Three Ways to Measure How AI Is Shaping Your Brand’s Perception

By Dig Insights

  • article
  • Agile Quantitative Research
  • Agile Qualitative Research
  • Behavioural Insights
  • Brand Research
  • Concept Testing
  • Diary Studies
  • Innovation Research
  • Trend Monitoring
  • Trend Scouting
  • Questionnaire Design
  • AI
  • Artificial Intelligence
  • Brand Tracking

Summarise with AI

ChatGPT
Claude Logo
Gemini Logo
Perplexity Logo

At Dig Insights, once we named System Zero, the layer of consumer decision-making that people now hand over to AI, the obvious next question was how to measure it. Our answer is that there are now two markets to consider, not one. There’s the traditional consumer market, the one we’ve always researched. And there’s an AI-mediated market that sits alongside it. We’re not dropping the human from the picture, but we do need to evaluate these two markets differently because they operate under different machinery.

In practice, this means that when we talk to the human layer, we’re no longer just asking people their opinions, what they consider and how they consider it. We’re also asking how AI plays into that choice. We explore AI’s role in discovery, consideration and recommendation, and run funnel analysis and equity mapping across both consumer surveys and AI-mediated environments. 

We evaluate human decisions and considerations as we typically would, identify the moments when delegation to AI happens and then query AI directly to capture how that process influences the outcome. Where before we started and stopped with consumers, we’re now capturing the AI world as well.

To capture both worlds, we’ve developed three distinct methods, and most of our studies now draw on at least one of them.

Asking Consumers About Their AI Use 

The first method of measuring System Zero is overt. This is simply asking consumers directly about their AI use, to understand their conscious behaviour: which tools they’re using, and how much they trust the information they receive. It’s about understanding their relationship with those tools and with the recommendations those tools give, and translating that attitude into behaviour. 

We include this in most of our studies now, because it’s critical to understand what proportion of a journey is being delegated to AI. The results aren’t necessarily consistent between segments, so it requires different levels of observation depending on the audience.

Using AI Within the Research Experience

The second System Zero measurement method is contextual. Here we use AI within the research experience itself to help reveal latent preferences through qualitative research or other observational means. This is less about studying AI usage directly and more about using AI to help consumers articulate their needs, priorities and trade-offs. It gives us a snapshot of someone’s decision-making process as it happens with AI. 

For example, we might look at whether someone clicks through thirty questions, provides just one prompt or forty or contests the results the AI gives them. We might look at whether they check other websites first, after or at all, and how that relationship with the tool is genuinely forming.

Querying the AI Systems Directly

The third method of measuring System Zero is querying the AI systems themselves. Here we ask large language models thousands of consumer-like questions, built from distinct consumer profiles, making sure those profiles aren’t communicating with each other, and we quantify how brands are represented and associated as a result. 

This third method has two basic approaches. The first is to ask the models a question and measure what comes back: we generate hundreds or thousands of prompts that mimic what a real consumer might ask. Dig Insights has been doing brand funnel work for fifteen years, so we know that awareness, consideration and recommendation sit at different points in a funnel and are likely to be engaged with differently by a model. We build that understanding into how we construct the questions. 

The second approach is forced choice, used to assess preference within the models. Direct questions to a language model often produce generic, overly positive answers because these models tend toward the diplomatic. Forced choice requires the model to reveal which brand it associates more strongly with a specific attribute, which gives us a much sharper read.

A Facial Moisturiser Case Study

We put all three methods to work in a case study on facial moisturisers in the UK and France. On the consumer side, we fielded a traditional survey through our DigOne platform, measuring the familiar funnel of awareness, consideration and recommendation, and running a brand equity mapping exercise where people tied key brand attributes to brands in a randomised carousel grid. 

On the AI side, we spoke to the equivalent of 500 consumers, weighting the sample by relative usage in each market. For the funnel analysis, we used realistic, consumer-style category prompts and coded which brands were surfaced, considered and recommended, asking thousands of unique prompts in total. For the equity analysis, we used forced-choice comparisons between brands on specific attributes, such as whether a brand uses high-quality ingredients.

Where Consumers and AI Diverge

Running both halves of this approach side by side lets us compare the two worlds directly and assess the gap between what consumers believe and what AI is actually presenting back to them. In our case study, that comparison showed real differences. 

Some brands, such as CeraVe and The Ordinary, were amplified in the AI-mediated results relative to how consumers themselves ranked them. Others, such as Nivea, were underrepresented in both markets we tested. 

A brand’s performance on a specific attribute with consumers didn’t always carry over into the AI-mediated results either, which is exactly the kind of gap this two-part approach is designed to surface.

Keeping Pace With Changing Models

One thing worth being honest about is that the models themselves are always changing. A new model comes out, and the tools and question sets we use don’t necessarily apply in quite the same way as before. 

That’s why we build in continuous monitoring rather than treating this as a one-off study; regular rechecking and using consistent profiles and question sets, so that we understand how the models are currently operating to the best of our ability and can query accordingly. 

We also don’t need to remeasure the whole research programme every time, since the consumer side doesn’t shift at the same pace. Often it’s enough to remeasure the AI environment specifically.


From Measurement to Action

The point of measuring all of this is ultimately to act on it. We think of this as a simple cycle: measure and diagnose how AI currently represents your brand, prioritise the attributes you can credibly own, activate against that positioning and then remeasure the AI environment to confirm that the activation has actually worked. That last step matters. It’s not enough to assume a messaging change has shifted how a model talks about your brand. You have to go back and check.


Author

Learn more about

Scroll to Top