Capture authentic customer voices from the web with AI

This is a clear, actionable guide to capturing real customer voices and turning them into useful input for product opportunities.

It shows how to structure research and give you an idea on how to set up a system to get regular user feedback easily. The examples are real prompts I used for Beyond, but you can adjust them to fit your needs.

Before using AI for research, I set the stage and let it determine the core problems and ideal customer profile (ICP) on its own. I do this to reduce my own bias. First-principles thinking helps here.

-

Step 1: Set the scene and context for ChatGPT

I begin by giving ChatGPT context to lay the groundwork before diving into specifics. This has consistently helped me get high-quality, focused responses. Don’t skip this step.

To check if ChatGPT understands my goal, I often have it generate something, like a new approach. Most of the time, the output isn’t particularly useful for the end results but it helps to put the AI into the zone.

Model

ChatGPT o1 Pro + Deep research

Time

This step usually takes around 20 minutes

Prompt

You are a startup founder looking to invent a completely new way for product managers to create decision trees.

Before jumping into solutions, start with ground research:

• How product managers currently use decision trees
• Common pain points and limitations
• Existing tools and their drawbacks
• Relevant insights from decision-making science

Once you have a clear picture, propose a new approach. Focus on what would make decision trees more effective, intuitive, and valuable for product managers. Go!

-

Step 2: Define ideal customer profile (ICP)

To ensure objectivity, I avoid focusing on specific products or solutions at this stage. Instead, I define the characteristics of the ideal customer profile (ICP).

Model

ChatGPT 4.5 + web

Time

This step usually takes a few seconds

Prompt

Browse the web and understand how the ideal ICP for this should look like.

• Who currently uses similar solutions
• Their roles, challenges, and needs
• Industry trends and common pain points
• Any existing insights from research or discussions

Summarize the key characteristics of the ideal ICP based on your findings.

-

Step 3: Clarify core customer problems and how they solve it today

Model

ChatGPT o1 Pro

Time

This step usually takes a few seconds

Prompt

Great, now focus on the niche ICP: product managers who use decision trees in their daily work.

Summarize briefly in bullet points:

• ICP: Who they are (role, company type, experience level, etc.)
• Core problem: What they’re trying to solve with decision trees (e.g., OST, impact mapping)
• Current solutions: How they approach it today and the limitations they face

-

Step 4: Capture authentic customer Voices from the Web

Model

ChatGPT o1 Pro + Deep research

Time

This step usually takes a while. You can run it multiple times as well.

Prompt

Research real customer feedback on tools used today to solve this core problem. Focus on where current tools fall short. Specifically, look into Vistaly, Miro, FigJam, DoubleLoop, and similar tools.

Key requirements:

• Only use direct user quotes, don’t invent anything
• Provide a link to each quote for full context
• Capture both product managers’ perspectives and stakeholder complaints (e.g., Head of Product)
• Cast a wide net but stay focused on the ICP
• Take as much time as needed. I need extensive user feedback

Format the findings as bullet points with user quotes and links. Go!

-

Step 5: Analyze & identify competitive gaps to inform opportunities

Model

ChatGPT 4.5

Time

This step usually takes a few seconds

Prompt

Based on the collected user feedback, identify potential opportunities for my product.

Key focus areas:

• Gaps and pain points in existing tools
• Unmet needs mentioned by users
• Recurring frustrations that could be solved differently
• Any workflow inefficiencies or stakeholder challenges

Keep it directly tied to the user feedback, no assumptions or invented insights.

Next steps

The result should be a list of real customer voices. You might need to refine the approach a few times to get it in the right format.

You can also include extras like a “relevance score” or “authenticity” rating. This was my first time testing deep research, and I think a few more iterations could improve the output even further.

Listening to customers isn’t a one-time effort, it’s an ongoing process that drives lasting success. Automating this is the next step.

Ideally, you’ll reach out to some of these people for real interviews to dig deeper. This is just the starting point to get a broad view of how your ICP sees the problem, their current solutions, and where they struggle.