DeepSeek's search API gives every page on a host the same score: Reveneau

DeepSeek's search API gives every page on a host the same score

SEO consultant Suganthan Mohanadasan ran 110 calls against DeepSeek's new search and fetch APIs for Search Engine Journal, and reports that every URL from one host gets the same authority score, that 11 percent of the index has no date, and that the fetcher pulled three pages their robots.txt tells it not to.

Why it mattersThe authority score DeepSeek returns is a score for the whole domain, so a page can only move its score by moving the score of the whole site, and the page on its own carries whatever the host carries.

An SEO person looking at the authority score DeepSeek returns for one of their pages is reading a number that covers the whole domain, with every other URL on the same host carrying the same value. That is the main finding in a 110-call test SEO consultant Suganthan Mohanadasan ran against DeepSeek's new search and fetch APIs for Search Engine Journal. He queried the API from Dubai, read back 4,322 results, and reports that every URL from one host carried the same authority number: suganthan.com and growth-memo.com both scored 1, ipullrank.com, brodieclark.com, medium.com and reddit.com all scored 2, ahrefs.com, moz.com, searchengineland.com, aleydasolis.com and linkedin.com all scored 3, and zenodo.org scored 4. The one break he found was a Reddit subdomain, where qh.reddit.com scored 1 while www.reddit.com scored 2.

The scores and what they mean

DeepSeek's own guide describes the authority field as a tier from 0 to 4, with "values of 3 or above indicating highly authoritative sources". Mohanadasan's run breaks down the 4,322 results as 2 percent at 1, 32 percent at 2, 60 percent at 3, and 6 percent at 4. Each result also carries a rerank score that DeepSeek says combines relevance, authority and freshness; across 1,000 results from 20 queries the median rerank was 1.63, the range was 0.75 to 2.70, and 68 percent scored 1.5 or above. Only 196 distinct rerank values appeared in those 1,000 results, so the output is bucketed into fixed values.

The gaps in the spam filter

DeepSeek says its index "filters SEO spam, GEO manipulation, and low-quality AIGC through strict detection mechanisms". Mohanadasan reports that the test turned up "Free Robux Generator" PDFs hosted on oig.dc.gov with authority 4 and a rerank score of 1.42. The host carried a government domain, so the uniform host-level score lifted spam content on it to the top tier.

The fetcher reads pages it is told to leave alone

The second surface, the Fetch API, took URLs the author named and returned their contents, including three disallowed by the sites' own robots.txt: httpbin.org/deny, Wikipedia's Special:Search page, and google.com/search. The requests arrived from five IPs on AS136907, Huawei Cloud Hong Kong, carrying the user agent "Mozilla/5.0 (X11; Linux x86_64; rv:149.0) Gecko/20100101 Firefox/149.0". The string contains nothing that names DeepSeek, so a webmaster blocking the fetcher has to match on that whole fingerprint, since there is no declared bot name to use.

The date filter removes undated pages

Eleven percent of the unfiltered results carried a null date, and all of those URLs disappeared when a date filter was applied. Every returned date ended with "T16:00:00Z", midnight in Beijing. A page without a machine-readable date is therefore invisible to any DeepSeek user who narrows a query by time, which is the common path for recent news or product queries.

The whole 110-call run cost $0.08 at DeepSeek's $0.75 per 1,000 successful requests. A publisher who cares what score a given page carries can ask the API the same question for a cent, which is cheap enough to run against every page a site owns and store the result.

Source

I Tested DeepSeek's New Search API, Here's How It Ranks And Scores Websites, Suganthan Mohanadasan, Search Engine Journal.

This item was written by an AI system from the linked source. Reveneau is responsible for what it publishes.