Call · 30 min
MarketingMattia EspositoUpdated 10 November 202613 min read

How to rank on DeepSeek. It runs its own search engine, and its API shows you the score it gives your pages.

DeepSeek documents no crawler and no console for site owners. Since October 2026, though, its search engine has been open to everyone. Here is the sheet to test it on the searches of your own trade, in Excel and PDF.

In brief

To rank on DeepSeek a page has to be found by DeepSeek’s in-house search engine, which anyone can query through the Search API since 17 October 2026. How the engine discovers pages, DeepSeek does not say: no named crawler, no robots.txt rule.

For every result the API returns a score from 0 to 3 and an authority tier from 0 to 4. So says DeepSeek’s official documentation, which, together with the rulings of the Italian data protection and competition authorities, is the basis of this page.

In Italy the data protection authority has limited the processing of users’ data since 30 January 2025, and as of 10 November 2026 no revocation has been published.

The test sheet is free to download below, in Excel and PDF, with no email required. You run the test yourself, with your own key.

Written by: Mattia Esposito, AI and marketing consultant, founder of Itria AI. From Puglia, based in Turin since 2023, he works remotely in Italy and abroad. LinkedIn profile.

The DeepSeek search engine test sheet, free to download

An Excel sheet to run five searches from your trade on DeepSeek’s engine and record, for each one, whether your site appears, in what position and with what score. A second sheet records what comes back when DeepSeek reads five of your pages. You need an account on the DeepSeek API platform with a little credit: ten requests cost less than one US cent.

The DeepSeek search engine test sheet: five searches with position, rerank_score, authority and date, five pages read by the Fetch API and what each field means
FileContentsLink
Excelto fill in, in Italian

The Test, Fetch, Fields and Requests sheets, with totals that calculate themselves.

come-comparire-su-deepseek-prova.xlsx

PDFto print, in Italian

The same test, to fill in by hand while you run the requests.

come-comparire-su-deepseek-prova.pdf

SheetWhat you enterWhere it comes from
Proof

Five searches in a client’s own words, your site’s position among the first 50 results, rerank_score, authority and date.

Search API

Test, check

Whether the summary the model will read says what your page says, and who comes first.

summary field, url field

Fetch

Five of your pages as DeepSeek reads them: outcome, title, contact details and key facts in the text returned.

Fetch API

Fields

What each field means, in the documentation’s own words.

Search & Fetch guide

Requests

The two requests to paste into the terminal, with a slot for your key.

Search & Fetch guide

On 7 August a friend sent me what ChatGPT said about me: it could find no Mattia Esposito linked to itria.io. Since then I have looked at every assistant as a reader with its own library. If my site is not on the shelves, it can be beautifully written and still never be cited.

With almost every assistant you have to guess the library from the answers. In October DeepSeek started selling access to its own: anyone can ask the engine for a search and see the list the model would read, with a score next to each page. For anyone working on visibility in AI answers this is rare data, and the reason for this page.

The sheet holds no results for itria.io. The key costs money, DeepSeek’s privacy policy places data processing in China, and the decision to run the test belongs to you: here you get the tool and the rules for reading it.

Does DeepSeek have its own search engine?

Yes, according to its documentation: the Search API “is powered by DeepSeek’s in-house search engine”. The Search and Fetch APIs were opened to everyone with the release note of 17 October 2026, which describes the engine as serving hundreds of millions of requests a day worldwide. DeepSeek states the figure; nobody outside can check it.

The figure counts requests, not pages. You may read online about an index of hundreds of billions of pages: no DeepSeek page read for this guide says so. On the size of the index, then, there is no first-party figure today.

There is, however, a contradiction between two DeepSeek documents. The privacy policy, updated on 10 February 2026, says the company integrates third-party APIs for its search services and shares the keywords you type with them. The October documentation describes an in-house engine. The app may use both routes, or the policy may predate the engine: the public texts do not settle it.

WhatWhat DeepSeek saysDocument
Search engine

In-house, described as serving hundreds of millions of requests a day.

