Does AI know your company?

We send the same 3–5 buyer questions to each AI assistant that has an API key configured on this site. Examples: “What are the best {category} in {city}?” and “Which {category} companies would you recommend?” The business name is not part of those questions. In the same round we also ask what the assistant knows about the business. ChatGPT uses the OpenAI web search tool, Perplexity uses Sonar with web search, and Claude uses its web search tool. Assistants without a key are skipped. The score comes back in about 10 seconds.

Mention (70%). 100 points if your business is first in a list, minus 12 points for each step down (floor 20). 50 points if you are named but not in a list. 0 if you are not named. Matching ignores capitalization and suffixes such as Inc, LLC, and Ltd. Your domain counts as a mention when the answer identifies your business with it. A sentence that only repeats the name or domain while saying you were not found does not count.

Citation (30%). 100 points if that assistant cites your domain as a source, otherwise 0. This needs the website field. A source on a subdomain counts. A different site that merely contains your domain text does not.

One recommendation answer scores 70% of the mention points plus 30% of the citation points. Those answers are averaged. Knowing you is separate: an assistant counts when its answer names your business, including a shorter form of the name, clearly describes it, or cites your domain. UNKNOWN, a disclaimer, or a sentence that only repeats the name while saying you were not found does not count. The score on this page is three quarters that recommendation average and one quarter how many of the assistants that answered already know you. If every assistant knows you and none recommend you, the score is about 25. If fewer of them know you, that part is smaller in proportion. If none of them have heard of you and none recommend you, the score is about 0. A real recommendation still counts for most of the number. Timeouts and errors are left out, so a failed call does not look like a miss. If Claude’s web search returns no answer, that question is asked again without search. The free result shows the score, whether each assistant knows you, whether it recommends you, and a few businesses they named. It does not show the answers. The full report lists each question, the answer, what the assistant knows, the sources, and what to fix. If every assistant fails, or none are connected, the page says score temporarily unavailable and does not fill in a number. Identical checks are reused for 24 hours. Very short or generic names can match more often than a distinctive brand.