DeepSeek opened its in-house hunt motor to developers on Oct. 17, and the company’s changelog says the motor filters SEO spam, “GEO manipulation,” and low-quality AI-generated content.
Each outcome from the new Search API includes an authority tier for the leaf and a relevance score, and the API accepts site: queries that bounds results to one domain.
What DeepSeek Says About Its Engine
DeepSeek’s changelog says its in-house motor handles hunt on its website and in its API, indexes hundreds of billions of web pages, and takes in hundreds of millions of hunt requests a day from about the world.
GEO (generative motor optimization) is whenever you form your satisfied to be cited by AI answer engines. DeepSeek’s changelog states that they credit strict finding as the origin of filtering; however, it has not defined what GEO manipulation is or described how their finding procedure operates. On May 15, Google said “our spam policies additionally use to generative AI responses in Google Search.”
The privacy policy of DeepSeek was final revised Feb. 10, which states that the business integrates “third-party APIs to provision hunt services” and shares users’ input keywords alongside these companies. A Chinese-based hunt API business called Bocha states they are the authoritative web hunt provider of DeepSeek in the client division of a GitHub README for one of their products. DeepSeek does not citation Bocha in their initiate pages. The changelog and the privacy guideline do not province if any web hunt queries are motionless sent to a third gathering service. Dan Petrovic from DEJAN referenced the two the privacy guideline and the GitHub README in his post concerning the DeepSeek launch.
What The Search API Returns
Each Search API outcome contains an authority field; authority sectors range from 0 to 4 alongside -1 being unknown. According to DeepSeek’s API reference, pages alongside tiers of 3+ are described as “highly authoritative” sources.
The rerank mark (another field) is a measure of 0-3 and incorporates signals that contain relevance, authority and freshness. A rerank mark of 1.5+ is considered elevated quality. The guide and the citation do not explain how a leaf receives its tier or score.
Each outcome additionally comes alongside a snippet for display, typically 100 to 200 characters, and a summary meant for a example to read, typically 600 to 800 characters. Page satisfied comes rear too whenever a petition asks for it and the satisfied is available. Queries can be constricted alongside site: and filtered by tongue and date published.
Search and Fetch all disbursal $0.75 per 1,000 prosperous requests. Searches that come back no results aren’t billed, and requesting 50 results expenses the identical as requesting fair one.
What DeepSeek’s Pages Say About Crawling
DeepSeek’s Fetch API was released alongside Search. Fetch provides a sole Markdown leaf and “is topic to the mark site’s preparedness and admission policies,” as per the guide.
According to the EU summary of the data used to train DeepSeek-V4, the crawler is “designed to regard robots.txt instructions and another norm web protocols, anywhere applicable.” However, the summary does not acknowledge which crawler this refers to.
Additionally, none of the initiate pages (the guide, API references, changelog and announcement) provision acknowledgment of a crawler for the hunt indicator nor do they provision data on what person delegate Fetch uses.
In supplement to providing data concerning multiple crawlers, the Known Agents bot directory lists a DeepSeekBot person delegate and describes it as an AI training data crawler.
Why This Matters
With a funded DeepSeek API account, idiosyncratic site: searches are under $0.01. Each site: hunt can come back up to 50 pages from your website; all leaf has the two an authority tier and a relevance mark for that query. If you see a missing result, it could be since the leaf was either exterior the indicator or caught by filters described in DeepSeek’s changelog, or since it was excluded for several another reason; neither the person guide nor the API citation provides a method to fig this out.
Robots.txt rules intended to obstacle DeepSeekBot are targeted against a crawler that the Known Agents bot directory describes as being connected alongside training data. It does not appear from DeepSeek’s first initiate pages if specified a regulation volition power what is included in DeepSeek’s hunt index.
Looking Ahead
If you use Claude Code, Codex, or DeepSeek Harness alongside the DeepSeek API, your server-side searches and leaf fetches are already handled through the Search and Fetch APIs at no cost. Billing for these server-side calls volition commencement on Oct. 31 at 16:00 UTC (noon ET).
Watch DeepSeek’s docs for a definition of GEO manipulation, particulars on how tiers are set, and a named crawler for the index.
