On October 17, 2026, DeepSeek said search in its web chat and its API now runs on an in-house search engine that, according to the company, filters out SEO spam, GEO manipulation and low-quality AI-generated content. It also opened that engine to developers as a paid API. It matters to anyone who wants to show up in AI answers.
What we know
- The engine: according to DeepSeek's API change log (October 17), search in its web product and its API is powered by its own engine. The company says it handles hundreds of millions of searches a day and indexes hundreds of billions of pages. These are DeepSeek's own figures, with no outside measurement.
- What it filters: the same entry says the engine filters «SEO spam, GEO manipulation, and low-quality AIGC» through strict detection. It does not define any of those terms.
- The two APIs, per DeepSeek's Search & Fetch guide:
- Search returns results for a query and supports, among others, the `site:`, `-site:`, `lang:`, `after:` and `before:` operators. Each result includes, among other fields, a relevance `rerank_score` from 0 to 3 (1.5 or above counts as high quality) and an `authority` score from 0 to 4 (-1 when unknown).
- Fetch retrieves one page and returns it as Markdown. The guide says it depends on the target site's availability and access policies, and a URL "blocked by site policy" returns a 404 error.
- Price: $0.75 per 1,000 successful requests for each API. Rate limit: 5 requests per second per account for each.
- Calls made through DeepSeek Harness, Claude Code or Codex are free until 00:00 Beijing time on November 1, 2026.
- The model: the same day, DeepSeek upgraded V4.1 Flash to DeepSeek-V4.1-Flash-1017. Calls to `deepseek-flash` now get the new version at the same price. DeepSeek reports gains on coding benchmarks such as DeepSWE v1.1, from 74.2 to 77.0.
What changes and what doesn't
What changes: another AI assistant now answers web questions from its own index rather than someone else's, and anyone with an API key can query that index. For the first time, site owners can see from the outside which of their pages DeepSeek holds and what authority score it assigns them.
What doesn't: the documentation still gives no crawler name, doesn't say whether DeepSeek honors robots.txt and offers no opt-out. It also doesn't say whether the mobile app uses the same engine; the entry mentions only the web product and the API.
How to check if it affects you
- See whether you're indexed. With a DeepSeek API key, a `site:yourdomain.com` query to the Search API shows which pages it holds and their `authority` score. At $0.75 per 1,000 requests, a check costs a fraction of a cent.
- Read your server logs. With no documented crawler name, look for user agents that mention DeepSeek and note them. Until you know the agent, a robots.txt rule is guesswork.
- Audit anything that could look like manipulation. DeepSeek doesn't spell out what it penalizes, but the direction is clear: mass-produced AI pages with nothing original, text written only to get a model to cite your brand, and "best of" lists you always win are likely to be filtered.
- If you build agents or chatbots, the Search API is a low-cost way to give them live web results. Budget before billing starts: 00:00 Beijing time on November 1 is 16:00 UTC on October 31, which is 12:00 in New York.
Related: Cloudflare and OpenAI: ChatGPT Now Finds More Pages
Sources
- DeepSeek, API change log, October 17, 2026 entry
- DeepSeek, Search & Fetch guide
- DeepSeek, Fetch API reference
Updates: if DeepSeek publishes its crawler name or a way for sites to opt out, it will be added here with a link.