Kagi Search Empowers AI Models to Perform Better?

Pleonasm

Kagi asserts that "AI models can perform up to 80% better when sourcing data using Kagi Search compared to other search engines." What is the evidence for this assertion?

To learn more, I asked the Kagi Research Assistant this question: “Does the use of Kagi Search improve the quality of AI answers?” and it replied as follows…

Kagi prominently claims that "AI models can perform up to 80% better when sourcing data using Kagi Search compared to other search engines". However, my research found no independent studies or peer-reviewed research that validates this specific claim. The statement appears to be a marketing assertion without publicly available supporting methodology or data.

It is an important point, in my opinion. If Kagi truly has evidence that Kagi Search (as opposed to the use of other search engines) empowers AI models to perform meaningfully better, then that fact becomes a strong point-of-differentiation for Kagi Assistant versus the competition.

P.S.: I am sincerely curious about this claim by Kagi, and am not seeking to indirectly criticize the company.

jadams9

The only public benchmarks comparing it I've seen is their blog post: https://blog.kagi.com/kagi-assistants
In the section "Kagi Tops SimpleQA, then gives up". I'd be interested if anyone knows of any other comparisons. Anecdotally I do think I get better results in kagi assistant, than using the same models with Linkup websearch tool.

Pleonasm

If Kagi wishes to strengthen its claim that "AI models can perform up to 80% better when sourcing data using Kagi Search,” it could do so by hiring an independent marketing research firm to conduct an experiment.

In this experiment, consumers would evaluate pairs of query result outputs from an AI model (ideally Kagi Assistant itself), where one output is generated using Kagi Search and the other is generated using only Google search. Participants would be asked to identify which output was preferred on a rating scale (e.g., Strongly Prefer Output 1, Prefer Output 1, Prefer Neither, Prefer Output 2, Strongly Prefer Output 2).

In this way, Kagi would (hopefully) have evidence that AI outperforms when it is powered by Kagi Search.