DeepSeek Opens Search Engine to Developers, Promises GEO Spam Filtering

DeepSeek opened its in-house search engine to developers on Oct. 17, and its changelog claims the engine filters SEO spam, "GEO manipulation," and low-quality AI-generated content. The claim arrives without a definition of GEO manipulation or any description of how the detection works.

The launch matters for search professionals. DeepSeek's changelog says the engine indexes hundreds of billions of web pages and processes hundreds of millions of search requests per day worldwide. Those figures come from the company itself, not from an audited third party, and no methodology accompanies them.

What does the Search API actually return?

Each result carries two proprietary scores:

• Authority tier: a scale from 0 to 4, with -1 meaning unknown. DeepSeek's API reference calls pages at tier 3 and above "highly authoritative" sources.
• Rerank score: a scale of 0 to 3 that combines relevance, authority and freshness. A score of 1.5 or higher counts as high quality.

Neither the guide nor the reference explains how a page receives its tier or score. Each result also includes a display snippet, typically 100 to 200 characters, and a model-facing summary of 600 to 800 characters. Queries accept site: operators, language filters and date-published filters.

Pricing is flat: Search and Fetch each cost $0.75 per 1,000 successful requests. Empty searches are not billed, and a 50-result request costs the same as a one-result request.

The takeaway: two opaque numeric scores now gate visibility in a system serving hundreds of millions of daily queries, and nobody outside DeepSeek can audit them.

Is DeepSeek really running its own index?

The company's privacy policy, last revised Feb. 10, states DeepSeek integrates "third-party APIs to provide search services" and shares users' input keywords with those companies. The changelog and the privacy policy do not say whether web search queries still go to a third-party service.

One third party has already claimed the supplier role. Bocha, a Chinese search API company, states in the customer section of a GitHub README for one of its products that it is the official web search supplier for DeepSeek. DeepSeek never references Bocha on its launch pages. Dan Petrovic of DEJAN connected the privacy policy and the GitHub README in his post about the launch.

There is precedent for the filtering stance, though not from DeepSeek. On May 15, Google said "our spam policies also apply to generative AI responses in Google Search."

The takeaway: an "in-house" engine that shares query keywords with unnamed third parties deserves scrutiny before it earns the label.

Which crawler builds the index?

DeepSeek released a Fetch API alongside Search. Fetch returns a single Markdown page and "is subject to the target site's availability and access policies," according to the guide.

The crawling picture has three gaps:

• The EU summary of data used to train DeepSeek-V4 says the crawler is "designed to respect robots.txt instructions and other standard web protocols, where applicable," but never names the crawler.
• None of the launch pages — guide, API references, changelog, announcement — identify a crawler for the search index or disclose the user agent Fetch uses.
• The Known Agents bot directory lists a DeepSeekBot user agent and describes it as an AI training data crawler, distinct from any documented search crawler.

This creates a practical trap. A robots.txt rule blocking DeepSeekBot targets a training-data crawler per Known Agents, and DeepSeek's launch pages do not say whether such a rule affects search index inclusion.

Why site owners should test now

With a funded API account, individual site: searches cost under $0.01. Each search returns up to 50 pages from a domain, each page carrying its authority tier and rerank score for that query.

Missing results carry three possible explanations and no way to distinguish them: the page sits outside the index, the changelog's filters caught it, or some unnamed exclusion applies. Neither the user guide nor the API reference provides a diagnostic method.

What changes on Oct. 31?

Free server-side search ends soon. Users of Claude Code, Codex or DeepSeek Harness with the DeepSeek API currently run searches and fetches through the new APIs at no cost. Billing for those server-side calls starts Oct. 31 at 16:00 UTC, noon ET.

Watch three items in DeepSeek's documentation next: a definition of GEO manipulation, the method behind the authority tiers, and a named crawler for the search index.