Search guide, note of 17 October 2026

Search in the app

Third-party APIs, with your keywords shared.

Privacy policy, 10 February 2026

Named crawler

None documented.

No document

Rating fields

rerank_score from 0 to 3, authority from 0 to 4, publication date.

Search guide

Reading pages

Subject to the site’s availability and access policies, with a per-domain request limit.

Search guide

Price

$0.75 per 1,000 successful requests, for Search and for Fetch.

Search guide

Search and accuracy

With the “Search” function on, accuracy may improve to some extent.

Terms of use

Where the data sits

In the People’s Republic of China.

Privacy policy

How a page reaches a DeepSeek answer with search on: the question goes to the in-house engine, the engine scores each page, the model reads the summary and answers The questionsearch on The engineDeepSeek in-house The scorererank and authority The summary600-800 characters The answerwith its sources Search columnRankingrerank, authority summary, dateFetch sheet
The path documented by DeepSeek’s Search API, and which column of the test sheet measures each step. How the engine discovers pages, DeepSeek does not say.

How DeepSeek’s search engine rates a page

Every Search API result carries a rerank_score from 0 to 3, which DeepSeek says combines signals such as relevance, authority and freshness, and an authority tier from 0 to 4. A score of 1.5 or above marks a high-quality result; an authority of 3 or above marks highly authoritative sources. How either is calculated, the documentation does not explain.

Next to the two scores sit other fields worth a site owner’s attention. The publication date is null when the engine does not know it, and the same goes for the site name. The summary, usually between 600 and 800 characters, condenses the page around the query and is written, says the guide, to be read by a model.

Search accepts the operators of a classic engine: site: and -site:, an exact phrase in quotation marks, lang: for language, after: and before: for dates. With lang:it you see only pages in Italian; if you sell abroad, repeat the same searches with lang:en. The count field goes up to 50 results per search.

What raises the two scores, DeepSeek does not say. The general rules on what gets a page cited, from the GEO study to Google’s guidance, are in the guide on how to rank on ChatGPT and in the one on generative engine optimization. The sheet shows where you stand today, with the engine’s own score rather than an estimate.

Does DeepSeekBot exist?

DeepSeek’s documentation names no crawler: no user agent, no published IP addresses, no robots.txt rule. The name DeepSeekBot appears in bot lists kept by third parties, with strings that change from one list to the next, and some of them admit there is no first-party source.

A line reading “Allow: DeepSeekBot” in robots.txt therefore rests on nothing DeepSeek has written. It does no harm, and it guarantees nothing. DeepSeek’s guide never mentions robots.txt: it says reading a page is subject to the site’s “access policies”, without defining them.

The practical rule lies elsewhere. A robots.txt that blocks every bot except a list of known names also shuts out anything without a documented name. How to write a robots.txt that keeps search and training apart, with the verified names of the other assistants, is on the page about robots.txt for AI crawlers; the short definition is in the glossary under AI crawler.

What can stop DeepSeek reading your site

DeepSeek’s Fetch API reads a page and returns it as Markdown, and its error codes tell you what stopped it. For a site owner this is the most direct check: if the page does not come through, or comes through without the contact details, the assistant cannot pass them on.

CodeWhat the documentation saysWhat to check
200

The page was read and its content returned as Markdown.

That contact details and key facts are in the text

403

The Fetch request was rejected by the security policy.

Run the test again; if it repeats across several pages, note it

404

No usable content: URL unreachable, URL blocked by site policy, an error on the site, an unsupported format, a page with no extractable text.

CDN, firewall, anti-bot service, pages built only from scripts

429

Requests-per-second limit exceeded, including the per-domain one.

Wait and retry

402

Insufficient account balance.

Top up the account

A 404 is the most ambiguous, because it lumps different causes together. On a company site, look first at a CDN or anti-bot service turning away automated visitors, then at pages that only show their text after running JavaScript or are made mostly of images. How Cloudflare treats AI crawlers is covered in the ChatGPT guide, in the section on the CDN.

