I started my experiment with Brave search a few weeks ago. I set it as default on all of my devices, Firefox on Linux and iOS. I’m switching from Startpage. This post is a collection of disorganized issue reports, some duplicates and plenty of my personal opinions (so quite fitting for this community).
Switching region could be easier
Google fallback mixing is enabled by some weird criteria and “Find elsewhere” doesn’t have startpage option
Video results cannot be disabled
Search goggles have interesting behaviour
Setting goggle as enabled by default
I disabled summarizer
Web discovery project
I noticed that switching region is quite annoying since it requires multiple clicks. I’m from Finland but I only occasionally wan’t Finnish results, as a software developer, Finnish results are meaningless, but sometimes I need to live like local.
2 Google fallback mixing is enabled by some weird criteria and “Find elsewhere” doesn’t have startpage option:
Even thought I subscribed to the brave search premium and I’ve tried to click the “find elsewhere” button I don’t see the option in my brave search settings on Firefox but got it almost immediately on brave browser.
I also really wish the “Find elsewhere” buttons in the middle of the results included Startpage, but understand if that’s due to some rate limiting on their end etc.
The video results are mostly from youtube. The video results would be fine addition if they did not take as much space on the results page and on my work video results are generally not something I look for. Often “the correct” result is in top 5 first results but the video results often take up that space being on the way.
I tried to derank videos by deranking youtube but got left with p*rn taking up the first results, more about this on next section.
In my opinion the search goggles is the best feature of brave search. It actually gives something fresh and useful.
However there are some things I noticed:
It could be that a set of results is returned from the brave index and reranking is applied
Or brave search index is legitimately lacking index of the sites I’m trying to boost and this causes the results to be very lacking
For example my “Unix + developer” goggle seems to work really well even though some improvement is still needed on my ended.
I tried to create a “personal search goggle” to derank such “social media” services which aren’t social for those who do not have account for that service. And derank those ones which I do not use and boost the ones I do.
While experimenting with this, I noticed that deranking youtube will cause the video result sometimes still to appear, but instead of showing peertube or other independent videos the videos are pure adult entertainment even for such search terms which have nothing to do with such content.
A hilarious example is searching for my student union of university of eastern finland, abbreviated to “isyy”, if youtube is deranked heavily, it results in p*rn. From this quite vague answer, I don’t know if that’s something already known or if there’s some way of report safe search issues: Brave safe search set to Strict and p*rn websites still come in the results - #7 by Mattches
Also, for some reason I was seemingly unable to derank Facebook results unless I completely discarded them.
While creating the personal goggle I got an idea, what if Fediverse was prioritized over proprietrary services. Brave brags being privacy friendly and they have an official mastodon profile, so I could hope for nice results.
I constructed the goggle by extracting domain lists from multiple fediverse instance lists and by applying boost to them. With other additions, like deranking for proprietrary social media, the goggle turned out to be over 3800 lines long.
However this “Fediverse boost” failed to return Fediverse results. For example “dog” query only returns one fediverse result. I assume this is caused because brave search index just does not have anything from my hyper specific sources or the default ranking overrides so heavily. Or the goggle is applied on top of an existing result collection, meaning my additional boosts didn’t return new results from the index and since the initial collection the goggle was applied to didn’t contain enough results to have more fediverse results, meaning no more results could’ve been returned.
I then tried to disable all other results not coming from these three and half thousand fediverse domains and I feel like something totally went wrong and discarding all other results didn’t work, so I ended up downranking all domains and boosting fediverse but this didn’t seem to work either. I named this “Fediverse only” goggle.
All the sources are here: https://codeberg.org/jimbo/bravegoggle/src/branch/main
I didn’t publish any of them since I feel like either something is broken or I didn’t understand some part of the docs. But it’s still beta so understandable I guess.
I’m still planning to leave the Unix + developer goggle enabled by default since it actually gives a lot of value already, even though rough. The “enablement by default” is a hack, described below.
This is just a tip to make goggles even more useful. I for example enabled my work goggle this way on my work computer. I think brave search could have separate opensearch definitions for the goggles, meaning from browsers perspective each goggle would look like completely separate search engine.
An interesting approach would be to make these “search profiles” shareable. Meaning sharing such would include the region, settings, goggles etc. This way I could create my own privacy respecting way to sync my settings without locking these features behind some proprietrary account system.
browser.urlbar.update2.engineAliasRefreshand toggle it to true
Go to settings → search
On the search engines there should be “Add” button
The url should be:
Save and set your preferences as you normally would
Summarizer results were often quite long and since I cannot trust text generating systems (meaning not only brave), I decided to disable it.
I decided to go with the dark appearance since it fits my system, but it lacks contrast. The light theme doesn’t seem too contrasty either. I recently did some investigation on another website and it turns out that in some cases, at least on Linux, Firefox rendering is less contrasty compared to WebKit or Chromium, in this case however I don’t see huge difference.
I’m not asking for that e-ink like high contrast appearance but an improvement to the result item descriptions and result item paths.
There are two sides for this issue, I feel like I’m able to read the content unlike in some cases where contrast has been an issue, but in this case the Lighthouse test seems to flag quite many of the contrast issues, so there’s measurable reason to consider this as a problem.
I did go through the code a little bit, however didn’t spend that much time to get proper understanding what’s going on. Even thouhg it claims to be private and share the data anonymously, it’s like having a keylogger or backdoor in encryption. If it works it works, but having such in the first place is a huge risk for malicious misuse.
As an idea, it seems great, but I don’t think there’s a way to implement it truly privately and that’s why I really wish there was another approach. But there may be limits in terms of computing costs since at least the web frontend runs on AWS . Luckily there’s the premium subscription (genuinely rather direct payments than ads since those AWS bills are big)
I was quite happy with the results. However I sometimes still fall back on Startpage on my daily queries.
And for those hyper specifig use cases, wrapping something like stormcrawler on top of elasticsearch is still probably more transparent than using goggles, but goggles definitely do have something which is easy to approach and at this stage of brave search, it is the best feature of the entire service in my subjective opinion.