A Whole New World? Key Learnings from ‘Gen AI in Social Listening’
23 May 2024

Author: Deanna Tserkezie, Director - Research Innovation, Pulsar 

The MRS Advanced Insights and Analytics Council teamed up with Dr Jillian Ney at The Social Intelligence Lab to discuss the application of Generative AI (Gen AI), and AI more broadly, with social listening data.

AI is no short-term trend, and social listeners have used it for some time. With the growth in familiarity and accessibility of Gen AI, organisations recognize the opportunity that these tools pose for improving workflows. Supporting this industry-wide development, the event centred on exploring the benefits and limitations of social listening.

While it's still fresh, we wanted to share what we learned from the day.

Levelling up Social Listening with Gen AI simulations

Richard Preedy and Kelly McKnight, Executive Directors at Verve, started the event by re-introducing Character AI, a chatbot service that generates human-like text responses, letting you interact with fictional, historical, and celebrity figures. Seeing the opportunity for their clients to improve accessibility of insights and bring to life social audience personas, Verve built their own character AI tool using a research-led framework. Presenting three case studies alongside their cumulative learnings, they shared that validating results is challenging. However, curating large datasets with the right data and personalising models improved results and got closer to real-life outcomes.

1. New ways to access data

Like the Character AI example, the team made their own Chef AI, an easy tool to inspire chefs. It helps them create new recipes using a database of ingredients from recipes, menus, and dishes found in social data. Some ideas were strange, but the tool was an innovative deliverable that made exploring the huge dataset accessible for inspiration, idea generation, and hypothesis testing.

2. Enriching personas

In another example, Verve used AI to group social conversations into themes. Where one theme sparked further interest, they further clustered the conversation by the personal needs they represented to discover the personas present. Assigning a name, image, and the key needs to each persona based on the data, they then used Gen AI to extrapolate beyond the dataset to establish other factors, such as their lifestyle, where they live and what else they like. Developing the personas by inferring these new variables sped up the insight process to craft stories and narratives that landed.

3. Speed and depth of insights

Verve's final example aimed to tackle the question: What are the underlying personas for Gen Z that show in the many styles, aesthetics, and cores? Verve collected data from trendsetting accounts and hashtags on Pinterest, TikTok, and Instagram, then clustered the conversation into themes and personas. They created a Character AI for each persona; for instance, the "Quaint Cuteness" persona includes women who embrace nostalgic trends like balletcore and cottagecore. The AI personas could then provide valuable understanding without hesitation for round-the-clock queries, even when faced with challenging or confidential questions, enhancing the effectiveness of social listening for quick answers.

Training Cultural Insights into Image AI

Shifting from Character AI to AI-based image analysis, Crowd DNA's Benjamin Long raised similar benefits and challenges.

Social listening approaches require AI to find elements within images for analysis at scale. However, image AI often only captures some of the meaning of the elements in context, requiring a researcher to move from facial expressions to emotion and tags to cultural meaning. In two examples shared, Long demonstrated how Crowd DNA uses AI-powered semiotics paired with human interpretation to size and track clusters of image-based tags.

For a cafe brand wanting to understand how to drive greater brand loyalty, this workflow involved collecting social images from cafes, coffee brands, grocers, and other loyalty-led industries and running it through image analysis to identify customer angle and tone of voice. Cultural strategists and semioticians then decoded these findings to reveal a shift in loyalty dynamics. They discovered that loyalty was evolving from mere point collection to fostering more engaging, inclusive interactions between brands and customers. This approach not only tracks emerging trends but also identifies new opportunities, providing a deeper understanding of consumer behaviour.

Training cultural insights into algorithms was raised as a critical challenge that takes time and isn't always perfect, underscoring the ongoing need for human expertise alongside AI in social listening.

Key Takeaways

1. Enhanced Efficiency to Complement, not Replace

AI speeds up data processing and analysis, which helps researchers handle vast social media data and find patterns and trends at scale. But it can't grasp human emotions and cultures well, limiting its effectiveness. While AI is a valuable tool, it can't replace the depth of insight and contextual relevance human researchers provide.

2. Dependence on High-Quality Data

The performance of AI in social listening relies on the quality and size of the dataset it's trained and applied to. Large, well-curated datasets are still needed to unlock AI's full potential, as their absence can result in incomplete or misleading insights.

3. Future Skill Sets

Increasing opportunities for integrating AI in social listening underscores the need for researchers to have skills in best practices to gain meaningful outputs. As with any data, understanding how AI models work is crucial for communicating their limitations. While bias in AI and datasets is a discussed issue, market researchers, with their expertise in identifying and mitigating bias, become even more critical in ensuring accurate and fair insights.

To ensure you are the first to know about future Data Anayltics events like this, please visit your MyMRS online account and update your email preferences under the Settings tab.