I Tested DeepSeek’s New Search Api, Here’s How It Ranks And Scores Websites

DeepSeek has opened admission to its own hunt motor through a community Search and Fetch API. It returns an authority mark for all outcome and says it filters out SEO spam, GEO manipulation, and low-quality AI content.

That is a important publish for SEOs.

The announcement on Oct. 17, 2026 came alongside an update to its Flash model. DeepSeek says the hunt motor already powers hunt on its website and through its API, indexes hundreds of billions of pages and handles hundreds of millions of searches a day.

We can now query that scheme directly, inspect the sources it returns, and see several of the scores attached to them. That gives us item tangible to examine before an AI example turns those sources into an answer.

Naturally, I wanted to see what it idea of my site. And whether the spam filtering held up.

I spent $0.08 finding out.

My location came rear alongside an authority mark of 1. Spam PDFs on authorities and university hosts came rear alongside authority 4. The Fetch API additionally presented itself as Firefox from Hong Kong and returned satisfied from URLs disallowed by robots.txt.

These are my preliminary findings.

What I Tested

I ran 110 calls on Oct. 17, 2026 from Dubai: 93 Search calls and 17 Fetch calls. Around 103 were billable under the documentation’s billing rule, costing about $0.08.

The tests covered mark distributions, authority throughout distinct hosts, six queries connected to my own tools and research, localization, date filters, and fetch behavior.

I used the Search and Fetch documentation as the reference. Where the responses differed from the documentation, I’ve kept the observed results clear.

DeepSeek Gives You 2 Scores

Each hunt outcome includes two sectors value paying notice to:

These definitions arrive from the Search API reference. The authority tier is DeepSeek’s own assessment, and the records describes it as a asset of the page.

Across 20 SEO and GEO queries at count=50, I collected 1,000 results. Their rerank scores ranged from 0.75 to 2.70, alongside a median of 1.63. Results were consistently sorted by that score, and 68% scored 1.5 or above.

Only 196 distinct scores appeared throughout those 1,000 results. The returned values looked bucketed, although that solitary doesn’t inform us whether the underlying scores are rounded or calculated in fixed steps.

Authority Looked Tied To The Host. Mine Was 1

Across all 4,322 results in the run, authority tiers -1 and 0 never appeared. The allocation looked akin this:

Higher authority tiers came alongside higher median rerank scores in this sample. That doesn’t isolate the consequence of authority from the another ranking signals, but it makes the site value investigating.

I afterward ran 16 site: probes throughout 16 hosts. On all host, all returned URL had the identical authority score. Here are several examples:

So although the records calls this a leaf authority tier, it behaved akin a presenter mark in my tests.

The Reddit subdomains were particularly helpful here: qh.reddit.com scored 1, during www.reddit.com scored 2. That points towards the presenter being the applicable unit, fairly than the complete registrable domain sharing one score.

A hunt for site:suganthan.com filled the 50 outcome cap. It returned my homepage, /resources/, notes and blog posts. DeepSeek was finding satisfied throughout my site. Those pages all carried authority 1.

Being indexed was plainly lone part of the story.

I Lost My Own Tool Queries To Registries And App Stores (Lol)

I ran six mark queries alongside count=50 from the UAE.

Package registries, app stores, and market listings did fine on these tool queries. PyPI, npm, Apify, the Chrome Web Store, WordPress.org, and GitHub were the kinds of hosts surfacing.

My Gemini part and FanoutFox the two had authority 1. Their rerank scores of 1.14 and 1.37 additionally cut below DeepSeek’s documented 1.5 high-quality threshold.

For my site, having satisfied immediately concerning the item being searched for wasn’t adequate to attack those hosts. The low authority tier is a plausible part of the explanation, although these tests don’t inform us its exact weighting.

The Spam Filtering Still Let Some Questionable Stuff Through

DeepSeek explicitly says its motor filters SEO spam, GEO manipulation, and low norm AI content. I wanted to see how that assertion lined up alongside the results.

Here is what motionless appeared:

I additionally saw self-published “frameworks” on zenodo.org obtain authority four and position #one and #three for “generative motor optimization.” That is another study concerning how authority behaved. Self-publication solitary doesn’t create a leaf spam.

The authorities and university examples are difficult to quadrate alongside the commitment of filtering this material out. Those uploads motionless received the highest authority tier.

This is why the apparent use of a presenter mark matters. A scheme that carries a host’s authority throughout its pages can end up giving parasite content the identical care as lawful matter on that host.

The results don’t inform us how much spam DeepSeek removed. They do display that these examples got through, and that a elevated authority tier wasn’t a dependable judgement of the idiosyncratic page’s quality.

How To Run The Same Kind Of Search

Here is a petition using an API key stored in an surroundings variable:

curl -s https://api.deepseek.com/v1/tools/search -H "Authorization: Bearer $DEEPSEEK_API_KEY" -H "Content-Type: application/json" -d '{"query":"find out who Gemini cites for your query","count":50}'

A trimmed outcome from that query:

{ "url": "https://suganthan.com/blog/gemini-citations/", "rerank_score": 1.14, "authority": 1 }

Search and Fetch all disbursal $0.75 per 1,000 prosperous requests. Under the billing rules, unsuccessful requests, bare hunt results, and timeouts are not billed. Returning 50 results expenses the identical as returning fewer.

There is a distinct impermanent exemption for the server-side Search and Fetch calls already made whenever using the DeepSeek API through Claude Code or Codex. Those remain liberated until one November 2026 at 00:00 Beijing time, which is 31 October at 8:00 PM GST. Per request, billing starts then. My straightforward API tests were billable.

