LLMs, machine visibility, and the future of social data

Social data has long faced a credibility question: do the conversations we analyse meaningfully represent wider populations, or only those motivated to speak?

That question is now resurfacing in a new form. As large language models increasingly surface, summarise, and repurpose online content, the issue is no longer just who is speaking, but how and for whom people are sharing content online and expressing themselves. People are beginning to anticipate machine audiences, adapting how they write in public spaces in ways that prioritize clarity, structure, and perceived usefulness, and businesses are trying to get ahead of the curve and optimise for LLM exposure across social media platforms. 

For those of us who analyse social and internet data, this sift demands a reassessment of what our data actually represents and how confidently we can interpret it. For instance, Reddit has been long valued for depth, nuance, and people thinking out loud, and is now showing subtle changes in what is shared and how contribution is shaped by the expectation of downstream machine reuse. These changes are not driven by platform design alone, but by a broader optimisation towards machine visibility that echoes the early days of search engine optimisation.

This session explores how LLM-driven optimisation is reshaping social data, what this means for data quality and interpretation, and why social intelligence teams need to calibrate how confidently they read and represent online conversations.

This is a discussion about behaviour, incentives, and what social data now represents, and the responsibility that comes with analysing it. You’ll learn:

• How GEO is changing what people and brands share online, and what this means to data quality 
• Why Reddit is an early signal of a wider behavioural shift
• How to spot machine-orientated expressions in your datasets
• What this means for representativeness and confidence