Summary
Every strong brand strategy rests on brand research: understanding customers, the market and the gap between how a company sees itself and how others see it. AI tools now make it practical to process volumes of material that used to be out of reach for mid-sized teams: reviews, interviews, competitor sites, open-ended survey answers. They also invent, generalize and miss nuance. This article shows how to run brand research that combines human conversations with AI-assisted analysis, and how to turn findings into insights you can build a brand on.
A marketing manager at a mid-sized company is asked to rebuild the strategic foundation of the brand. She has three customer surveys from recent years, hundreds of online reviews, transcripts of sales calls and a clear sense that the current message isn't landing. What she doesn't have is time to read all of it, or a method for turning the pile into direction.
That's the starting point for brand research. At Karnaf Studio, active since 2003, research has always been the foundation of our branding work. What has changed recently is how much material we can process, and how quickly we can move from raw data to the right questions.
Brand research vs. market research
Market research asks how many customers there are, how much they spend and who the competitors are. Brand research asks different questions:
- How are we perceived today? Not how we want to be seen, but how people actually describe us.
- What are customers really buying? A company may believe it sells on price while customers talk mostly about responsiveness or personal attention.
- Where is the white space? Which message, tone or stance is still unclaimed in the category.
- What does the team believe? A brand employees don't identify with won't last.
The output isn't a thick report. It's a handful of sharp insights that drive positioning, brand promise, tone of voice and, eventually, the visual identity.
The research process: what's human and what's AI
1. Define the research questions (human)
"What do customers think of us?" is too broad. "Why did customers who compared us with a competitor choose us, and why did others walk away?" is answerable. This step requires knowing the business and its goals. There's no shortcut.
2. In-depth interviews (human, with AI transcription)
An open conversation with a customer, employee or partner reveals what no survey will: a hesitation, a recurring word, a story about the moment the company surprised them. A person runs the interview, someone who knows when to follow up and when to stay quiet. AI comes in afterward for transcription, first-pass summaries and quote tagging.
Important: before uploading transcripts to any AI tool, get participants' consent, remove identifying details and check the tool's data-retention policy.
3. Processing existing material (AI as accelerator)
Reviews, open-ended survey responses, support tickets and transcribed sales calls are a goldmine most companies ignore because nobody has time to read them all. A language model can cluster thousands of sentences into themes, surface recurring vocabulary and compare customer segments.
4. Competitor mapping (AI collects, humans interpret)
Gathering competitors' messages, taglines and visual language is technical work AI handles well. Interpreting it, what competitors actually promise and where those promises are weak, needs a trained eye.
5. Synthesis (human, with AI as a sparring partner)
This is the most important step: turning findings into insights. A finding is "customers mention availability." An insight is "customers don't buy our expertise, they buy peace of mind." A model can propose phrasings and challenge assumptions, but people who understand the context write the insight.
6. Personas and insight map (shared)
From the research you build personas: key customer groups, their motivations, concerns and language. AI helps draft and organize, but every persona must be grounded in research data, not imagination.
The trap: synthetic personas and invented data
It's tempting to ask a language model to "imagine our customers" and answer on their behalf. The result sounds convincing, but it isn't research. A model returns what is statistically plausible, not what your specific customers think.
NIST's Generative AI Profile, a companion to its AI Risk Management Framework, names this risk "confabulation": the confident generation of false or misleading content. In brand research it's especially dangerous, because an entire strategy can end up resting on a quote nobody said or a figure that doesn't exist. That's why we check every AI-surfaced finding against its source: does the quote exist in the transcript, does the review really say that, does the number come from real data?
An example: from finding to insight
Imagine a company that installs commercial air-conditioning systems. Leadership is convinced its edge is price. AI analysis of a few hundred reviews shows "price" barely appears, while "showed up on time" and "explained what they did" keep recurring. Interviews reveal that office managers choose the company because they never have to chase it. The insight: the company sells certainty, not price. That shapes the message, the tone and even calmer, more orderly design choices. (This is an illustrative scenario, not a description of a specific client project.)
Common mistakes
- Interviewing only happy customers. Lost customers often teach you more.
- Asking leading questions. "What did you love about our service?" assumes they did.
- Confusing volume with importance. A frequent topic in reviews isn't necessarily what drives choice.
- Trusting summaries without reading sources. An AI summary is an entry point, not a substitute.
Pros and limitations
| Pros | Limitations |
|---|---|
| Processing volumes of material that used to be impractical | Models can invent quotes and data |
| Fast detection of recurring themes and language | Missed nuance, irony and cultural context |
| Comparison across customer segments | "Synthetic personas" that don't reflect real customers |
| More time for interviews and synthesis | Privacy concerns when uploading transcripts |
FAQ
How many interviews do we need? There's no magic number. Keep going until interviews stop adding new themes.
Can a survey replace interviews? Surveys are good for measuring; interviews are good for understanding. Brand research needs both, especially the latter.
Can we build personas with AI alone? You can draft them with AI, but they must rest on real data. A persona created without research is a hypothesis, not a tool.
In summary
Good brand research combines two strengths: human listening that catches what goes unsaid, and AI-assisted analysis that lets you read far more than any person could. Combine them well, verify every finding against its source, and you get a strategic foundation that can carry a brand for years. At Karnaf Studio, every branding project starts with research, and only then moves to design.
Want a brand built on a real understanding of your customers? See our branding portfolio or talk to us.
Sources
- NIST, Artificial Intelligence Risk Management Framework: Generative Artificial Intelligence Profile (NIST AI 600-1), July 2024 – https://www.nist.gov/publications/artificial-intelligence-risk-management-framework-generative-artificial-intelligence
Let’s talk about your brand
Tell us about your project and we’ll get back to you.
Contact Karnaf Studio
