DUBAI — DeepSeek launched a public Search and Fetch API on October 17, 2026, alongside an update to its Flash model, introducing a unique authority scoring system for search results.
The new API returns an “authority” score from zero to four, with minus one indicating an unknown status, and a “rerank_score” from zero to three, aiming to combat SEO spam, geo-manipulation, and low-quality AI-generated content.
Initial tests of the API revealed that the “authority” score appeared to function as a host-level metric rather than a page-level one. Higher authority tiers generally correlated with higher median rerank scores, according to an analysis.
Testing showed that some spam PDFs hosted on government and university domains, such as oig.dc.gov and business.columbia.edu, received an authority score of four. Conversely, the author’s own site was assigned an authority score of one.
DeepSeek’s Fetch API identified itself as Firefox from Hong Kong during testing. It also demonstrated the ability to retrieve content from URLs that were explicitly disallowed by robots.txt files, indicating a potential bypass of standard web crawl directives.
The API’s spam filtering capabilities were not entirely effective in early assessments. Despite DeepSeek’s stated objective to filter out low-quality content, some spam results persisted.
Queries for specific software tools frequently yielded package registries, app stores, and marketplace listings in the search results. This occurred even when directly relevant content from the author was available, suggesting a preference for broader platforms.
Organizations mentioned in the context of the API’s functionality and comparison included Apify, Chrome Web Store, Writesonic, Resoneo, Google, SEJ, Bounteous, WordPress.org, PyPI, n8n, Akamai, npm, and GitHub.
Source: Search Engine Journal
