Battle of the privacy-focused search engines: Kagi vs. DuckDuckGo
Google’s search engine currently handles over 90% of the world’s web searches. Microsoft’s Bing handles about 3%. A bunch of comparatively niche search engines fight over the remaining 7%.
A problem with Google’s search, and Microsoft’s to a lesser extent, is that these companies use tracking technology to record your entire search history. They use that information to supply their colossal advertising infrastructure. Many people don’t want their browsing history recorded and, even if you don’t worry about this, you might not want to contribute to turning the whole web into an advertising banner.
Consequently, we need a search engine that respects our privacy – one that doesn’t associate search requests with a user, nor share user profiles with advertisers. This article is about two such search engines: DuckDuckGo and Kagi. A relative newcomer, Brave, also merits consideration, but that’s a job for another day.
Funding and subscription
Running a search engine is expensive – colossally expensive, if you do the whole job yourself. We know how Google and Microsoft fund their search engines: by monetizing users’ personal information. What about search services that don’t do that?
Kagi’s funding model is easy to understand: it’s a paid, subscription service. It doesn’t rely on advertising in any way, only on the fees its users pay.
DuckDuckGo, however, is free to use because it’s supported by advertising. It isn’t targeted advertising a la Google, but it’s still advertising. DDG’s advertising is driven entirely by the search query, not the user’s history (which DDG does not store). So, for example, if you search for “Volkswagen Polo” you might see links to sites that sell these cars, but you shouldn’t see advertisements for things you saw in your browser yesterday.
Presenting sponsored links in search results can be irritating but, if you’re in the market for a new car, it might be useful. If you click a sponsored link, DDG gets a tiny financial reward from the advertiser. Is it enough to fund the service? It seems to be: DDG has been profitable for at least ten years. At one point it received commercial sponsorship from companies that wanted to provide their staff with a non-advertising web search; I haven’t been able to find out whether this is still the case.
DDG allows users to turn off even its own non-targeted advertising links. In practice, I find that I don’t see a lot of ads with DDG, even when I leave advertising turned on. This might be because I don’t search for things that people sell; but DDG advertising seems to be pretty non-intrusive, by all accounts.
Kagi’s basic subscription costs $54 per year, which allows for up to 3600 searches. Since I sometimes make more searches than that in a month, I need the more expensive “professional” service, at $108 per year. This level of service also provides access to Kagi’s translation services, which I’ve found to be quite good. There’s an even more expensive service tier that includes additional, premium AI support. As yet, I’ve not found much use for AI-based search, but I appreciate that some people like this.
When I was a working man, I wouldn’t have hesitated a moment to subscribe to Kagi for $108 per year. Now I’m retired, it’s an outlay that I had to think hard about. You really have to be sure that Kagi offers better search results than DDG, to pay even this modest amount. Fortunately, Kagi offers a free trial, so you can find out.
While DDG is free to use, it does have a subscription tier, which provides a VPN, and extended AI support. I haven’t used it, so I don’t know if it represents good value at $99 per year.
It seems to me that both DDG and Kagi balance on a financial tightrope. They both serve a comparatively niche market, and small changes to their user base could have profound operational consequences. Still, both are hanging in there, and seem to be doing reasonably well.
Kagi’s subscription model requires the user to authenticate, which is a potential source of problems that I’ll discuss later.
In the baldest financial terms, DDG wins over Kagi because it’s free of charge. Kagi isn’t terribly expensive, but free is a tough price to beat.
Privacy issues
DuckDuckGo’s privacy model is very simple: it doesn’t record anything about user interaction except a settings cookie. Removing this cookie will lose any custom settings you’ve made, but it won’t stop the search engine working. Unless you’ve made significant custom settings, it won’t be a problem if you configure your browser to dump all cookies periodically. DDG doesn’t save any search history, or bias the search results according to previous user behaviour. It’s pretty much as private as a search engine can possibly be.
Neither Kagi nor DDG uses search history or user behaviour to bias the search results, which means that you won’t get the algorithmic “confirmation bias” you see with Google’s search. Google’s bias can actually be useful, though, as well as harmful. Kagi compensates for the statelessness of its user interaction by allowing the user to add intentional bias, as I’ll explain later.
Kagi follows the same privacy model as DDG to the extent that it can. However, being a subscription service, it needs to know the user has paid, so it has to provide a way to authenticate, and to store the authentication status. It uses a cookie for this, as most websites do. Dumping the cookie won’t hurt your relationship with Kagi, but you’ll either have to log in again next time you want to do a search, or find some other way to maintain a logged in session (more on this later).
Kagi loses slightly to DDG in privacy because of the need to associate payment details with an account. You’ll need a working email address, although it can be a burner. Still, if you use a credit card for payment, Kagi will receive credit card details and be able to match them to an account, should it wish. This means that Kagi users rely for privacy on the integrity of Kagi’s staff. I see no reason to doubt Kagi in this respect but, if you’re not convinced, Kagi takes anonymous payment using cryptocurrencies and the like.
