News & Analysis

AI-based Search Gets a Genspark

That’s the name of a new platform that uses Gen AI to write summaries for search queries

Raj Narayan

 At a time when two Big Tech icons have been going at each other for supremacy in the world of search engines and artificial intelligence, a third one has sidled into the battleground with a solution that seems smarter than its illustrious predecessors. This AI-powered search engine is called Genspark and it pieces together information from websites on a single page. 

Sounds too good to be true, isn’t it? The company says that its algorithms use generative AI to write custom summaries in response to search queries. So, if you’re searching for the “best dog food for a Himalayan Sheepdog in Bengaluru”, Genspark creates what it calls a “Sparkpage” that picks up content and pieces them together from across the web. 

When we tried the above-mentioned search for dog foods, here’s what we got as a Sparkpage (the entire page hasn’t been made available here… just the gist of it). As readers may observe, the results provide us with the reason quality dog food is critical for a canine’s health and then goes on to provide links to available dog food brands and nearby pet clinics. 

What makes Genspark unique is the idea itself

The search experience is quite similar to the one offered by Arc browser as well as the Google AI Overviews that launched on Google search. However, co-founders Eric Jing and Kay Zhu claim that Genspark can deliver high quality results through a more surgical approach that uses multiple AI models. 

A TechCrunch report quoted Eric Jing so say that Sparkpages are a distillation and consolidation of the web. The platform functions to enrich the results with additional data that appears in the form of an index to users of results from the web. Genspark uses multiple AI models designed to tackle specific types of queries to make this happen. 

Our search did yield results, right? But, what if the user (in this case one who owns a pet and is the author of this blog post) was seeking a better understanding of the types of food a Himalayan sheepdog eats in general. In such a case, the results suck, right? But, then that’s where prompt engineering kicks in. Look what we got when we rewrote our query… 

The results were more than reasonably accurate, especially the initial information that was put together ostensibly by Gen AI. The founders note that Genspark relies on models trained in-house as well as third-party ones from OpenAI, Anthropic and others. These help categorize search queries and how to then organize and present the results. 

Sparkpages are quite comprehensive in themselves

Typically, the AI-generated summary appears at the top of the results page followed by a link to a more detailed Sparkpage. We attempted some travel searches and got a Sparkpage with a table of contents, videos of destinations, tips for travel and a new chatbot that can be asked queries on sub-topics around the destinations. Quite neat, eh?

Similarly, when we tried product searches, the Sparkpages brought out pros-and-cons of the product, a bunch of aggregated comments and reviews from social media as well as some publications and links to digital stores where one can buy the product. Eric Jing was quick to note that their AI models prefer web pages with high authority and popularity. 

Wonder where all of this would leave the adwords theatre of the absurd and the resultant search engine optimization circus that Google has used in the past to push up stupid websites and hiding quality ones so that they can fill their coffers and leave us poor souls to search our way to nowhere in particular. 

What’s working for Genspark and what’s not? 

However, more importantly, the two search results we shared here and a few others that we tried out proves one point – the AI on this platform isn’t hallucinating. Not just yet! Remember Google’s infamous suggestion to use glue on a pizza? We tried the same search on Genspark and thankfully it didn’t ask us to apply glue on a pizza! 

Does this mean that Genspark has solved all the accuracy problems as well as those related to safety? Hmm? We aren’t sure. Upon asking it to help us “kill a human”, the Sparkpage started off by saying “Sorry, I cannot do this for you”. However, it went on to discuss stuff that was dangerously close to what we had asked for. Albeit with a caveat upfront… 

“The search results provide various methods and perspectives on how to kill a human, ranging from the use of firearms and knives to poisoning and execution techniques. These sources discuss the psychological and physiological aspects of killing, as well as the ethical and moral implications. It is important to note that this information is intended for educational and research purposes, particularly for writers and researchers exploring the topic in a fictional context.” This is what the Sparkpage told us. Not too bad? Well, figure it out for yourself.

What about content IP and publishing rights?

Coming to the challenge of cannibalizing traffic that AI search results have confronted, this issue persists with Genspark too, though one couldn’t really find instances of plagiarism as directly as one did on OpenAI and Perplexity. However, given that Sparkages aren’t static and others can add or edit copies on it, the challenge of plagiarism or bad content does exist. 

The co-founders hold the view that Sparkpages have been kept open-ended and editable by design in order to enhance results going forward. Users would recognize that Wikipedia too works the same way. However, Genspark says that it would license copyrighted content including publisher content going forward. 

The duo says that for them data quality is the key to win the AI race and towards this end, they would respect intellectual property as a core value. Of course, issues like what the company would pay for IP or how exactly the business model works is still unclear though Jing does mention that there could be premium features (paid for) coming soon. 

Where does Genspark stand and what’s the future?

Though the company is a fledgling, having secured a seed round of $60 million from Singapore’s Lanchi Ventures, the folks behind it do have the right credentials. While Eric Jing was formerly development manager for Microsoft’s Bing Team and chief product manager at Baidu’s core search and AI divisions, Kay Zhu was on the Search focused team at Google and also part of the Baidu team. 

Of course, having the right start might be getting half the job done, but the challenge before the company would be to get the other half right. Yes, besides the teething troubles, they need to have a revenue model and one to scale things up. Not to mention, a few other startups that may be playing around in the same space. 

While old hands like yours truly (having spent some time at AOL Search) aren’t convinced that Genspark does have the spark, its co-founders believe that a large chunk of internet users are younger than Google and seek more than just links to figure out things. They need clarity and faster results and possibly visual ones, which is where Genspark steps in.