3 Things To Account For When Testing

1. Set Your Localization

Leaving out localization produced the identical results as specifying a UAE IP throughout all six mark queries. I was calling from Dubai, which is accordant alongside DeepSeek using the caller’s IP as the default.

Set this site explicitly to an IP location in the market you desire to test. Otherwise, your audit may indicate anywhere the manuscript happens to run.

Using a UK or U.S. IP changed the rerank scores of 26 to 41 of the 50 results and exchanged 4 to 15 URLs. The average mark delta was small, about ±0.05. It mostly reshuffled the results, without substantially changing the mark scale.

A Chinese IP, 114.114.114.114, behaved likewise to the UK and US settings. I didn’t see a plainly distinct Chinese outcome set in these tests, although this doesn’t established how DeepSeek organizes its indexes.

2. The Original Question Made Very Little Difference

The ori_question site lets you provision the user’s first inquiry alongside the hunt query. DeepSeek says this can enhance query understanding.

Across my six queries, it changed the rerank mark for lone zero to two of the 50 results and exchanged at most one URL.

3. Date Filters Removed Every Undated Result

The after: and before: operators produced zero date violations throughout four queries.

They additionally removed all undated page. 11% of the unfiltered results had date: null, and those URLs disappeared whenever a date display was applied.

That matters if your satisfied needs to appear in searches for latest information. Put an suitable machine-readable date on the page, afterward inspect that DeepSeek really recognizes it. A leaf it treats as undated can disappear from a date-filtered search.

Every returned date that was current ended in T16:00:00Z, which corresponds to midnight Beijing time. The filtering behavior pointed to Beijing calendar days in these tests.

I additionally established two differences between the documented and observed content lengths:

That is value knowing if you are construction a tool about these responses. The summaries I received were considerably longer than the records suggested.

The Fetch API Arrived As Firefox From Hong Kong

The Fetch requests I inspected did not acknowledge themselves as DeepSeek. They used this user agent:

Mozilla/5.0 (X11; Linux x86_64; rv:149.0) Gecko/20100101 Firefox/149.0

The Accept-Language header was:

zh-CN,zh;q=0.9

There was no From header, bot token, Signature or Signature-Input. The requests presented themselves as Firefox, alongside no DeepSeek identifier to equivalent in the person agent.

Five fetches came from five distinct IPs, all identified as AS136907, Huawei Cloud, Hong Kong:

121.91.172.54 183.87.40.160 119.8.239.43 27.106.106.73 159.138.131.203

In the rendering test, it loaded CSS. It did not run JavaScript, burden images, or blaze the XHR beacon. Content that required JavaScript did not appear in those fetches.

These observations involvement the Fetch API. They don’t established how the crawler that builds DeepSeek’s hunt indicator behaves.

It Returned Content From URLs Disallowed By Robots.Txt

The Fetch API returned complete satisfied from three URLs disallowed for User-agent: *:

1. httpbin.org/deny
2. Wikipedia’s Special:Search
3. google.com/search

I could not verify whether it always requested /robots.txt. What I could verify was that the disallowed URLs motionless returned satisfied through the API. Those rules did not forestall the fetches I tested.

What To Look For In Your Logs

If your access logs document Accept-Language, this volition assistance exterior requests matching the observed headers:

grep -E 'Firefox/149.0' access.log | grep 'zh-CN'

A additional particular display combines the Firefox version, tongue header, and ASN. A Cloudflare tradition rule in that form would be:

(ip.src.asnum eq 136907 and http.user_agent contains "Firefox/149.0" and any(http.request.headers["accept-language"][*] contains "zh-CN"))

Treat that blend as a fingerprint to investigate. It is according to what I observed in these requests and could change.

There is no DeepSeek-specific person delegate here, and blocking the Firefox cord solitary can capture lawful browsers. AS136907 is additionally a broad Huawei Cloud network. Blocking the entire ASN can power another tenants, so don’t assume all petition from it belongs to DeepSeek.

What I Would Do With This As An SEO/GEO/AEO/AI SEO

The helpful part of this publish is that we can inspect the hunt results and scoring directly.

Here is anywhere I would start:

1. Run your genuine mark queries against POST https://api.deepseek.com/v1/tools/search. Record rank, rerank score, and authority together.
2. Check your site’s authority tier alongside site:yourdomain.com, including any applicable subdomains.
3. Set localization explicitly so repeated tests portray the market you attention about.
4. Check your dates. Add suitable machine-readable dates and verify whether DeepSeek returns them. Compare results alongside and without date filters.
5. Test applicable allocation for your tools. PyPI, npm, Apify, the Chrome Web Store, WordPress.org, and GitHub performed fine in these queries. If your tool belongs on one of those platforms, a lawful listing is value evaluation alongside your own site.
6. Monitor the fetcher in your logs using the blended Firefox/149, zh-CN and AS136907 pattern, during checking matches before treating them as DeepSeek.

For me, the most engaging finding is the blend of apparent authority tiers, feeble rankings for several first tool content, and apparent spam receiving the highest authority tier. It gives us particular questions we can now test.

This was one day, 110 calls, six mark queries, and one calling region, alongside additional probes for scores, hosts, filters, and fetching. I motionless couldn’t verify whether the fetcher requested robots.txt. Treat the numbers as directional, please.

More Resources:

• How ChatGPT Actually Picks Sources (I Read The Network Traffic, Not The Outputs)
• The Technical Signals AI Search Uses That Most SEOs Still Aren’t Optimizing
• Checking A Page Is Part Of A Retrieval Pipeline For AI

---

This article was initially published on Suganthan.

---

Featured Image: Fahmi_Ruddin_Hidayat/Shutterstock