For most of my career, I’ve worked with material that many people would have described as social data. I’ve used that phrase too. Yet even then, I wasn’t convinced it quite described the material that I was analysing.

My very first project explored the experiences of volunteers at mega sporting events, like the Olympics and Commonwealth Games. I couldn’t interview the volunteers directly, so I turned to the digital traces they had already left behind online. At the time, those traces came largely from blogs, forums, and online communities. Sure, Facebook and Twitter existed, but they weren’t yet the centre of the internet. Looking back, what I now see is that while I labelled this as social data, I was using the material as evidence.

The internet changed, but our language didn't

Today, the online material available to researchers is far more varied. We analyse search behaviour, reviews, creator content, behavioural traces, online communities, AI-generated summaries and even synthetic datasets. Increasingly, we’re expected to understand not only what people have created, but also what machines have generated on their behalf.

None of this changes the principles of good research. Researchers have always evaluated evidence, considered context, and recognised that different sources have different strengths and limitations. What has changed is the range and nature of the digital material available to us, while much of our language comes from a simpler internet.

Different questions require different evidence

For years, one of the biggest debates around digital research has been representativeness. Is social media representative? Can we generalise from people participating in a particular online community? These are important questions, particularly when making claims about a wider population, but representativeness isn’t the only way evidence can be useful.

Researchers have never expected every form of evidence to answer every question. An interview tells us something different from an observation; an experiment tells us something different from ethnography. We don’t question whether one is better than the other, but whether it is appropriate evidence for the question we’re trying to answer and the claim we’re trying to make.

The same principle applies online. Search behaviour might provide evidence of information seeking, a product review might provide evidence of a reported experience or evaluation, an AI-generated summary may provide evidence of how available information has been retrieved and synthesised. None give us transparent access to what someone thinks, means, or does simply by existing. What they allow us to understand depends on how the material was produced, the question we’re asking, and the claim we’re trying to support.

Looking beyond the source

We tend to organise digital material according to where it came from – TikTok, Reddit, Google, Trustpilot, LinkedIn. While platform design can influence who participates, what becomes visible, and how people behave, the source alone doesn’t necessarily tell us what the material can provide evidence of.

An Instagram post might be personal expression, paid influence or organisational communication. A Reddit thread might contain lived experience, collaborative problem-solving, or customer support. Two pieces of material from the same platform can therefore have very different evidential properties. An AI summary is different again; it isn’t simply another source of evidence, it’s a synthesis of other material, influenced by what the system retrieves, selects, and presents. Depending on how it is produced, it may also contain a layer of interpretation.

I’ve started to think we need to consider both the conditions under which the material was produced and what it might provide evidence of.

Beyond terminology

This is why I don’t think this is simply a question of replacing one label with another. The issue isn’t really the term ‘social data’ – it’s that the term can encourage us to treat very different forms of online material as though they belong to a single category.

That becomes harder as researchers work with search behaviour, AI-generated summaries, and synthetic data alongside material created directly by people. The range of material we’re being asked to interpret has expanded beyond many of the assumptions that accompany the term social data. For me, digital evidence is an attempt to recognise that change. Whether the terminology sticks is less important than whether it helps us think more carefully about the material we’re working with.

Dr Jillian Ney is a digital anthropologist and founder of The Social Intelligence Lab