📎 How to Use AI in Brand | Part 1: Using LLMs to Understand Your Audience

Stop guessing what your customers want, use AI to find out

Good morning, readers. Today I’m launching a multi-part series I’ve been excited about for a while: how we should be using AI to support brand and marketing work. Most of the resources out there over-index on how to create content faster and cheaper—the result being a lot of generic dreck. 🚮

I think one of AI’s best use cases is helping us better understand the people we’re marketing to and pushing our creative thinking further…not being creative for us. How do we do that? Let’s start with audience research because everything else—your positioning, your messaging, your creative—falls apart without that foundation.

Perfect timing, because Typeform just launched the next installment of their Get Real series, and we’re teaming up to ask the exact questions I’ve been wrestling with. They’re digging into how generative AI is reshaping our industry—the stigma around using it, the fear of being replaced by it, and the pressure to prove we’re still human. They’re capturing actual marketer perspectives through writing, video, and audio responses to get the nuance behind the data, not just numbers. You’ll see yours truly 👋🏼 in the survey asking (and answering) the exact question I was getting at above—how do we make sure our use of AI actually makes our marketing better for our audience, not just easier for us as marketers?

👉🏼 Share your honest take by September 26—I predict a spirited final report.

If someone sent you this post and you’re not subscribed, join those people learning how to tactically advocate for brand at your company. 📬

---

🧷 A disclaimer before we begin: I’m not one of those people who is obsessed with prompting optimization, and I’m still (always) learning. These aren’t intricate or complex prompts, and I guarantee each one I share below can be improved and refined, let alone customized to your use case—so take them as a starting point, see what works for you, and tell me what you learn as you go!

---

If you’re using AI to help produce any type of customer-facing creative output, there’s critical context you need to feed your tool. And whether you’re a marketer, a founder, a CEO, a sales manager, whoever—in order to create effective marketing, you need to understand your audience deeply, from demographics to firmographics to psychographics. It’s the responsibility of any marketer, but especially of the brand leader, to get past “our customers have these titles and work at companies with $X in revenue and have these functional problems” to understanding the underlying motivations, emotional drivers, and cultural context surrounding the people with whom we aim to connect.

My goal in this post is to share how I’ve used AI to build (and maintain) a tidy foundation of context to quickly ramp the LLM assistants you (should) set up for each skill area or workstream. Using these methods will help you get to strategic plans, campaign budgets, and asset production readiness faster—and makes it less likely you’ll wind up with the same garbo as anyone else with a free ChatGPT account. 😛

This post is focused on foundational context inputs. The following posts in this series will build on this foundation, building up to a strategy for using AI systems empowered with your unique context to build a robust, measurably successful brand strategy (stay tuned for a Part 2).

AI audience research tools

Both Claude (Anthropic) and ChatGPT (OpenAI) are great for analyzing qualitative data like interview transcripts, support tickets, and survey responses, or for processing structured data and creating research frameworks. Perplexity is where I tend to go first for competitive research, advanced sourcing, and trend analysis (but all the major LLMs are getting better at citing sources). There are tons more tools out there, and I’m exploring them in real time with you.

The benefit of these tools compounds when you layer context over time within the same thread, like with Claude Projects (my personal preference) or Cursor Agents. The first step is to set up a project environment with a purposeful goal, like for customer interview analysis and persona development.

Technique 1: Mine your existing customer data

Homework: Gather/export any and all data you have related to your goal—sales call transcripts, support tickets, customer survey responses, review site comments, social media mentions. CSV files are perfect (and you can have your LLM clean those up before analysis, if needed). This works because you’re not asking AI to guess what customers want—you’re asking it to organize and analyze what they’ve already told you.

The prompt:

I'm sharing [X, Y, Z types of data] from our customers. Please analyze this content and identify:

1. The top 5 pain points mentioned most frequently
2. The language customers use to describe these problems (exact phrases, note any interesting nuances)
3. The outcomes and/or ulterior benefits they're seeking
4. Any differences in priorities between different customer segments
5. Emotional undertones (frustration, excitement, confusion, etc.)

Format your response as a table with direct quotes as evidence for each insight.

📋 Pro tip: Analyze both positive and negative feedback. Pay attention to complaints and support tickets where customers got frustrated—this usually shows you an opportunity via a gap between what they expected and what they experienced, which often points to positioning or messaging issues. I’ve found some of my best brand insights in Zendesk tickets where customers said things like “I thought this would...” or “Why doesn’t it just...”

