AI NewsInfrastructureAnnouncement

DeepSeek search API grades every page from 0 to 4

DeepSeek opened its in-house search engine to developers this week, and each result carries a page authority tier from 0 to 4 that the company says is cleaned of SEO spam and GEO manipulation.

AI News

Editorial3 min read

LinkedInX

Why it mattersAn SEO or GEO practitioner can now query a general-purpose AI search engine and read back the exact authority tier it has assigned to each page, which is a signal most engines keep hidden.

A content team that cares about being cited by an AI answer engine can now ask DeepSeek, in a single API call, what grade it has given a given page. The grade is a number from 0 to 4, and DeepSeek says a 3 or 4 marks a highly authoritative source.

DeepSeek's own API changelog states that the Search and Fetch APIs are open to every developer. The release note describes DeepSeek Search as "an AI-native general-purpose search engine that indexes hundreds of billions of web pages", and says it "filters SEO spam, GEO manipulation, and low-quality AIGC through strict detection mechanisms, ensuring authoritative and trustworthy information for users and LLMs". AIGC in that line is DeepSeek's own term for AI-generated content.

What a search response carries

The Search and Fetch guide documents the fields each result returns. Alongside the usual title, URL, snippet and site name, each row carries a rerank_score from 0 to 3 that DeepSeek says combines relevance, authority and freshness, and an authority field. The guide states the authority tier is "0 to 4 (-1 means unknown)", with "values of 3 or above indicating highly authoritative sources". The guide does not say what each tier level means or how a page earns one, and does not list an appeals or correction route for a page tiered lower than its owner expects.

Each result also carries a 100 to 200 character snippet, a 600 to 800 character summary, the full page content when a request sets need_content: true, the site's name and icon, and a date field in RFC 3339 format. The Fetch API returns the full content of a given URL on demand.

The pricing and the free window

Both endpoints cost $0.75 per 1,000 successful requests, with DeepSeek billing only HTTP 200 responses that return a non-empty results array. Server-side use of the same two tools inside DeepSeek Harness, Claude Code and Codex is listed as free for every DeepSeek API user through 16:00 UTC on 31 October 2026, after which the per-call pricing begins to apply.

What is new for an SEO or GEO team

For a team already optimising for AI answer engines, the visible authority tier is the practical change. A content owner can call the API against their own pages, their competitors' pages, and the sources their AI answers tend to cite, and read a number for each from the same index that DeepSeek's own chat models draw on. That removes a step of guessing from the current GEO workflow, which so far has used indirect signals.

DeepSeek has not described how the detection of SEO spam, GEO manipulation and low-quality AI-generated content works in practice. Matt G. Southern of Search Engine Journal reports that DeepSeek has not defined what GEO manipulation is in its documentation, and the guide does not state a limit on how often a page is re-tiered or a route to request a review. A page that returns a 0 or 1 against the API is a page the engine does not currently treat as a source worth promoting to the chat model, and the owner of that page has no stated path to find out why.

The one engine most mentioned alongside DeepSeek in this market, Google, publishes no equivalent per-URL authority number. A GEO practitioner reading a surprising tier for a client's page therefore has one data point from DeepSeek and no comparable one from the AI engines they cannot query programmatically.

Source

Updates and Search & Fetch guide, DeepSeek API Docs. Press coverage: DeepSeek Opens Its In-House Search Engine To Developers, Matt G. Southern, Search Engine Journal.

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

Share
LinkedInX
Start a project