nobodywasishere
Thank you for replying. The issue with the first query is not only the number of results. It is a combination of result completeness, exact-phrase matching, and filetype:pdf not being applied consistently.
The query was:
filetype:pdf "curriculum vitae" "ireland"
Google returned many more results, and I manually checked all of its result pages. In that specific case, the results I checked were relevant: they were PDFs and contained both exact quoted phrases.
On Kagi, the same query returns substantially fewer results. Using “More Results” increases the number, but it still does not reach the set of relevant results found through Google. At the same time, some Kagi results do not appear to respect the requested file type: I see .doc files and ordinary webpages rather than PDFs.
So, to answer your question directly, the problem is:
The number of matching results is much lower than Google’s for this specific query. It would be very helpful if Kagi - especially as a paid service - returned more of these relevant results and could serve as a more complete alternative to Google for this type of search. In this case, I need to find and compare other people’s résumés with my own, so I still have to use Google to check for additional relevant documents. Ideally, Kagi should aim to ensure that users do not need to return to Google to verify whether a constrained search has missed important results.
filetype:pdf is not always respected: some returned results are .doc files or normal webpages rather than PDF documents.- Some results returned by Kagi appear less relevant than PDF results omitted by Kagi but returned by Google. For example, Kagi may return a
.doc file or an ordinary webpage despite the filetype:pdf filter, while omitting PDFs that contain both exact quoted phrases.
- Since
filetype:pdf, "curriculum vitae", and "ireland" are explicit constraints, I would expect Kagi to return all - or at least substantially more - known matching PDF results before returning items that do not satisfy the requested file type.
I am not saying Google is always better or that every Google result is relevant. I am describing a reproducible case where Google returned a larger set of manually verified, constraint-compliant results, while Kagi both omitted many of them and included results that appear not to meet the filetype:pdf constraint.
Regarding the second query:
"useful" "Apple" "It is crucial that the system tailors" "ireland" filetype:pdf
Thank you for explaining that Kagi post-processes key terms after obtaining results from upstream providers, and that it cannot fetch every candidate page within the available time budget. I understand the performance trade-off.
However, this also appears to explain the failure mode: Kagi may remove a result that really does contain all quoted phrases because the title, URL, or generated snippet does not expose every phrase. In this case, the target PDF contains the requested text, but Kagi returns no result for the complete query.
That means the post-processing can produce false negatives for exact-phrase searches, especially for PDFs and other documents whose snippets may be incomplete or poorly generated. A result should not be excluded merely because every quoted term is not visible in its title, URL, or snippet if the upstream source identified it as a candidate match. Google wins here. ='(
TL;DR: For this kind of research query, I prefer recall and constraint compliance over aggressive filtering (well, I guess everybody would prefer this here, and if you were in my situation, you would too). If I ask for PDFs containing exact phrases, I would rather see all plausible matching PDFs - including lower-ranked ones - than receive a shorter list containing non-PDF or less relevant results while known matching PDFs are omitted.
You should check yourself the Google results that were missing from Kagi if that would help.