I lived through the case of text that only appears after a script on my own site. Cloudflare’s email obfuscation stripped the address from the HTML and put it back with JavaScript: to an automated reader the contact page had no contact details. The story is in the piece on the email AI assistants cannot see, along with the same flaw found on 14 food producers’ sites out of 121.

DeepSeek in Italy: what the data protection authority decided

On 30 January 2025 the Italian data protection authority, the Garante, ordered, as a matter of urgency and with immediate effect, a limitation on processing the data of Italian users against Hangzhou DeepSeek Artificial Intelligence and Beijing DeepSeek Artificial Intelligence. The decision rests on Article 58(2)(f) of the GDPR and applies “nelle more”, pending the investigation opened on the same day.

In the decision the Garante records that the two companies had stated “di non essere entrate e di non avere pianificato di entrare nel mercato italiano”, that they had neither entered nor planned to enter the Italian market, and that they had pulled the app from the Italian stores. The Garante found that on 30 January the app was no longer in the Apple and Google stores, while the service could still be reached through the website.

As of 10 November 2026 no revocation has been published on the Garante’s website. The accurate wording is “limitation of data processing”, the term the decision uses. The Garante’s press release uses “blocks” in its headline.

WhereWhat happenedDate
Italy, data protection authority

Limitation on processing Italian users’ data; investigation opened. No revocation published.

30 January 2025

Italy, competition authority (AGCM)

DeepSeek’s commitments on informing users about “hallucinations” made binding; proceedings closed without a finding of infringement.

16 December 2025

Germany, Berlin

The Berlin data protection commissioner reports the app to Apple and Google as illegal content under the Digital Services Act.

27 June 2025

Germany, outcome

Apple and Google decline to block the app in their stores, according to Berlin’s annual report.

2025 report

European Union, EDPB

The task force on AI enforcement becomes the forum for coordination on DeepSeek.

11-12 February 2025

Two Italian authorities, then, have dealt with DeepSeek in different ways. The data protection authority limited data processing; the competition authority, which opened proceedings on 2 April 2025 over inadequate information about hallucinations, made DeepSeek’s commitments towards Italian users binding. For the United Kingdom and the United States this page has collected no rulings: the rules cited here are Italian and European.

Where your clients’ questions end up

DeepSeek’s privacy policy states that personal data is collected, processed and stored in the People’s Republic of China. The controller is Hangzhou DeepSeek Artificial Intelligence; for the European Union and the United Kingdom the policy names a representative under Article 27 of the GDPR, Prighter. In its January 2025 decision the Garante had noted that no representative had been designated in writing.

The policy also says DeepSeek collects personal data publicly available online to train its models. For the test in the sheet the consequence is simple: your searches use the words of a typical client, never the name of a real client or any personal data.

Is ranking on DeepSeek worth the effort for a small business?

Working only for DeepSeek makes sense for hardly any Italian small business, because the official app has been out of the Italian stores since January 2025. The engine serves more than the app, though: the Search API feeds the tools that developers and companies build on DeepSeek models, and the October note mentions searches run from DeepSeek Harness, Claude Code and Codex. How much they count in Italy, no source says.

The good news is that the work is the same as for the other assistants. A page with its text in the HTML, a written date, the company name spelt the same everywhere and facts with their sources works for Google, for Copilot, for Perplexity and for DeepSeek.

The difference is that here you can read the engine’s score, so measuring visibility becomes a sheet rather than a hunch. How to measure it on the other engines is explained in the guide on how to measure GEO.

In September I read the websites of 142 Italian companies that also sell abroad. In 61 of them a foreign buyer had no real way to get in touch, and only 28 out of 143 had the file that explains the company to an assistant, in 24 cases written by a plugin. The point, as I wrote at the time, is that “an assistant does not browse a site: it reads it, and reports what it finds written there”. DeepSeek is no exception: the full post.

Three cases measured on itria.io

The email that was not there. An address put back into the page by a script is invisible to anything reading only the HTML: the case is in the piece on the invisible email.

From one page to five. On the morning of 25 September 2026, for Claude’s web search, itria.io had a single page; by the evening, after asking Brave to recrawl seven, it had five. In Brave and Claude.

