DeepSeek Opens Search API to Developers With Spam Filters and Authority Tiers

DeepSeek opened its in-house search engine to developers on Oct. 17, and the company's changelog says the engine filters SEO spam, "GEO manipulation," and low-quality AI-generated content.

The Chinese AI firm disclosed few operational details. Its changelog credits "strict detection" as the source of the filtering. It has not defined what GEO manipulation is, and it has not described how the detection process operates. GEO (generative engine optimization) is the practice of shaping content so AI answer engines cite it.

For comparison, Google took a public position on May 15, stating that "our spam policies also apply to generative AI responses in Google Search."

How large is the engine behind the API?

DeepSeek's changelog says the in-house engine handles search on its website and inside its API. It indexes hundreds of billions of web pages. It takes in hundreds of millions of search requests a day from around the world.

The company has not named the crawler that builds this index. Its launch pages — the guide, API references, changelog and announcement — provide no crawler identification and no user agent for Fetch, the companion tool released alongside Search.

What does the Search API return?

Each result carries structured fields that let marketers audit their own pages:

• Authority tier: a value from 0 to 4, with -1 meaning unknown. DeepSeek's API reference describes pages at tier 3 or above as "highly authoritative" sources.
• Rerank score: a 0-3 scale combining relevance, authority and freshness. A score of 1.5 or higher counts as high quality.
• Snippet: typically 100 to 200 characters, meant for display.
• Summary: typically 600 to 800 characters, meant for a model to read.

The guide and the reference do not explain how a page receives its tier or its score. That opacity limits the API's value as a diagnostic tool.

Queries support site: operators that restrict results to one domain, plus filters by language and publication date. Page content returns when a request asks for it and the content is available.

What does it cost?

Search and Fetch each cost $0.75 per 1,000 successful requests. Searches that return no results are not billed. Requesting 50 results costs the same as requesting one.

That pricing makes self-audits cheap. With a funded API account, an individual site: search costs under $0.01 and can return up to 50 pages, each with its authority tier and relevance score. A missing result has three possible causes — the page sits outside the index, a filter caught it, or another exclusion applied — and neither the user guide nor the API reference provides a method to determine which.

Who supplies the search results?

DeepSeek's privacy policy, last revised Feb. 10, states that the company integrates "third-party APIs to provide search services" and shares users' input keywords with these companies. A Chinese search API company called Bocha states in the customer section of a GitHub README for one of its products that it is the official web search supplier of DeepSeek.

DeepSeek does not reference Bocha in its launch pages. The changelog and the privacy policy do not state whether web search queries still go to a third-party service. Dan Petrovic of DEJAN referenced both the privacy policy and the GitHub README in his post about the launch.

What about crawling and robots.txt?

The Fetch API returns a single Markdown page and, per the guide, "is subject to the target site's availability and access policies." The EU summary of the data used to train DeepSeek-V4 says the crawler is "designed to respect robots.txt instructions and other standard web protocols, where applicable." The summary does not identify which crawler it means.

The Known Agents bot directory lists a DeepSeekBot user agent and describes it as an AI training data crawler. A robots.txt rule blocking DeepSeekBot therefore targets a training-data crawler, and DeepSeek's launch pages do not say whether such a rule affects the search index.

When does billing start?

Developers using Claude Code, Codex or DeepSeek Harness with the DeepSeek API already have server-side searches and page fetches handled through the Search and Fetch APIs at no cost. Billing for those calls begins Oct. 31 at 16:00 UTC, noon ET.

Three gaps define the near-term agenda for anyone tracking this launch: a definition of GEO manipulation, the method behind the authority tiers, and a named crawler for the search index. Until DeepSeek publishes those, marketers auditing their pages on the platform work with scores they cannot explain and filters they cannot see.

By Elena Vasquez