In short, I don’t really see much to choose between DDG and Kagi on privacy grounds. With both services, you’re trusting that the companies will comply with their own privacy policies. However, Kagi handles a little more personal information than DDG, so breaches of its privacy policy could be a bit more serious, and it’s slightly more of a viable target for villainous hackers.
Search performance and ranking
No search engine is neutral; they all present their results in the form of a list, so they have to make decisions on ordering. Neither DDG nor Kagi operate completely independently of other search services (more on this later), so there will always be some inherent pre-selection of results. However, DDG and Kagi also apply their own rankings.
Users judge the effectiveness of a search engine by how well it delivers relevant content. The challenge for the search provider is to work out how to do that. All the major search engines, including Google’s, say they up-rank sites that contain “genuine content”, and not just click-bait and AI slop. DuckDuckGo got into trouble in 2022, when it was seen to down-rank sites that it deemed to contain “misinformation”, but some users welcomed this move. Today DDG’s approach is to down-rank sites that exhibit “low journalistic standards” which it assesses in a number of ways. I presume that deliberate disinformation remains an indicator of poor journalism.
Kagi favours sites that limit advertising and tracking; this, again, is something that many users welcome, but others see as a kind of censorship. It also down-ranks sites that appear to be AI-generated. Kagi consistently includes more hits from my own site than other search engines do, so it must be measuring quality properly ;) Ahem.
It’s almost impossible to rate search engines on effectiveness, other than on entirely subjective grounds. I’d like to give just one example which is, of course, not at all representative. I’ve been having trouble with the fit of my rucksack when hiking, and I did a web search for “adjust rucksack”. For the sake of comparison, I used DDG, Kagi, and Google.
All three engines returned what I consider to be useful results. Google, as you might expect, filled at least 60% of each page with links to sites that sold rucksacks, which wasn’t at all what I wanted. Still, I found useful results in the first three pages, when I’d mentally filtered out all the shopping links.
Kagi and DDG both returned the same links as one another in the first two pages, with no links to shopping sites. What I considered to be the most useful link for my purposes was the first one Kagi showed; that same link was somewhat lower down in the results from DDG, but it was there.
I’ve found this to be broadly true of most searches: for my purposes, Kagi usually puts the links I find most useful a little higher up the result list. Still, I’ve generally found that Kagi and DDG show the same links in the first two pages – unless I apply a lens in Kagi.
To my mind, lenses are one of Kagi’s most useful features. You can deliberately bias the search results according to location, date, keywords, and other things. You can then name this group of bias settings, and apply the same bias to any search. So, for example, I have a “Recent UK” lens, that emphasises results from the UK in the last week. There are also built-in lenses that favour forum posts, news sites, and other things.
I can filter results by criteria in DDG, but I can’t group the settings and name them for future use. If you use custom search biases – and I do, all the time – Kagi is more convenient than DDG.
Kagi’s lens settings are stored as part of the user’s account. If you’re a hard-core privacy enthusiast, you might balk at Kagi storing even this small amount of personal information; but if you did, I think you’d be losing one of Kagi’s most useful features.
AI assistance
Both DuckDuckGo and Kagi provide “AI” assistance. You can type questions in natural language, and get answers from a bunch of commercial LLMs. In a sense, the search engines act as anonymizing proxies for the LLM providers. You’ll find all the usual suspects: ChatGPT, Gemini, and so on. To use the latest and best LLMs in either DDG or Kagi you’ll need a specific subscription.
Kagi’s “Quick” model answers most questions in under five seconds. Anything beyond that can take significant time. In my tests, waiting longer didn’t elicit noticeably better answers, but I guess this depends on the question.
I’m not an expert on the use of LLMs, and I’m definitely not a fan. The DDG and Kagi AI assistants both output reasonably well-written, apposite text. However – at least with the standard LLMs in the base subscription – none of the available models could answer questions that required significant research. If I want general information with reasonable veracity about common subjects, I can get it from Wikipedia easily enough. Hard questions remain hard, with LLM support or without it.
I’m sure there’s a market for these “AI” tools, but I don’t think I’m part of it, so it’s hard for me to comment on which of DDG or Kagi performs better in this area.
Search sources
Let’s address the elephant in the sitting-room.
There’s only a handful of companies in the world with the resources to maintain a full index of the whole world-wide web. In the English language there’s really only Microsoft, Google, and Yandex. Yandex is problematic for a host of reasons, including its connection to the Russian state. DuckDuckGo and Kagi maintain a partial index, as does the newer Brave search engine.
Over time these indexes will undoubtedly expand, but it’s worth keeping in mind that it took Microsoft, with all its colossal resources, nearly twenty years to develop a search index that can begin to compete with Google’s.
This means that most most English-language search services use Microsoft and Google as back-ends for general searches, and perhaps Yandex.
That doesn’t mean they’re limited to these sources, of course. DuckDuckGo says that it uses 400 different information sources, including Wikipedia, TripAdvisor, and Wolfram Alpha. Kagi also uses multiple information sources, not very different to DDG’s.
