume on their own.

If you hit it regularly:

Space executions out, and put a Wait node between iterations of a loop
Reduce Max Results and avoid re-running the same workflow in tight cycles
On n8n Cloud or shared hosting you share an outbound IP with other tenants, so the threshold is reached sooner
Enable Cache Settings so repeated identical queries do not reach DuckDuckGo again
The Search Suggestions operation uses a different, much lighter endpoint and is the least affected by this limit

Image Search: VQD token missing

If DuckDuckGo’s image search page does not return a valid VQD token (a session token required to query image results), the node throws:

> DuckDuckGo image search token (VQD) could not be extracted. Image search may be temporarily unavailable. Please try again later.

No fake or placeholder image URLs are produced. The result item carries the error message.

Image Search: 403 Forbidden

If the image results endpoint (i.js) returns HTTP 403, the node throws:

> DuckDuckGo image search returned 403 Forbidden. The search token (VQD) may have expired or the request was blocked. Please try again.

This typically indicates a VQD token expiry or a temporary block by DuckDuckGo.

Web Search: Parser failure

If directWebSearch receives an HTTP 200 response with a large body but cannot parse any result blocks from it (indicating DuckDuckGo may have changed its HTML structure), the node returns an error item (success: false) with this message in error:

> DuckDuckGo web search response could not be parsed. The page structure may have changed. Please try again later.

This is distinct from a genuine no-results response, which returns an empty array without an error.

Extract Page Content: refused address

Extract Page Content and Fetch Page Content retrieve a URL supplied by the caller — and when the node runs as an AI Agent tool, the agent chooses that URL. The node therefore refuses any address that is not publicly routable:

> Refused to fetch a private, loopback or link-local address
> Refused to fetch a non-HTTP(S) URL (file)

Refused targets include localhost and 127.0.0.0/8, cloud instance metadata (169.254.169.254), the private ranges (10/8, 172.16/12, 192.168/16), IPv6 loopback / link-local / unique-local addresses, and every non-http(s) scheme. Redirects are re-checked at each hop, so a public URL cannot redirect into a private one.

This stops a prompt-injected agent from making your n8n host read internal services — including n8n’s own API — and hand their contents back to the model. Fetching internal addresses is deliberately unsupported.

Empty results

An empty result array ([]) is a valid response when DuckDuckGo genuinely finds no results for the query. This is not an error condition.

Since v32.10.0 an empty array means only that. A bot-detection challenge is reported as the error above instead of being reported as “no results”.

—

🤖 AI Agent Usage

This node supports usableAsTool: true, making it available as a tool inside n8n AI Agent workflows.

Setup:

1. Create an AI Agent node in your workflow
2. Add DuckDuckGo Search as a connected tool
3. Configure the default operation and parameters
4. The agent will invoke the tool autonomously when a search is needed

Notes for AI Agent use:

Output fields are consistent and predictable, suitable for LLM processing
isFallback: true on news/video results signals lower-confidence data to downstream logic
Specific error messages (VQD missing, 403, parser failure) help agents self-diagnose and retry
Some older n8n versions may require the environment variable N8NCOMMUNITYPACKAGESALLOWTOOL_USAGE=true to expose community nodes as agent tools

—

🔍 Search Operators

Advanced query operators are supported for Web Search only. Enable useSearchOperators and the operator string is appended to the query before being sent to DuckDuckGo.

Example:

{
  "operation": "search",
  "query": "machine learning",
  "webSearchOptions": {
    "useSearchOperators": true,
    "searchOperators": "site:arxiv.org filetype:pdf"
  }
}

—

🌍 Supported Regions

Pass a DuckDuckGo locale code as the region parameter. The node does not apply a separate language filter — locale code controls both language and region together. Examples:

Many locale codes are supported. Use wt-wt for no regional bias. Region codes follow the DuckDuckGo kl parameter format.

—

🔀 Migration Notes

If you are upgrading from an older version of this node, read the following.

