Request
This has been mentioned in previous related forum posts, but my feature request is to make the AI Quick Answer feature Opt In by default- i.e. a user has to choose to add Quick Answer to search results, not the current Opt out pattern.
Support impacts from current Opt out pattern
There are folks who are excited to use and try LLMs. There are many more folks who don't know what it is at all, and don't really care. There are also a number of folks who want absolutely nothing to do with them, never want to interact with them, and become increasingly irate when they are forced to interact with them. Managing the emotions of the latter group is service labor that Kagi employees will need to perform in these forums as long as Quick Answer is Opt Out by default.
As Kagi grows its user-base, there will be an increasing number of folks who don't work in tech but do care about the quality of their search results. They will be confused when Quick Answer gives them a completely wrong answer and assume that this is the fault of Kagi in general (not the LLM) being used. New Bugs will be filed in this forum where the root is the LLM that is outside of Kagi's control. This isn't a good user experience and has the potential to weigh down internal Kagi resources answering bug reports.
General business risks to current Opt out pattern
LLMs will never not "hallucinate," and research shows that their inability to summarize correctly can "mislead audiences." This is a risk to user trust in Kagi (again, because regular working people will not recognize the difference between Kagi and an LLM. Kagi is using an LLM, therefore Kagi owns the answer from the LLM by their logic). Research is also beginning to show that as user trust in AI increases, critical evaluation of answers decreases, which means there's a real risk that someone will believe a dangerously wrong answer generated by an LLM and tragically harm themselves or others as result.
Opt out patterns are considered by many a "dark pattern" that can be deceptive to users. Government bodies (FTC, EU) have been releasing research about these design practices and how they impact consumers.
Default setting changes are hard
I understand it is very difficult to make a switch in default behavior, but now might be the best time to make this change before Kagi's user base becomes even larger.
Desired future behavior
When a new user starts to use Kagi, their results should look the way they currently do for any user that has the AI Quick Answer setting turned off:
If a user wants to use Quick Answer, they can click it. There can be text mentioning that they can have Quick Answer always on by updating their AI Settings, with a link to their settings.
I would also suggest bolder, more clear text to the beginning of a Quick Answer that mentions these answers are generated using language models built on statistical probabilities of parsed historical texts, and therefore can be wrong. This way users are presented information that could help prevent unconscious blind trust in the accuracy of the answer.
Thanks for your time reading this request!
