Everyday more than 5.6 billion searches happen on the Worldwide Web. There are so many search engines for users and search algorithms that have evolved, following the quality and depth of searches, and the information that is searched for.

Broadly, search queries are classified around education, entertainment, and shopping.

Evolving from prototype to prototype, search engines have moved on from text to enable conversational searches that are aided by AI. Since its inception in 1998, Google has been synonymous with Internet search worldwide. Yahoo and Bing followed. Now, the same companies have launched generative AI-driven search engines and more keep coming up: SearchGPT, AI Overviews, Bing AI, Perplexity AI, DeepSeek, Grok - a mad race is on for minimising searches and producing personalised results.

Technology development is indeed a source of excitement to all. But, “The primary enthusiasts for search AI are knowledge workers—students, researchers, and professionalsseeking concise, direct answers without navigating through multiple links,” says Jibu Elias, Country Lead, RCC, Mozilla Foundation or Fellow at Mozilla Foundation.

Workings of search AI

When a user posts a query, the search AI processes the question in natural language- that is without keywords, interprets the entire query and not individual keywords, understands the relationship between words, phrases, scours the internet, and brings contextual responses that directly solves the query. The questions need not be simple but can be highly specific ranging from, “I am going abroad; what can I see there?” to “The best brand smart TVs in 2025.”

Different search engines cater different user intents. This is a time of reimagined search, says Mrutyunjaya Mohapatra, AI product manager, Humantic AI, a personality assessment start-up. “The web-results do not pertain to text responses alone. There are images, videos, dictionary answers, quotable quotes, pictorial proofs, data to support the responses,” he tells us. Of course, this saves the time of the user and the results are also well-aligned with the query.

Though AI-driven search enginesenhance user experience, they also introduce variability and potential biases. Based on the users’ location, online presence/behavior, preferences and internet footprint, AI alters its algorithm to map the context of the query and produce relevant responses. “The lack of consistent answers can affect the reliability of information, thus emphasizing the need for mechanisms that ensure accuracy and fairness in AI outputs,” explains Elias.

Dileep Jacob, Co-Founder, Nofrills.ai, says the world’s first search engine supporting Indic languages, “Nofrills does not use Google or traditional search Application Programming Interfaces (API), the interface that connects computers or software programs to each other. Our search engine is built using a classifier model that is assigned the URLs of websites and data in different classes such as historical data,current affairs,Wolfram Alpha style facts,PubMEd Factual Data from where the model can directly generate content. This helps us to update accurate information. Our search algorithm is embedded with its own indexing servers and spiders- programs used to scan the web for relevant content. With our architecture, we provide relevant answers to user queries without altering the search algorithms.”

Search AI’s use-cases

Basically, any application that is designed through aggregation of data might become a use-case for search AI. In the market, edtech, travel and hospitality, healthcare, media, marketing and gaming, and data-oriented research have heavy uses for search AI. Thinkers, founders, small business owners are keen to build such an AI that caters to these data-rich sectors.Indeed, search AI’s capabilities are vast. Not just answers but the AI- driven engine can give you insights, data, and metadata for accurate and personalised results.

Zero-click and other challenges

Even though AI is touted to be a super force, building and deploying AI agents and running them at scale is faced with intricate problems.

The first and foremost are data privacy and intellectual property.

“Large Language Models (LLMs) reproduce proprietary content without permission, and this underscores the urgent need for clear regulations and ethical guidelines in AI deployment. AI-driven search engines are providing answers directly without redirecting to source websites. This ‘zero-click’ approach threatens traditional web traffic models, potentially impacting publishers and content creators who rely on site visits for revenue. The challenge lies in balancing user convenience with the sustainability of content ecosystems.”

Jacob states a problem faced by Nofrills.ai. Indic languages, in which Nofrills disseminates content requires approximately 10 times more tokenization- breaking down a stream of text into smaller pieces to convey the same message as in English. An Indian tokenizer hasmany letters and complex geometrical structures, because, in Indic languages, words change form depending on gender,tense and use case.This requires substantial change to transformer architecture.The limited availability of digital data also adds to this challenge.” Jacob adds that Nofrills.ai has addressed this by adding a language helper and translation helper tools.

“Talentto build efficient AI and solve problems is scarce,” says Mohapatra. “It is very tough to get augmented data to train AI for a seamless performance,” he adds.

Search AI’s user base is growing

Yet, the adoption is growing slowly yet steadily; while conversational AI search engine users in the U.S have grown from 1% to 8%, the growth rate is still higher in India. As per IMARC Group, a market research firm, it is expected to reach more than $4.9 billion in 2033. Perplexity AI, launched in 2025 has reached 15 million monthly active users as of late 2024, up from 10 million earlier that year. ChatGPT has over 300 million weekly users, and Bing AI has 100 million everyday users. These new AI-driven search engines have challenged Google’s dominance too with its market share dropping to 89.98% in November 2024 from 91.38% in March 2024. With all the technological advancement, points out Mohapatra, “In the next five years or so, we could land a single AI interface on a phone or laptop that assists individuals as the search-and-task-disseminator, a personal bot.”

“The ease of assigning tasks and the accessibility to information are the keys; the additional tools of the search engines are an absolute bonus,” says Vaishnavi, RA in a global development company. “Especially in shopping, instead of search-writing specific designs and styles- for example, black top with sequins and a flared bottom, I can simply upload the picture and buy it online,” she adds.

Does AI mean only doom?

No, says Mohapatra. Not in all domains. In fact, Generative AI is aiding great strides in software development and Natural language Processing (NLP) and Predictive analytics, the latter being very impactful in healthcare industries.“It has simplifiedand accelerated software development processes fornow and in the long run it is likely to support system designand develop efficient software systems to take over menial tasks, solve complex issues and boost productivity,” he tells us. 

Thanks to generative AI- so many people have gained access to knowledge and a platform to get quality, reliable and up to date information for any topic of their choice, says Subash Sasidhara Kurup,CEO of Norfills.ai. “Further, we want to break down linguistic barriers and make AI progress accessible to all,empowering individuals from diverse backgrounds to explore, learn and grow.” 

When it comes to search AI agents, it is important to raise questions about accountability and decision-making authority, says Elias. “Ensuring that AI agents operate within ethical boundaries and under human oversight will be crucial as we navigate this new landscape,” insists Elias.

“That is why we are focussing on building smaller language models or SLMs for specific tasks and training AI on very limited and relevant data,” says Mohapatra.