Removed UI options (silently ignored)

The following UI options were removed because they had no effect on execution or were never implemented. If these fields exist in saved workflow JSON, they are silently ignored at runtime — no workflow changes are required.

| Removed option | Reason |
|—————-|——–|
| useApiKey / API credential | DuckDuckGo has no public Search API; the credential was never read |
| searchBackend | Only the direct HTML backend is used; the selector had no dispatch logic |
| proxySettings | Proxy configuration was exposed in the UI but never applied at runtime |
| reliabilitySettings | The reliability manager was initialised but search calls were never wrapped with retry or circuit-breaker logic |
| Separate searchFilters collection | The UI collection was exposed but never read or applied |
| Web Search timePeriod | The date-filter parameter was not included in the web search request body |
| Image filter options (size, color, type, layout) | The i.js filter parameter is undocumented; smoke testing showed unreliable/silent behavior |
| Video filter options (duration, resolution, publishedTime) | Removed with the reliability and filter cleanup |
| Top-level locale option | Redundant with each operation’s Region option; the per-operation Region (or the wt-wt default) is used instead |

Output field changes

New output fields

Correct field names (if your workflow used old examples)

The following field names were incorrect in previous documentation. Use the names in the right column:

—

🆘 Troubleshooting

Empty results

Try a broader or simpler query
Check region/language settings — some queries return fewer results in non-default locales
Verify safe search settings are not filtering valid results
DuckDuckGo may temporarily return no results for some queries; retry after a short delay
If every query comes back empty after a few successful runs, you are most likely rate-limited — see bot-detection challenge

Search returns a bot-detection challenge error

Your n8n instance’s outbound IP address has been temporarily blocked by DuckDuckGo — see Bot-detection challenge for what triggers it and how to reduce it. Wait before retrying; the block lasts tens of minutes and immediate retries prolong it. This is not a fault in the node or in your query.

Before v32.10.0 this condition returned an empty result list with no error, which is the cause behind most historical “web search returns nothing” reports.

Image search fails with VQD error

DuckDuckGo image search requires a session token (VQD) extracted from an initial page load. If this extraction fails, retry the request. This is a temporary DuckDuckGo availability issue.

Image search fails with 403

The VQD token may have expired or DuckDuckGo has temporarily rate-limited requests from your IP. Wait a short period and retry.

Web search returns parser failure error

DuckDuckGo may have changed its HTML page structure. This means the node’s result extractor cannot parse the page. This is a known fragility of HTML-based scraping. Please open a GitHub issue if this persists.

News/video results marked `isFallback: true`

The primary search path failed and the fallback HTML path was used. Results are real but may have less metadata (e.g. no relative time or thumbnail). If high-quality metadata is critical, retry or filter on isFallback: false.

Configuration errors

Verify the query field is not empty
Ensure region uses a valid locale code (e.g. us-en, not en-US)
maxResults must be between 1 and 100 (1 to 50 for Search Suggestions)

—

📄 Page Content Extraction

By default, Web Search returns DuckDuckGo’s short snippet (the description field). DuckDuckGo controls that snippet length, and the node already returns all of it.

To get more text, enable Fetch Page Content (available on Web Search and News Search, off by default). The node then fetches the actual page behind each of the top results and extracts its main readable text into a pageContent field.

How it works: extraction is three-tiered — (1) Mozilla Readability over a lightweight linkedom DOM pulls the main article text and drops nav/boilerplate; (2) when Readability finds no article, a DOM heuristic removes boilerplate and high link-density blocks (menus not wrapped in

); (3) if DOM parsing fails, a dependency-free regex heuristic is the last resort. The result feeds clean text to downstream nodes or AI agents.

Extraction runs in a worker thread, one page at a time, with a 10-second limit per page that is separate from pageContentTimeout. Some HTML is slow to parse on purpose or by accident – Readability takes seconds to minutes on a few kilobytes of deeply nested elements, and the regex fallback is quadratic on an unclosed

Included Nodes

Description

DuckDuckGo Search Node for n8n

✨ Features

🤔 Why this node?

📦 Installation

Via n8n Interface (recommended)

Via npm

🚀 Quick Start

🔍 Operations

Web Search

Image Search

News Search

Video Search

Extract Page Content

Instant Answer

Search Suggestions

⚙️ Configuration Reference

Common Parameters (Web, Image, News and Video search)

Operation-specific parameters

Cache Settings

📋 Output Field Reference

Web Search

Image Search

News Search

Video Search

🎯 Ranking Rules

How it behaves

Example

🔗 Result URL Handling

🔄 Fallback Behavior

⚠️ Error Behavior

Bot-detection challenge (rate limiting)

Image Search: VQD token missing

Image Search: 403 Forbidden

Web Search: Parser failure

Extract Page Content: refused address

Empty results

🤖 AI Agent Usage

🔍 Search Operators

🌍 Supported Regions

🔀 Migration Notes

Removed UI options (silently ignored)

Output field changes

New output fields

Correct field names (if your workflow used old examples)

🆘 Troubleshooting

Empty results

Search returns a bot-detection challenge error

Image search fails with VQD error

Image search fails with 403

Web search returns parser failure error

News/video results marked isFallback: true

Configuration errors

📄 Page Content Extraction

News/video results marked `isFallback: true`