DeepSeek says its search engine filters ‘GEO manipulation’

DeepSeek has put “GEO manipulation” alongside SEO spam and low-quality AI-generated content in the announcement opening its search API to developers. It says its engine filters all three, without giving examples of the GEO tactics it catches.

The claim appears in the company’s October 17 changelog. DeepSeek says the same in-house engine powers search on its website and through its API. Developers can now retrieve ranked web results and use that material in their own applications.

Generative engine optimization, or GEO, describes work intended to make brands and content more visible in AI-generated answers. The launch note offers no definition of manipulation or test results showing how the filtering works.

Google already applies spam rules to AI answers

Google’s spam policies explicitly cover attempts to manipulate generative AI responses in Search. They also identify specific abusive practices, including hidden text placed to manipulate search engines and producing large numbers of pages that add little value for readers.

At the same time, its guidance for AI Overviews and AI Mode continues to recommend ordinary SEO work: making important information available as text, helping crawlers access pages and keeping structured data consistent with visible content. Google says no special optimization is required for those AI features.

Those documents give site owners both recommended actions and examples of abuse. DeepSeek’s announcement supplies a filtering claim without comparable detail about the publishing practices it targets.

The API exposes scores for returned pages

DeepSeek is more specific about what developers can inspect. Its Search API reference documents two scores attached to each returned page:

The API accepts site: queries and can return up to 50 results. A publisher could use it to inspect which pages from its domain appear in a search and the tiers attached to those pages.

That gives marketers a way to examine source selection before an AI application writes its answer. Whether a returned page becomes a visible AI citation is a further outcome to check in the answer itself.

DeepSeek’s published pricing lists Search and Fetch at $0.75 per 1,000 successful requests each. Search finds ranked results; the separate Fetch API retrieves an individual page as Markdown text.

The response schema contains no field identifying a spam-filter decision. If a page is missing, the returned data leaves its absence unexplained.