How this check works

Enter a URL and the questions your customers ask. The tool checks which pages DeepSeek Search returns. It does not collect chatbot answers, so it cannot tell you whether DeepSeek cites or recommends your company.

Run your first check

1. Paste the website or product-page URL you want to track.
2. Add your company name so the tool can separate questions about you from questions about the product category.
3. Enter one customer question per line. No tags needed.
4. Add competitor URLs if you want a comparison. Use product pages on both sides, or whole websites on both sides.
5. Review what counts as a match and how many API calls the run needs. Add technical checks or location hints if needed, then run.

The form starts blank. It works for any public website or product. Use the run label to recognise a test later. It does not translate your questions.

Which URLs count?

Automatic matching tracks an exact page when your URL has a path or meaningful query parameters. A root address tracks the website. You can change this choice.

Page matching ignores fragments, leading www, trailing slashes, HTTP/HTTPS differences and common tracking parameters: utm_*, gclid, fbclid and msclkid. Other query parameters matter, but their order does not. Path case matters. Redirects, canonical tags, translations and renamed URLs are not resolved.

The results also count other pages on the same website. Your blog might appear while your product page does not. A third-party article mentioning you does not count as your website.

What the numbers mean

A category question does not name your configured company or hostname. Questions naming your company stay in the results but are left out of the headline percentages.

If company name is blank, only your hostname is used to identify company questions. Review the question list before running. The tool does not automatically recognise competitor names or every product name as your company.

Failed searches are left out. A successful search without a match counts as absence, including a search returning no pages. With no successful category searches, the result is "Not measured". Ranks follow the API's returned order.

Check the scope beside each competitor. A whole website has more chances to appear than a single product page. Several companies can appear in the same search, so these percentages are not market share. Older reports may use different question sets for each product.

What the technical checks tell you

Authority scores

The optional authority check runs a site-filtered search for each distinct domain under each location hint. It groups returned pages by host and shows their scores. This is a sample. It cannot score the whole domain, count all indexed pages or prove deindexing when no pages return.

DeepSeek supplies an authority tier from 0 to 4, with −1 for unknown, and a relevance score called rerank_score from 0 to 3. These scores do not tell you the chance of a chatbot citing a page or independently establish its quality.

Dates

Search results are grouped by their dates relative to the run: previous 365 days, older, unknown or invalid, and future. Dates for matching pages are shown separately. These dates come from DeepSeek; the tool does not verify publication dates.

Actual date-filter behavior is not tested automatically. No after: or before: query is added. An unknown date does not prove a filter would exclude the page.

The text DeepSeek reads

Fetch requests your URL and shows the title and text DeepSeek returns. Read it. A long response can still miss the product copy you care about.

An error or empty text can have several causes; it does not prove blocking. Fetch does not check search inclusion, crawler permissions or JavaScript rendering against a browser. It has no location setting. The returned text is displayed as text, never executed.

What location hints can tell you

Default runs send no location hint and are labelled Unlocalized. DeepSeek may use its own defaults. Neither that label nor the server's location makes this a controlled country test.

Experimental comparisons send the same questions with a public IPv4 hint for India, Japan or China. Check each IP's country independently and record who verified it, when and using which source. The tool rejects private or reserved addresses but cannot verify an IP's country.

DeepSeek does not confirm the location it applies. An IP hint is not a VPN and does not reproduce a local user's account or search experience. Country labels describe the hints you supplied.

The first table shows successful searches and errors under each hint. The same-question table uses only category questions that succeeded under every hint. A failed question is removed from every hint in that comparison. The table is unavailable until the run finishes, or if no questions succeeded under all hints.

One run shows differences between requests. It does not establish a country effect, statistical significance or a trend. Keep questions, scopes, language, depth and IPs constant for repeat tests. Test translated questions in separate runs.

Download reports

• Team report / PDF downloads an HTML report. Open it and choose Print → Save as PDF.
• Text briefing downloads a Markdown summary with the settings, results and limits.
• Spreadsheet CSV contains result rows marked search, authority or fetch. Page matches and website matches are separate. Calculate presence per successful category question, not as a share of all result rows.
• Full evidence JSON includes settings, IP verifications, raw responses, Fetch text and errors. Check it before external sharing.

Downloads include the whole run, even when you filter the screen. "Reuse these settings" copies a run back into the form for review. Older reports keep their original scope and limits.

Saved runs and API allowance

No login is needed. A secure cookie lets this browser open its reports for 30 days. Clearing cookies or switching browsers loses access. A shared browser shares report access. Existing named-account reports remain private.

You can reserve up to 50 calls per run, 100 per browser session or named account per day, and 200 shared per day. Two runs can be active at once. The daily allowance resets at 00:00 UTC, or 05:30 IST. Clearing cookies cannot reset the shared limit.

Each question and location hint needs one search call. Each authority check and Fetch check adds one call. The full batch is reserved before starting, including calls that fail or are never collected.

The cost estimate uses $0.75 per 1,000 billable Search or Fetch calls, checked on 22 October 2026. Failed calls, empty search results and timeouts are not billed by DeepSeek, but still use the tool's reserved allowance. This estimate is not an invoice and does not cover other clients using the same API key. Calls are not retried automatically.

A run continues if you close the page. "Complete" means every planned call was attempted; check for errors before sharing. An interrupted run keeps the results already recorded and its reserved allowance. It does not resume automatically.

The key stays on the server. Questions and selected URLs go to DeepSeek. Read the Search and Fetch API documentation.