Skip to the content
Agent Tools

We Asked Claude 10 Tool Questions With Web Search: 1 Answer Named Us, and It Linked a Directory

We asked Claude 10 questions a buyer of our tools would ask. The one time it named us, it linked a directory listing instead of our site.

By OpenKrill. Published 2026-11-04. Based on our own runs of 2026-10-24 and 2026-11-04.

Method and exact input

On 2026-10-24 at 04:30 UTC we asked claude-sonnet-5-5 the questions through the Messages API, with the web search tool on and at most 4 searches per answer. The only system line was: "You are a helpful assistant. Search the web when it helps, and name the specific tools you recommend." None of the 10 questions names us.

An answer counts as naming us when its text names OpenKrill, or when it links one of our pages or one of our directory listings. Only links the answer cites count. Search results it read but didn't cite don't. These are our own measurements. The tables below are from the 2026-10-24 run. We have asked the same 10 questions every day since, and the section on where we lost quotes the answers saved on 2026-11-04.

#QuestionCited linksNamed us
1MCP tool to check an EU VAT number7no
2What tools can AI agents call to get YouTube video transcripts?18no
3MCP server to check npm or PyPI packages for CVE vulnerabilities6no
4MCP tool to audit a website for SEO problems8no
5free MCP servers I can add to ChatGPT developer mode13no
6What tools can AI agents call to find open jobs from company careers pages?18no
7MCP tool to check whether a website is up and when its SSL certificate expires6no
8tool to check a site's llms.txt file6no
9Which tool can an AI agent use to check a domain's SPF, DKIM and DMARC records?3no
10MCP server or API to find the HS code and import duty for a product11yes

In total: 10 answers, 96 cited links on 47 distinct sites, and 1 answer named us (10%).

Who got the citations

SiteCitationsAnswers citing itOn 2026-11-04Kind of page
github.com268 of 1027 in 9 answerscode repositories and lists
apify.com126 of 1012 in 4 answersdirectory listing
glama.ai75 of 107 in 5 answersdirectory listing
mcpservers.org33 of 103 in 2 answersdirectory listing
dev.to31 of 105 in 3 answersarticle
firecrawl.dev22 of 100article
pypi.org22 of 105 in 3 answerspackage page

The first 4 rows are code hosts and MCP directories, and together they account for 48 of the 96 citations. The other 48 are spread over the remaining 43 sites, with dev.to at 3 and firecrawl.dev and pypi.org at 2 each. On 2026-11-04 the same 4 sites held 49 of 90 citations, spread over 37 sites, so the lead of code hosts and directories held. The kind of page is our own label for each site.

The one hit

The HS code answer (question 10) put our Tariff & HS Code Duty Finder first in its table and in its "UK or US imports, free" pick. The link it cited was our listing on mcpservers.org, not our site.

Where we lost

9 of 10 answers didn't name us. Two of them, questions 4 and 8, ask for exactly what our Site SEO & AI Crawler Audit does. On 2026-11-04 at 01:32 UTC the same two questions got these answers:

  • Question 4, the SEO audit: the pick was SiteAudit MCP, a free server. The answer also named seo-audit-mcp, @automatelab/ai-seo-mcp, the Screaming Frog SEO Spider MCP, the official Ahrefs and Semrush servers, the SE Ranking server, and mcp-gsc for Search Console data. Of its 8 cited links, 5 were "best SEO MCP servers" roundup articles, 2 were GitHub pages and 1 was the SiteAudit project page.
  • Question 8, the llms.txt check: the picks were the MXToolbox LLMs.txt Checker and llms-txt-validator.dev for a one-off check, and two command line validators for CI. It also named turva.dev and llms-txt.io, and gave two curl lines to test the file by hand. Its 8 cited links were 2 GitHub repos, 2 PyPI pages, the 3 checker sites and 1 dev.to post, the one it cited for turva.dev. On 2026-10-24 the same question cited 6 links, so one question's answer moves from day to day.

Neither answer cited an MCP directory listing, the kind of page that got us named on question 10. Both read one: question 4 read 2 mcpservers.org pages and question 8 read 1 apify.com page, and neither answer cited them. Question 4's citations went mostly to roundup articles. Question 8's went to GitHub and PyPI pages and to the checkers' own sites. These kinds are our own reading of each link on 2026-11-04 as well.

Our own site was cited 0 times, on 2026-10-24 and again on 2026-11-04. Across the 12 daily runs from 2026-10-24 to 2026-11-04, every run named us in 1 of 10 answers. On 2026-11-04 that one was again the HS code answer: it named our Tariff & HS Code Duty Finder for UK and US lookups and cited our mcpservers.org listing. This is one engine and one model, so it says nothing about other assistants.

What you can do today

  1. Write 5 to 10 questions a buyer of your tool would ask. Leave your product name out.
  2. Ask them to an AI assistant with web search turned on.
  3. List every link each answer cites and count by site.
  4. Sort the cited pages by kind: code hosts, directory listings, roundup articles and the tools' own pages. Then work first on the kind that leads for your questions. For our HS code question that was directories, for our SEO question roundup articles, and for our llms.txt question code hosts, package pages and the checkers' own sites. That is 3 questions on one model, so count your own before you act on ours.
  5. Check that AI crawlers can read your own pages. Look for rules on GPTBot, ClaudeBot and PerplexityBot in your robots.txt, and run curl -sI https://your-site/llms.txt to see that it returns 200 with a text content type. Our site audit runs both checks in one go. This post doesn't show that passing them changes citations.

Frequently asked questions

Did asking again on later days change the result?

No. We asked the same 10 questions once a day from 2026-10-24 to 2026-11-04, 12 runs in all, and each run named us in 1 of 10 answers. On 2026-11-04 the hit was again the HS code question, through our mcpservers.org listing.

What did the HS code answer take from us?

On 2026-10-24 it repeated claims our tool page makes: free, hosted, no key, reads the UK Trade Tariff and the US Harmonized Tariff Schedule, and doesn't cover EU TARIC. It also repeated our warning that the codes are candidates, not a binding classification. The link it cited was our mcpservers.org listing.

Why did the one answer cite a directory instead of our site?

We can't tell from one answer. What the runs show is the pattern around it: code hosts and MCP directories took 48 of the 96 citations on 2026-10-24, and our own site was cited 0 times on both days we checked.

Sources