DeepSeek: Chat Web Search, Search API and Fetch API
DeepSeek runs its own search engine for its chat app and its Search and Fetch APIs. What is documented, and what site owners can control.
DeepSeek is an AI company that makes the DeepSeek models and the DeepSeek chat app. Its homepage lists Search as a feature of the app. In October 2026, DeepSeek opened a Search API and a Fetch API to all API users. DeepSeek says that both run on its own search engine. DeepSeek publishes no crawler name, user agent or robots.txt rule, so a site owner has no documented way to opt in or out.
Web search in the DeepSeek chat app
DeepSeek added Internet Search to its web chat at chat.deepseek.com on December 10, 2024. The release note for DeepSeek-V2.5-1210 says that users turn on the Internet Search option to get real-time answers. DeepSeek introduced the DeepSeek App on January 15, 2025, and its feature list includes web search. These release notes do not say how the chat shows or links its sources.
Search and Fetch API
The DeepSeek Search & Fetch guide describes two endpoints:
- The Search API finds web pages for a query and returns structured results. DeepSeek’s in-house search engine runs it.
- The Fetch API gets one web page and returns its content as Markdown.
Each search result has a title, URL, snippet, summary, site name, site icon, date and, on request, the page content. It also has two scores:
| Field | Range | Meaning, as DeepSeek describes it |
|---|---|---|
rerank_score |
0 to 3 | Combines signals such as relevance, authority and freshness. 1.5 or above counts as a high-quality result |
authority |
0 to 4, or -1 if unknown | The authority tier of the page. 3 or above means a highly authoritative source |
A query can use operators such as site:, -site:, lang:, after: and before:. So an app can limit results to some sites or exclude others.
The guide also says that DeepSeek Harness, Claude Code and Codex already call these APIs as server-side tools when they use the DeepSeek API. The October 17, 2026 release note announced the APIs for all users.
What DeepSeek says about fetching pages
The guide says only that fetching depends on the availability and access policies of the target site. It names no user agent and does not mention robots.txt. The Fetch API returns HTTP 404 when it cannot get usable content. One example in the guide is a URL that site policy blocks.
DeepSeek does not describe how its search engine collects pages. It publishes no IP ranges and no reverse DNS host name for its requests.
What a site owner can do
DeepSeek documents no robots.txt token. The section below covers DeepSeekBot, a name from third-party bot lists.
A Disallow rule in the User-agent: * group applies to any crawler that follows robots.txt and has no group of its own. But that rule also blocks every search engine crawler without its own group. And DeepSeek has not said that its search engine follows robots.txt.
The DeepSeekBot name
Third-party bot lists and security vendors use the name DeepSeekBot for a crawler that they connect to DeepSeek. DeepSeek does not document this name, and the lists do not agree on one user agent string. The README of the DeepSeek-LLM repository says that the model training data includes “self-collected data respecting robots.txt”. The README does not name the crawler. If you want to block the reported name, add this group:
User-agent: DeepSeekBot
Disallow: /This group works only for a crawler that uses the token DeepSeekBot and follows robots.txt.