How to Teach AI to Understand Your Customers

If only you could know your customers—really, truly know them. You could see through their eyes as they browse, shop, and buy. You could know what they want, why they want it, and how they’re going to get it. Just about every marketer dreams of reading their customers’ minds. The best that marketers have been able to do is ask them what they want, how they feel, and why—until generative AI, that is. In this piece, the first in a four-part series for Every, copywriter Chris Silvestri expands on a framework he built called empathy engineering. It’s for simulating those customers using AI—a cheaper, quicker, and perhaps more effective way to understand them.—Kate Lee

Was this newsletter forwarded to you? Sign up to get it in your inbox.

---

Good communicators know their audience. A presentation to hundreds of conference-goers sounds different than a sales pitch over lunch. It’s why you keep it casual when chatting with a friend and you’re more formal on a call with your boss.

Knowing your audience is also necessary when communicating with large language models. Having that level of insight is what empathy engineering—a framework for using artificial intelligence to understand your customers—can help you achieve. I developed this framework over eight years spent helping more than 50 companies including Moz, Shortstack, and Pantheon convert prospects into customers. AI just made this approach easier and more effective.

But it starts with laying a foundation. Before you even think about prompting AI, you need to think about your customer.

How to get started with customer research

To demonstrate how this works, I’m going to use a made-up company called TeamFlow. TeamFlow is an AI-powered task management platform for remote teams—think Trello with Slack, plus AI. The company built a solid product and has a growing user base, but it wants to improve its marketing and convert more prospects into customers. You can take a look at the full company profile.

TeamFlow’s marketing team has hired us to help the company refine its messaging, uncover hidden opportunities for growth, and connect with its target audience on a deeper level. We’re going to help the company clarify and better communicate what it does, who it does it for, and how uniquely or better it is than anyone else—so that its value and product become the obvious choice for the right people. 

Gathering the context

No matter how powerful AI becomes, it can't replace the value of real-world customer research. AI is a tool, not a magic wand. It can write content, but it lacks context—which is what we have to provide. 

Empathy engineering starts with understanding your customer, your product, and your market. To do this, we'll focus on three pillars of research: internal research (your team), user research (your prospects and customers), and market research (your competitors).

Internal research: Mine your team’s knowledge

Your team is a gold mine of customer insights. They're on the front lines, talking to customers, solving problems, and witnessing firsthand how people use (or struggle to use) your product. Here’s how to extract that precious data:

• Tap into their expertise: Conduct interviews with team members from the product, marketing, sales, and customer support teams. Ask them about common customer questions, pain points , and popular features.
• Analyze support interactions: Review transcripts from support calls or chats. Look for recurring themes, confusing product features, and unmet customer needs.
• Dig into internal data: Explore your customer relationship manager (CRM), website analytics, user behavior patterns, and other internal data sources for insights into how customers are using your product.

User research: Go straight to the source

There's no substitute for talking directly to your customers. User and customer research helps you understand their needs, motivations, and experiences in their own words. There are a few ways to get this data:

• Conduct customer interviews: Ask open-ended questions about their pain points, goals, and how they currently solve their problems.
• Send out surveys: Gather qualitative and quantitative data on customer satisfaction, feature usage, and demographics, and run website surveys to understand their pains, motivations, and desired outcomes.
• Observe user behavior: Conduct testing to see how customers interact with your product and identify areas for improvement, or look at user data like heatmaps to learn how they use your website.

Market research: Know your competition

Understanding the competitive landscape is essential for differentiating your product and crafting effective marketing campaigns. You can learn both what works and what doesn’t by looking at what others in the space are doing. There are a few methods for doing it:

• Analyze competitor websites: Examine their messaging, pricing, features, and target audiences.
• Read industry reports and reviews: Stay up to date on what’s new in the industry.
• Pay attention to customer reviews: See what people are saying about your competitors' products—both positive and negative. This can reveal valuable insights into their strengths and weaknesses.
• Ask your customers: Ask them what alternatives they’ve considered or tried—unless you're in a new or niche market, in which case, prioritize understanding your customers' needs and pain points without introducing the idea of alternatives.

