What’s next for the search box
AI’s impact on the web’s most-used input
Since the early days of the web, the search box has been a daily fixture in our lives, allowing for the discovery and navigation of the internet’s sprawling stores of information. Search has consistently evolved over the past three decades, but no change has been so dramatic as the new conversational search paradigm introduced by ChatGPT.
The application of language models to create better chat-based search interfaces will have a broad impact across businesses, but applied incorrectly, will create more fragmented and frustrating experiences for users. Good UX planning and meaningful technical integration is critical to making conversational interfaces work for your users.
An abridged history of the search UI
Around 2000, Google created a simple and user-centric alternative to competing search portals, focused around an input box and a couple buttons. At the time, this was a much needed innovation, prioritizing fast load times, minimal bandwidth consumption, and quick search results. This, combined with the quality of search results, launched Google on its rise to search dominance.
Predictive text
The next innovation in the search box came in the form of predictive text, suggesting words or phrases to the user as they type. In addition to reducing the number of key presses needed to arrive at a search, the predictive text also provided users with a view of the search zeitgeist, a peek into the internet’s subconscious through the lens of common search phrases.
From search results to answers
Even with faster load times and less typing, searches still often felt like a lot of work. You had to actually perform the search, scan the results, find a website that looked reliable, and click through to find the information that you needed to actually answer your question. In 2012, Google released its Knowledge Graph, reducing the time-to-answer for certain types of information dramatically, in some cases before the user even finishes typing the search prompt.
I provide this history through the lens of Google’s UI because Google has, for the past 20 years, enjoyed a position of dominance in the search industry. Even if its rollout of these features were not the first, their impact has certainly been the broadest due to the sheer size of Google’s market share. That is, until…
Along came ChatGPT
ChatGPT, in partnership with Bing entered the search marketplace in 2023 as a new paradigm for search. Bing + ChatGPT uses Large Language Models (LLMs) to understand and respond to user searches in a conversational manner. The conversational search can summarize findings across multiple websites, provide annotations, and even answer contextual followup questions, all in chat format. Only 30 days after its release, Bing’s new chat has surpassed 100m daily active users.
And just like that, the search box has transformed into a chat interface. While this change hasn’t been without its bumps, it’s likely that the impact on user expectations for search will be widespread and irreversible.
One step forward, two steps back
Unlike many past innovations in search coming from big technology companies, the conversational model underlying the new Bing (along with alternative open-source and closed-source LLMs) is broadly accessible. This is a good thing — information that is scattered across the data silos of enterprises, organizations, and communities of the world can all benefit from better ways to make their data searchable and interactive to those who need it.
The problem is that the way that this technology tends to be incorporated into existing organizations and products results in a regression of the user experience. Take the following scenario as an example:
- Startups specializing in the new tech create a minimum viable product to help companies jump on the bandwagon. This is usually a bolt-on product that sits on top of an existing product or platform.
- Innovation and digital transformation leaders with a mandate to adopt new technology, but with limited staff and budget for implementation, work with the startup to “dip their toe in the water.”
- The bolt-on solution is incorporated as a test, without consideration of the impact to the user’s experience.
- You wind up with a user experience that looks like this:
- User engagement is low, the test is deemed unsuccessful, and the tech is written off or shelved
Doing things the right way
If you’re someone in charge of a product, service, or enterprise that uses search to expose information to users, it’s likely that you’re considering how to integrate LLMs like ChatGPT. And you should be. Rapid adoption of these technologies will mean that products and services using the “old” way of search will soon struggle with meeting user expectations.
But don’t just settle for a bolt-on. The next evolution of the search box should incorporate the best of what LLMs have to offer, without sacrificing the benefits of previous search innovations or adding clutter and confusion to the user experience.
Progressive Navigation Patterns
For e-commerce and retail destinations where users have fairly cemented expectations of how to browse and find things, the role of search is to be as useful as possible when required by the user. Progressive search patterns like the one pictured below start with traditional navigation, provide users with helpful suggestions once they begin a search, and then adapt to more open queries and questions with a conversational interface.
Conversational Navigation Patterns
For websites and products where search is the primary user intent (documentation sites, content portals, customer service portals, etc.), shifting to a conversation-first design is something to consider. By asking clarifying questions, providing a summary of indexed information, and presenting in-depth results, conversational search interfaces will help quickly answer specific questions while allowing users to delve in deeper where needed.
Multi-modal search patterns
One benefit of designing around conversational interfaces is that they can work in concert across multiple interfaces. A user interacting with a voice assistant can receive a spoken summary, a text with a link, and an emailed discount code relative to a conversational search.
While implementation will vary depending on the needs of each company or product and their users, I urge innovation and technology leaders to think twice before going for a quick off-the-shelf solution. This technology can truly be transformative for your customers, employees, and users, but execution matters. Starting with understanding your users, their journeys, and building experiences that incorporate LLM interactions thoughtfully will create meaningful engagement and lasting impact.