Technique 2: Competitive message analysis

Homework:

Collect homepage copy, ad creative, and marketing emails from your top 5-10 competitors. If you’re using an LLM with live search on, you can ask it to pull these for you, but it’ll be more limited to what it can find publicly online (like web pages, not an email series). IMO, it’s worth it to spend the extra time to gather this yourself—I keep a big Figma file of my “competitive landscape” on hand with all kinds of screenshots of various competitors’ marketing, and I add to it all the time.

Screenshot of one of the pulls I made at Later, compiling influencer marketing homepages

The prompt:

Review the attached marketing messages from these companies in or adjacent to the [industry] space. Analyze and identify:

1. What themes/benefits do most competitors emphasize?
2. What language patterns do they use repeatedly?
3. What pain points or customer needs are they NOT addressing? Order these by impact and importance.
4. What positioning territories seem unclaimed or underexplored?
5. How could we differentiate our messaging in this landscape? 

Create a competitive messaging map showing where competitors cluster and where white space exists.

Include notation around where you see contradicting information, are making assumptions, or may need more information to fully validate your analysis. 

📋 Pro tip: The output for the “white space” opportunities (bullet #3 above) will benefit the most from any context you gather on customer needs and opportunities, so you don’t just end up with a list of every possible applicable feature. Secondly, you don’t have to limit yourself to direct competitors in this exercise. Include companies your prospects might consider as alternatives, even if they’re in adjacent categories—whatever is ultimately competing for your audience’s time and attention.

As a follow-on prompt, you can also enter this snippet again so the LLM creates a visual map separate from the text artifact (and here’s a super rough example of the potential output):

Create a competitive messaging map showing where competitors cluster and where white space exists.

Technique 3: Social listening at scale

Homework: Gather Reddit threads, LinkedIn posts, Twitter/X conversations, industry forums (consumer Facebook groups, startup forums like Bookface or Hacker News, etc.), and review sites where your audience talks about their challenges. Perplexity, for example, can also help pull these for you, but do your own search and see how it differs from what the LLM finds.

The prompt:

I'm researching how [target audience] talks about [problem/category] on social media. Analyze these posts and comments to identify:

1. How they describe the problem in their own words
2. What solutions they've tried and why they failed
3. What they wish existed but doesn't
4. The emotional context around this problem
5. Who or what they blame for the problem
6. Their success criteria for solutions

Focus on authentic, unfiltered language they use when they think no brands are listening. 

Include notation around where you see contradicting information, are making assumptions, or may need more information to fully validate your analysis.

📋 Pro tip: Look for conversations where people are asking for recommendations or complaining about existing solutions, especially in industry-specific subreddits, LinkedIn groups, and Slack communities. You’re looking for phrases like “Does anyone know of a tool that...” or “Am I the only one who hates how [category] always...,” which are good indicators of unmet needs—and especially language patterns that haven’t been sanitized through product marketing.

Technique 4: Interview transcript goldmining

The homework: Customer interview transcript analysis has been adopted fairly fast as an AI use case (though the number of interviews are lower than you’d hope). My teams have benefited a lot from aggregating customer and prospect call transcripts from across the company—other teams are happy to share, and recordings are often available for download from a CRM. I take all of that, add it in a chat within my “transcript analyzer” project, and add our brand guidelines, content system, corporate vision documents—whatever will give the bot background with which it frames responses. When we’ve done this exercise, here’s one of the prompts I thought returned the most helpful output:

The prompt:

Analyze these customer conversation transcripts and identify:

1. Unprompted problems they mentioned (vs. direct responses to questions)
2. Jobs-to-be-done: what are they actually trying to accomplish?
3. Emotional moments: where did they seem frustrated, excited, or relieved, for example?
4. Success metrics: how do they measure if something is working?
5. Decision criteria: what factors influence their buying decisions?
6. Language goldmines: exact phrases they use that could inform our messaging

Flag moments where customer participants contradicted themselves or where their stated needs differed from their implied needs. 

Include notation around where you see contradicting information, are making assumptions, or may need more information to fully validate your identified insights.

📋 Pro tip: Were there any moments when your interviewer had to clarify or rephrase a question? These usually happen when there’s a disconnect between how you think about the problem and how customers think about it. Also look for times when customers used analogies or metaphors to explain their situation (“It’s like herding cats” or “It feels like throwing spaghetti at the wall”), which gives you emotional context that you can build brand insights around.

