Choosing the right evidence for brand research
Brand research helps businesses understand how people perceive their brand, what they associate it with, how it compares with competitors and what influences people to choose it.
It can measure brand awareness and consideration, uncover perceptions and associations, identify gaps between a brand promise and customer experience, and understand why customers stay or switch.
The value of brand research is not simply knowing what people think. It is using that understanding to make better decisions about positioning, messaging, customer experience, product and marketing. The question is how best to get that evidence.
Today, there are three broad approaches: analysing existing evidence, using synthetic research to explore audience responses, and recruiting real people.
At the LDN AI Lab, we’ve been testing how each of these approaches can be used in real projects, where they add value and where they don’t. What we’re learning is that AI isn’t necessarily changing what we need to research. It is changing how we can get there.
Using existing data for brand research
A surprising amount can be learned without conducting a new study. Industry research, syndicated data, customer reviews, search behaviour, competitor activity and social conversations can reveal how a brand is performing and how people talk about it.
AI makes it possible to analyse these sources at a much greater scale. We can identify recurring themes, compare competitors and understand the language customers use, while connecting findings back to their original sources. This can provide a useful picture of brand awareness, perception, consideration and competitive positioning.
We’ve been applying this approach in our own brand intelligence work, bringing together multiple sources to build a more complete picture of a market before recommending further research. It is particularly useful at the beginning of a project, when we want to understand what is already known and where the gaps are.
The limitation is that we’re working with questions that have already been asked. Existing data can tell us a lot about what is known, but it cannot answer a question that has never been asked.
What is synthetic research?
Synthetic research uses AI-generated respondents or audiences to explore how people with particular characteristics might respond to a question, proposition or concept.
The evidence is still developing, but early research suggests synthetic respondents can be useful for certain types of attitudinal and conceptual research.
For example, one study used 19,447 synthetic personas to reproduce findings from 45 studies published in the Journal of Marketing, reproducing 76% of the original main effects. Another study by PyMC Labs and Colgate-Palmolive compared synthetic consumers with 9,300 human responses across 57 product surveys and reported 90% of human test–retest reliability.
These results are promising, but they don’t mean synthetic respondents are equivalent to a recruited sample.
What we’re learning through the Lab is that their value is often in exploration. We’ve been using synthetic audiences to compare propositions, pressure-test messaging, explore different audience perspectives and identify hypotheses worth taking into human research.
It gives us a way to explore more possibilities earlier in a project, before committing the time and cost of recruitment.
It is also important to understand what synthetic research isn’t. A modelled audience isn’t a recruited sample, and generating thousands of synthetic responses doesn’t create a statistically representative population. We use synthetic research to explore and inform rather than to manufacture a level of quantitative certainty that isn’t there.
When should you recruit real people?
Some research questions need real participants. If you need to measure brand awareness, track perceptions over time or establish a benchmark for future comparison, you need a defined population and a consistent research methodology.
Real people are also important when the audience is specialist or difficult to model. If we’re trying to understand the experiences of a small group of senior decision makers, for example, qualitative interviews can reveal context and motivations that aren’t available through existing data or synthetic audiences.
This is an important part of what we’re learning. AI doesn’t remove the need for human research. In some cases, it makes the reason for doing it clearer.
Rather than recruiting a large sample to explore every possible question, we can use other evidence to narrow the problem first, then use human research where we need validation, depth or a measurement we can confidently return to.
The more significant the decision, the more important it is to understand the limitations of the evidence behind it.
Can AI and traditional research work together?
Yes, and this is where we see the greatest opportunity. Existing data can help map the market and identify the questions worth exploring. Synthetic research can then help us test propositions and narrow the field. Recruited research can provide robust measurement or deeper understanding where real participants are needed. We’re increasingly applying this combination to projects rather than treating AI research and traditional research as competing approaches.
The benefit is that human recruitment can be focused on the questions where it adds the most value. It can also make research more useful. Instead of spending a large study exploring every possible direction, we can do more of the exploration before people are recruited.
The result can be a more focused research brief, a better questionnaire and a clearer understanding of what we actually need to learn from participants.
What we’re learning at the LDN AI Lab
The biggest lesson so far is that AI isn’t a research methodology in itself. It is another set of tools we can use to make research more efficient, expansive and useful.
We’re finding that AI is particularly valuable for the work that happens around traditional research: understanding what is already known, exploring possibilities, analysing large volumes of qualitative data and helping us decide where human research will have the greatest value.
We’re also learning where we need to be more cautious. Synthetic audiences can produce convincing answers, but convincing isn’t the same as representative. The more a research project depends on precise measurement, a specialist audience or a benchmark that needs to be tracked over time, the more important human validation becomes.
That’s why we’re applying what we learn directly to client projects rather than treating the Lab as a separate experiment. Every project gives us another opportunity to test where AI genuinely improves the research process and where traditional methods remain the better choice.
How we approach brand research at Studio LDN
We don’t start by deciding that a project needs a research panel or that it should use AI. We start with the business question. What do you need to understand? What decision does the research need to support? How much confidence do you need before acting?
The answer might be existing research and data, synthetic research, qualitative interviews, quantitative research or a combination of methods.
The right approach is the one that produces useful evidence at the right level of confidence.
LDN AI Brand Intelligence
LDN AI Brand Intelligence brings these approaches together to help brands understand their position, their audience and their competitive landscape using evidence that already exists, analysed faster and more consistently with AI.
Because better research isn’t necessarily about asking more people. It’s about asking the right questions, using the right evidence and knowing when you have enough to act.
1. Yeykelis et al. — AI personas and replication
Full title: Using Large Language Models to Create AI Personas for Replication and Prediction of Media Effects: An Empirical Test of 133 Published Experimental Research Findings
The study generated 19,447 AI personas across 45 studies contained in 14 papers published in the Journal of Marketing. The AI personas reproduced 76% of the original main effects (84 of 111). When interaction effects were included, the replication rate was 68% (90 of 133).
2. PyMC Labs + Colgate-Palmolive — synthetic consumers
Full title: LLMs Reproduce Human Purchase Intent via Semantic Similarity Elicitation of Likert Ratings
This research tested the Semantic Similarity Rating (SSR) methodology against 57 personal-care product surveys containing 9,300 human responses. The researchers reported that synthetic respondents achieved 90% of human test–retest reliability, with distributional similarity above 0.85.
