Conversational Search

In developing a new conversational tool of our own, we embarked on a discovery phase that involved research into AI-powered conversational design and an analysis of the leading players in the industry. Through this process, we gained valuable insights into the best practices, trends, and user expectations surrounding conversational AI tools. Here are some of the key insights we discovered during the process.

When designing a conversational search platform, prioritising user efficiency is essential. This means addressing both sides of the UX/UI coin, ensuring that the platform’s interface and user experience are both easy to navigate, provide quick and concise results, and focus on supporting a smooth conversational flow.

When designing an efficient conversation interface, consider these best practices: keeping the UI simple and focused on the query’s input and response, summarising results concisely while highlighting only key information, and grouping suggestions into logical categories or lists whenever possible to make the information more digestible for users.

Perplexity exemplifies this approach with its minimalistic UI, prompt suggestions, and immediate access. By removing the need for users to log-in or sign up before searching, Perplexity offers a barrier-free experience that is both efficient and appealing for users.

UI elements should elevate both the experience and presentation of the conversational flow. A clean and intuitive landing page is crucial for converting passive viewers into engaged users. The hero section of the page should be simplistic and feature a prominent input field for the search query, a concise and attention-grabbing tagline, clear branding, and a primary call-to-action.

By prioritising these key elements, you can create an inviting entry point that sets the stage for a smooth transition from landing page into conversational search. Phind, an intelligent answer engine for developers, demonstrates this well.

A well-designed conversational search should replicate natural dialogue to create a human-like experience. The natural language processing (NLP) should be designed to understand and respond to queries in a way that mirrors human conversation. This includes understanding context around queries, avoiding unnecessary jargon, and acknowledging errors with empathy.

Chat GPT is a good example of this with its advanced NLP capabilities that enable it to engage in natural-sounding conversations and to handle complex queries. This level of human-like engagement fosters a sense of trust and reliability between user and platform.

Media richness refers to a medium’s ability to effectively communicate information. It’s evaluated based on key criteria; the capacity to include personal focus, the immediacy of feedback, the conveyance of multiple cues, and the variety of language used.

In conversational search, a media-rich design may include elements like images, videos, audio, links and maps, creating a more immersive and interactive experience. Whether this approach enhances the user experience compared to a leaner design depends on the platform’s goals and target audience. For instance, platforms like Chat GPT and Mistral adopt a lean, minimalistic approach, while others, like Copilot, opt for a more media-rich experience.

Conversational search is an iterative process where users engage in a back-and-forth dialogue to refine their results and help the system better understand their needs and preferences. An effective iterative process involves providing clear feedback, rephrasing or expanding on user prompts, and knowing when to stop iterating. This approach ensures an efficient and effective user experience.

The process is cyclical, comprising three stages; generating a response, evaluating and identifying areas for improvement, and refining the prompt. For example, Perplexity’s Related section demonstrates this well by offering contextually relevant follow-up questions that encourage further topic exploration.

AI-powered platforms can dynamically personalise search results based on a user’s behaviour, interests, search history and patterns. This type of granular personalisation enables AI to tailor the search experience to each individual, making it both seamless and personalised.

Komo integrates personalisation directly into its search landing page, offering users four search options: Ask, Search, Research, and Explore. Ask is an AI-driven search, Search functions as a standard search engine, Research looks through research resources, and Explore provides mixed media results (videos, images, articles). By choosing one of these, users customise their search experience right from the start.

Conversational search platforms should prioritise accessibility in their design. Key features to incorporate include keyboard navigation, straightforward designs that are compatible with screen readers, text-to-speech functionality, multi-language support, and customisable interfaces for adjusting fonts, colours, and contrast.

A good example of this is Claude.ai’s Chat Controls which allows users to choose their preferred chat model and font styles, including a ‘Dyslexic-friendly’ option. This feature not only improves accessibility but it is also available in the free version, Claude Sonnet.

Conversational history allows a large language model (LLM) to retain context throughout an ongoing conversation and to understand follow-up questions. It consists of two parts; memories and messages. Memories are the stored messages from previous interactions, while a message represents the question-and-answer pairings within a conversation.

Easy access to stored conversations improves the user experience by letting users quickly revisit past queries, ensuring continuity in interactions, and adding an extra layer of personalisation.

By integrating these insights and UX/UI best practices, an AI-driven conversational platform can deliver an effective, engaging and intuitive user experience. Prioritising efficiency, simplicity, and accessibility ensures that users can navigate the platform with ease and quickly find the information they need.