However, these sources are for specialist queries. For general search, DDG relies primarily on Microsoft’s Bing. At one time Kagi documented the search sources it used but, after a spate of complaints about some of its choices, Kagi now says only that…
Our search results also include anonymized API calls to all major search result providers worldwide, specialized search engines like Marginalia, and sources of vertical information such as Wolfram Alpha, Apple, Wikipedia…
It’s not clear what “all major search result providers” amounts to, but I’d be surprised if it didn’t include Microsoft or Google, or both.
DDG and Kagi both anonymize the searches they make on other providers, so I don’t think we need to be concerned that using Google, et al., as back-ends creates a privacy hazard. However there’s no getting away from the fact that using either DDG or Kagi amounts to funding the tech giants.
I wish it were not so, and I can only hope that, over time, independent search providers will develop their own indexing infrastructure that rivals those of Microsoft and Google. I doubt that will happen this year, and probably not this decade.
Translation, summarizing, and proof-reading
These services are not directly related to web search but, since the search developers are already invested in LLM technology, I guess they can readily use the same expertise to provide these supplementary services, all of which rely on LLMs.
So far as I know, DuckDuckGo does not offer any of these services. Kagi’s translation service seems to be at least as good as Google’s, inasmuch as I am able to judge. I tried some passages of classical and medieval Latin, which I’d been working on translating myself. Both Kagi and Google made a good job of the (medieval) Venerable Bede, producing readable English. Both stuggled with Virgil’s (classical) Aeneid, but that’s hardly surprising: Virgil’s highly-stylized poetry poses a challenge to human translators, too. I tried translating some of my website articles into French and German, with which I have some passing familiarity, and the results seemed good enough.
I believe that Kagi’s translation service is home-grown, based on several commercial LLMs. I don’t have any more information about how it works than that. It isn’t just acting as a proxy for Google Translate, because it doesn’t produce the same output.
Kagi’s summarizer will work with blocks of text, or whole web pages. I found it to be astonishingly good, when fed with competently-written text. To be honest, it’s not a tool I have much need for, but it’s impressive, nonetheless.
The proof-reading service is also impressive, although I found its suggested changes to my own writing to be slightly stuffy. Again, I don’t really use this kind of tool, but perhaps I should.
Authentication issues
Unless you’ve subscribed to get the more advanced AI support, DDG is free to use, and doesn’t require authentication. Kagi does, and this is a slight cause of irritation.
Kagi’s authentication status is stored in a long-lived cookie so, provided you allow your browser to retain such cookies, you’ll stay logged in.
Those of us who are really fussy about privacy, however, configure our browsers to dump cookies periodically. Doing this to Kagi’s cookie will log you out. If you want to stay logged in, you’ll need to tell your browser to exempt Kagi’s cookie from deletion.
This isn’t a big deal, really, except that I have to do it on every web browser on every device I use.
An alternative approach to maintaining authentication is to configure your browser to use Kagi with a session link: a random numeric parameter to add to the Kagi search URL, that automatically logs you in. This configuration is significantly more fiddly than simply telling the browser to keep Kagi’s cookie but, of course, the browser doesn’t need to store a cookie. In practice I use a combination of these (cookie and session link) techniques.
I see no way to avoid the minor inconvenience of authentication, if you’re using a subscription service. I presume the same problems would be present if you used DuckDuckGo with a subscription. Of course, for most purposes, you don’t need a subscription for DDG, and the problems of authentication don’t arise.
“Small web” support
The “small web” is that (ever-diminishing) portion of the world-wide web not dominated by advertising and commercial interests. It consists of websites (like this one) maintained by individuals and based on simple technologies.
Kagi’s search favours results from small web sites, and can be configured to show only those results.
I use this feature all the time: unless I’m actively shopping for something, I invariably get more apposite and interesting results from sites on the small web than from commercial operators.
Summary
DDG and Kagi are both privacy-respecting search engines. Although you’ll have to pay to use Kagi, you get more than just search for your money: Kagi offers excellent translation and summarizing services, for example. Both search services offer AI support, but I’m in no position to judge between them, because I don’t use them enough.
As much as I admire Kagi’s principles, I don’t find its search performance hugely superior to DuckDuckGo’s. Kagi is ad-free, but DDG doesn’t show me a lot of advertising, and I can turn off the little I do see. Judged subjectively, Kagi puts the most useful results (for me) a little closer to the top of the results list than DDG does, but both give me the same results in the first page or two. Named “lenses” are a killer feature of Kagi for me, along with its bias towards “small web” sites. I don’t know that I’d subscribe other than for these features, or unless I used the translation feature every day.
There’s also the minor irritation of needing to authenticate to use Kagi.
However, part of repudiating ad-tech’s assault on the Internet is being willing to pay for what you use. Every subscription to Kagi adds a drop of pure, clean water to the cesspool of poisonous bile the Internet has become.