Technique 5: Persona development and validation

The homework: Once you’ve gathered insights from your existing research and the results of the prompts above, use AI to help synthesize them into actionable personas. What you end up with here will just be a starting point—you’ll still want to think critically about organizing personas in a way that will be most useful to your team and business. I’ll share my Notion template on persona-building in the coming weeks—subscribe to stay tuned.

The prompt:

Using the context provided in the customer documentation I've shared, derive and generate 2-5 distinct "marketing audience persona" profiles that include:

1. Demographics and firmographics
2. Primary goals and motivations
3. Biggest challenges and pain points
4. Preferred communication styles and channels
5. Decision-making process and criteria
6. Success metrics and desired outcomes
7. Exact language they use to describe problems and solutions

For each persona, also identify:
- What messaging would resonate most
- What messaging to avoid
- Content topics they'd find valuable
- Potential objections to our product/service

Format as detailed persona cards with direct quotes from the research as supporting evidence. Do not invent direct quotes. 

Include notation around where you see contradicting information, are making assumptions, or may need more information to fully validate your recommended personas.

📋 Pro tip: Once you have AI-generated drafted personas, validate them by asking:

• Do these personas align with our highest-value customers?

• Can our sales team recognize these personas in their prospect conversations?

• Do these personas have distinct enough needs to warrant different messaging (and if so, in what channels)?

Putting it all together: The audience insight brief

If you’re taking the time to try out each one of these prompts, you’ll end up with a LOT of information—and the goal is to use that to aid your own learning process and understanding, but it may not be something you can expect stakeholders to digest or recall. So you might use this framework to organize everything you learn:

Audience Insight Brief Template

Using the context provided in the customer documentation I've shared, create an audience insight brief that follows the template below. Include notation around where you see contradicting information, are making assumptions, or may need more information to fully validate parts of the brief.

Template:

"Primary Audience: [Title/Role at Company Size]

Core Problem: [In their exact words]
- Supporting evidence: [Direct quotes]
- Emotional undertone: [Frustration/Fear/Excitement]

Secondary Problems:
1. [Problem + evidence]
2. [Problem + evidence]
3. [Problem + evidence]

Success Vision: [What they want to achieve]
- How they measure success: [Metrics they mentioned]
- Timeframe expectations: [When they need results]

Language Patterns:
- Words/phrases they use: [List]
- Words/phrases they avoid: [List]
- Communication style: [Formal/Casual/Technical]

Decision Process:
- Who else is involved: [Other stakeholders]
- Main objections/concerns: [What holds them back]
- Preferred information sources: [Where they research]

Competitive Context:
- Current solutions they use: [Alternatives]
- What they hate about current solutions: [Frustrations]
- What would make them switch: [Triggers]"

📋 Pro tip: This step can be like boiling the ocean depending on the amount of upstream data input, the level of revisions and polish you applied to smaller unit outputs, and general LLM bad habits. It’s not going to be perfect, but it will create discussion that can accelerate clarity and earn investment in a brand led vision.

Next up: Building brand foundations with AI

Once you have this rich foundation for audience understanding, Part 2 will walk through how you can use AI to help develop positioning, messaging, and brand strategy that maps to what you’ve learned about the people you’re building for. You’ll notice that the outputs from the prompts above will still feel pretty tactical and functional—not yet tapping into emotional needs and creative opportunities. So we’ll cover:

• Using AI to identify positioning white space and messaging hierarchy

• Generating messaging frameworks that speak your audience’s language

• Testing brand concepts before you build them

• Creating a brand strategy that differentiates based on real insights

Each step builds on the previous one. You can’t create compelling brand foundations without first understanding your audience. And you can’t create effective brand assets without solid foundations.

Coming next week: How to Use AI in Brand | Part 2: Building Brand Foundations

---

Made it this far? Here’s one more reminder to weigh in with your thoughts on how you see generative AI reshaping the marketing landscape. I can’t wait to hear what you think. 👇🏼

---

Thanks for reading.

Found an AI use case for better audience understanding? I’d love to hear about it! Consider:

• connecting with me on LinkedIn: Kira Klaas

• sharing this with someone who needs better AI prompts

• sending to a friend 💌 or coworker 💬