Case study: TeamFlow’s research

Let’s take a look at where TeamFlow is with its research. Here’s what we have to play with (thanks to ChatGPT-4o):

Internal research 

• Company profile: A description of TeamFlow, its product, and its target audience
• Team interviews: Conversations with the founder and key team members in marketing, sales, product, customer support, and customer success

User research 

• Customer interviews: One-on-one conversations with people who fit our target audience
• Website survey: Responses from visitors to the TeamFlow website
• Customer reviews: Feedback from people who are already using TeamFlow

Market research 

• Competitor research: An analysis of TeamFlow’s main competitors, their products, pricing, and marketing strategies

If you don’t know where to start, feel free to use the questions I ask and the structure of the interviews and survey.

Source: ChatGPT-4o.

Extracting the data gold

Sifting through pages of interview transcripts, analyzing survey responses, and scouring competitor websites can feel overwhelming. This is where AI comes in. Think of an LLM as the world's most eager intern—one that never sleeps, never complains about paper cuts, and can read troves of documents in record time.

By having a conversation with it and pointing it in the right direction, AI can sift through all your research, identify key themes, and highlight actionable insights. And it’s utterly at your mercy. It has no ego. If you think its analysis is off, challenge it, provide more context, or ask it to look at the data from a different perspective. It'll readily revise its work, give you new interpretations, and adapt to your guidance.

How to have a conversation with AI

Next, we have to provide the LLM with the context it needs. 

My AI tool of choice lately has been Gemini’s 1.5 Pro model. Google recently expanded the model’s context window to 2 million tokens, which gives users plenty of room to feed it their documents.

To start, I’ll create a new prompt in Gemini and feed it everything I have with some information around my goal and the purpose of this chat session. At this point, I don’t want to give it any particular instruction. I want it to process the materials and give me its initial feedback about what it finds. I’m simply providing some initial background information on the project and clarifying what the AI should expect with a conversational prompt: “I'm working on a messaging and copywriting project for TeamFlow, a B2B SaaS company. I'll share all our research materials so you can have the context you need to help me.”

The model responded with a detailed clarifying question about what I need help with.

Source: Google AI Studio.

The next step is to extract insights from our raw data that we’ll need to empathize with our customers and write copy that resonates. In my prompt, I’m doing the following:

• Setting the context and expectations: “Before writing anything, I will share [the following materials] and ask you for help.”
• Clarifying my goal: “Help me analyze [the materials] and identify key insights.”
• Providing the framework and questions to answer: “Focus on struggles and challenges, motivations and goals, desired outcomes and recurring themes.”
• Asking for a specific format: “Provide your analysis in bullets, highlighting…”

I try to be as descriptive as possible by writing to the AI like I would to an intern. It responds with a detailed analysis of the interviews.

Like with any intern, you might need to guide it, especially with long documents such as interview transcripts. So I ask it to pick up the analysis when, for example, it forgets to include the last two interviews in the previous result.

You can view our full insights from customer interviews and all other assets in this report. In a couple of seconds we were able to distill thousands of words into a few succinct bullet points, including quick summaries from the interviews. Your team should be able to use this information to get aligned on your ideal customers, their struggles, their motivations, and what they want to achieve. 

What’s next?

We’ve laid the groundwork. We’ve gathered our customer data and analyzed it for insights. Now, how do we give our AI assistant a voice, a personality, and a unique perspective that reflects the complexity of our ideal customer so we can have a conversation with it?

That's where the next step in empathy engineering—articulation—comes in.

In the next part of this series, I’ll show you how to build AI personas that think, feel, and react like your ideal customers.

---

Chris Silvestri is the founder of Conversion Alchemy and a conversion copywriter for B2B SaaS companies. 

To read more essays like this, subscribe to Every, and follow us on X at @every and on LinkedIn.

We also build AI tools for readers like you. Automate repeat writing with Spiral. Organize files automatically with Sparkle. Write something great with Lex.