Didn't like a Quick Answer response? post it here

Thibaultmol

To gather feedback on improving Quick Answer we made this thread where you can share screenshots of your quick answer responses that you didn't like the answer of. Please post an explainer for each screenshot and what you didn't like about it.
(This thread purely about the text response, not about the interface / UX around Quick Answer.)

xJustxMegx

Hi,

The answer should be in Dutch and for me only the second part is relevant and answers the question.

mpldr

Not sure if it's because the Model has been changed but the "new" Quick answers provides useless answers rather often while being overly verbose.

A quick example:

I asked what happens when I delete a branch.

a quick answer to my question, not to a different question. It's become something of a habit now to copy the quick-answer prompt, paste it into assistant and have it generate an answer, which usually works significantly better. (see https://kagi.com/assistant/774d20ab-ead3-4b0a-8492-392fd5137121)

mphi

Quick Answer Response (+additional followup question):
https://kagi.com/assistant/afacbcd0-38ae-48c9-ad47-510aa93ef457

It answered everything, but what I looked for - and added some filler word which are not specific enough to help with anything.

Especially the "Switching to DHCP" part is really bad: "This is done using the nmcli command".
I specifically asked for a nmcli command, and it tells me to use nmcli - thanks :-D.

And then it gives me a command to add a new connection, because DHCP is default for new connections... what?

After follow-up question, it got it right (that is Ki Quick then though).

pophob

Summary quality for this search query is poor despite all of the needed info being in the first Kagi result.

This result from Google is much closer to what I was looking for

deafbeef

Google directly returns formatReadableSize, while kagi does not mention it.

https://kagi.com/search?q=clickhouse+format+int+as++bytes%3F
vs
https://www.google.com/search?q=clickhouse+format+int+as+bytes&sca_esv=e5cbd14f40937cac&udm=50&fbs=AIIjpHw2KGh6wpocn18KLjPMw8n5ofjZQFej7XxDpLEwzn5636jPhZQ2SCHYNZOTHhCqX-lIANdm34OLsXpljMXvur-qOVxvA3NXiaKxlYQy-RRiasLGkzd4zqCUc8t8vI2F-h1nJA1utTYvqu2kpTmHUtQ7kaUGVvJcbPCfc9Trwsl4hdZsOuhb1DwK7z1ZfvqQTYc6nJeD&aep=1&ntc=1&sa=X&ved=2ahUKEwj_rraK1tuRAxX3RPEDHaPdGb8Q2J8OegQICxAE&biw=2051&bih=1031&dpr=1.82

Improve quality 🙂

TomatootamoT

This answer simply isnt correct. First site had the correct answer.

mpldr

TomatootamoT This is indeed the answer to your question. If you want 4% off, then it is indeed the opposite of what you wanted, but not of what you asked.

TomatootamoT

mpldr I want to know what the 600 is 4% of. Not what it four percent of 600.
The correct answer : 600 is 4% of 15,000.

mphi

Just give me the command switch. I know what it is, that's why I said "use". Just ssh -A somewhere in the answer as code block would have been perfect, but no, endless generic blah-blah.

Google KI mode does comparably better - everything I really wanted from the Quick Answer and more. It gave overview, examples for setup, usage and ssh config file. Nice. It probably saved somewhere that I can't remember simple command - not so nice.

Thibaultmol

TomatootamoT i'm afraid that's just a badly formulated question. I also didn't get it

Zambyte

Thibaultmol the question makes sense, and the quick answer is still wrong. This progression may help with understanding the question:

What is 6 half of?
- 12
What is 6 50% of?
- 12
What is 600 50% of?
- 1200
What is 600 4% of?
- 15000

For what it's worth, I checked Claude 4.5 Haiku, Qwen3-235B, GPT OSS 120B, and even Mistral Small (!!) and it each were able to answer correctly, so the question is perfectly reasonable.

igrek

Quick started gaslighting instead of answering a specific question:

Expected answer:

igrek

Literally on the same page search result:

westerntragopan

The response has three significant issues: (1) It incorrectly identifies the "next scheduled match" as Wolverhampton Wanderers on January 24th, when the actual next match is against Exeter City in the FA Cup on January 10th—a game that's even listed in the response's own "upcoming fixtures" section. (2) It mentions Manchester City playing Brighton & Hove Albion on January 7th as a future fixture, but this match already occurred yesterday (1-1 draw) since today is January 8th. (3) The response shows internal inconsistency by listing multiple matches that occur before January 24th (Exeter on Jan 10th, Newcastle on Jan 13th, Man United on Jan 17th, etc.) while still claiming Wolves on the 24th is "next"—suggesting the AI may be incorrectly prioritizing Premier League fixtures over cup competitions instead of sorting chronologically.

humphreyyy

This is just wrong. See here the correct response from kagi assistant with opus. I wish we could select a different model for quick answers..