DeepSeek opens its search API to developers, and says it already blocks GEO manipulation

DeepSeek has turned its in-house search engine into a paid API for developers, and the company says every result it returns has already been screened for spam and for content built specifically to game AI answer engines. The move, rolled out alongside a new model release on October 17, gives any team a direct, queryable window into how one more AI system sees their site — and a price tag for using it.

What actually launched

DeepSeek's October 17 changelog opened its Search and Fetch endpoints to all API users, built on what the company describes as its own in-house search engine rather than a licensed third party. Search returns ranked results for a query; Fetch pulls a single page and converts it to Markdown. DeepSeek's own documentation states each endpoint is priced at $0.75 per 1,000 calls, with calls made through DeepSeek Harness, Claude Code or Codex currently free. That free period ends when billing starts October 31 at 16:00 UTC, per the changelog. The company also says the engine is "serving hundreds of millions of requests daily worldwide," though that figure, along with a separate claim of an index spanning hundreds of billions of pages, comes from DeepSeek itself and has not been independently measured, according to an analysis by SEO research firm Dejan.ai.

Every result carries two scores on top of the usual title, link and snippet. An authority field runs 0 to 4, with DeepSeek's own API reference stating that a tier of "3 or above" marks a "highly authoritative" source. A separate rerank score runs 0 to 3 and blends relevance, authority and freshness, with 1.5 or higher treated as high quality. Developers can scope a query to one domain with a site: operator, which is the detail that matters most for marketers: it turns DeepSeek's index into something a content or SEO team can interrogate directly, the same way they already check Google Search Console or Bing Webmaster Tools.

The filtering claim, and why it's worth attention

Alongside the launch, DeepSeek has said its engine screens out "SEO spam, GEO manipulation, and low-quality AI-generated content," in the wording Search Engine Journal attributes to the company. DeepSeek has not defined what counts as manipulation or explained how its detection works, and Dejan.ai's review is careful to separate what DeepSeek claims from what outsiders can verify: nobody outside the company has measured the index, confirmed how authority tiers get assigned, or checked whether DeepSeek still routes any queries through third-party search suppliers it used before building its own stack.

That gap between claim and verification is the real story for anyone doing generative engine optimization. DeepSeek is the first major AI lab to put a number on an anti-manipulation policy and expose a tool — the site: query with an authority score attached — that lets outsiders probe it, even informally. Most GEO discussion so far has focused on whether AI answer engines can be gamed at all; DeepSeek's pitch is that its index already screens out the attempt. Whether that holds is untested, but a lab making the claim in public, with a priced API attached, changes the incentive to find out.

What a software marketer should do differently

Most GEO monitoring to date has centered on ChatGPT, Perplexity and Google's AI Overviews. DeepSeek's API gives a cheap, direct way to check a fourth surface without buying a third-party visibility tool: run a site: query against your own domain, read the authority tier it comes back with, and see what content DeepSeek has indexed versus what it has skipped. A gap between what you expect to be indexed and what the API returns is itself a signal worth investigating, even if DeepSeek won't explain the scoring.

• Run a handful of site: queries against your own domain and your top competitors' to see what DeepSeek has indexed and what authority tier it assigns.
• Treat the stated anti-manipulation filtering as a reason to avoid shortcut GEO tactics, not a settled fact — the company hasn't disclosed how detection works.
• Check your robots.txt rules for DeepSeek's crawler alongside the rules you already maintain for Google, Bing, OpenAI and Perplexity's bots.
• Budget for the API's per-call pricing if you plan to monitor at any scale past October 31, when the free tier for Harness, Claude Code and Codex calls ends.

The bigger shift is what DeepSeek's move signals about the direction of GEO tooling generally. A year ago, "optimizing for AI search" meant writing content and hoping ChatGPT or Google picked it up. Now a lab is shipping the inspection tool alongside the search engine itself, pricing access to both, and stating — without proving — that it has already closed the door on the cheapest manipulation tactics. For a marketing team deciding where to spend limited GEO attention, that combination of transparency and unverified claims is exactly the kind of signal worth tracking before committing a tool budget to it.