The pages that didn’t exist. Without a real 404 page, five URLs that had never existed were getting impressions in first position: the case of the phantom URLs.

The result. Between 11 October and 7 November 2026 Bing’s AI Performance report counted 268 citations of itria.io pages in Copilot, and Search Console 3,550 impressions in AI Overviews and AI Mode.

On 21 September I summed up what I had seen on my own site: “AI cites the body that issues the rule or the company doing the work. Never the people explaining it.” So this page quotes DeepSeek, the Garante and the AGCM in their own words, rather than a summary of summaries: the post.

The other guides are gathered in the index of marketing for small businesses, and the work on a real company is on the page about presence in search engines and AI assistants.

Questions and answers

How do I rank on DeepSeek?

Your page has to be found by DeepSeek’s in-house search engine, which powers its web search. DeepSeek documents no sign-up form, no console for site owners and no named crawler.

What you can do is test the engine with the Search API, see where your site appears and with what score, and keep your pages readable: text in the HTML, a date, the company name, sources.

Is there a DeepSeekBot crawler to add to robots.txt?

DeepSeek’s documentation, read on 10 November 2026, names no crawler, no user agent and no robots.txt rule. The name DeepSeekBot circulates in third-party lists, with strings that differ from list to list.

A rule written for that name has no first-party source. What matters is not blocking unknown bots across the board.

Is DeepSeek banned in Italy?

On 30 January 2025 the Italian data protection authority, the Garante, ordered a limitation on processing the data of people in Italy against the two companies behind DeepSeek, and the companies pulled the app from the Italian stores. As of 10 November 2026 no revocation has been published on the Garante’s website.

How much does it cost to test the DeepSeek Search API?

The documentation gives $0.75 per 1,000 successful requests, for search and for fetching a page alike. Failed requests and searches with no results are not billed. You need an account with credit: without it, the API returns error 402.

Where does the data from searches on DeepSeek end up?

DeepSeek’s privacy policy, updated on 10 February 2026, states that personal data is collected, processed and stored in the People’s Republic of China. So your test searches should never contain client names or personal data.

Notes on sources

  1. DeepSeek, Search & Fetch, API guide: in-house engine, result fields, operators, error codes, prices and limits; source of the quoted sentence. Read on 10 November 2026.
  2. DeepSeek, DeepSeek-V4.1-Flash-1017 Release, 17 October 2026: Search and Fetch opened to everyone, hundreds of millions of requests a day (the company’s own statement), calls from DeepSeek Harness, Claude Code and Codex.
  3. DeepSeek, Privacy Policy, updated on 10 February 2026, and Terms of Use. Read on 10 November 2026.
  4. Garante per la protezione dei dati personali, the Italian data protection authority, press release of 30 January 2025 and decision of 30 January 2025, doc. web no. 10098477, source of the quoted sentence. Search for later rulings on the Garante’s website on 10 November 2026.
  5. Autorità Garante della Concorrenza e del Mercato, the Italian competition authority, proceedings PS12942, meeting of 16 December 2025.
  6. Berliner Beauftragte für Datenschutz und Informationsfreiheit, press release of 27 June 2025 and 2025 annual report.
  7. European Data Protection Board, press release of 12 February 2025 and minutes of the 102nd plenary, 11 February 2025.
  8. The itria.io figures come from Search Console (generative AI report) and Bing Webmaster Tools, window 11 October to 7 November 2026. The count of the 142 sites dates from 7 September 2026.
·Who does it for you

Knowing which engine reads you is the first step. Getting chosen over your competitors is the work that follows. Positioning is built starting from the market.

Mattia Esposito, who wrote this guide, also works with Itria on positioning: in-depth market analysis to find where competitors are exposed and take that ground. He did it on his own site first. In August, asked who had founded itria.io, ChatGPT couldn’t find him. From 11 October to 7 November 2026 itria.io pages appeared 13,803 times in Google and Bing results and 3,818 times in AI answers: Google’s AI Overviews and AI Mode, and Microsoft Copilot.