Building Persona with AI

My Promting workflow:

1. Pull raw signal data with Gemini

2. Claude sorts the evidence

3. Claude builds the persona

Step 1: Raw Evidence Gathering

Goal:

I use Gemini's live-web capabilities to pull real, messy, unfiltered user complaints, workarounds, and direct quotes from the internet. I not use ChatGPT or Claude for this step, as they are prone to generating "designer speak" summaries rather than raw data.

The Prompt to use in Gemini

Role:

You are doing primary research for a product team. At a later date, we are going to build a persona designed to help us cut features from a roadmap, not just build more.

Task:

Pull raw evidence about what [Target User Group, e.g., fitness people/power users] are saying about [Industry/Specific Apps, e.g., calorie tracking apps or Competitor Name]. Use current live web sources.

For each finding, extract:

The direct quote (verbatim).
What the person was trying to do (intent).
What tools or workarounds is the user already using instead?
What they explicitly said they stopped using or do not care about.

Constraints: Find 25+ sources. Skip marketing content. Do NOT summarize or categorize into themes yet. We only want the raw evidence.

(Note: Turn on "Deep Research" / "Thinking" mode if available, then export the final output to a Google Doc or Markdown file).

Step 2: Pattern Extraction & Filtering

Goal:

Take the raw research document from Step 1 and feed it into Claude to isolate behavioral friction points, contradictions, and what users actually care about.

The Prompt to use in Claude:

Role:

You are doing primary research for a product team. At a later date, we are going to build a persona designed to help us cut features from a roadmap, not just build more.

Task:

Review the attached raw research document. Extract patterns from this data that we can use to kill features before they make it onto our product roadmap.

Produce 4 distinct lists, citing the source numbers from the raw research for every claim:

Jobs Already Done: Problems users solve fine with simple workarounds that they are not waiting for us to fix.
Accepted Friction: Things they complain about, but it doesn't cause them to switch products or churn.
Pay to Fix: Severe pain points that drive users crazy enough to switch products or churn.
Contradictions: Things users say they want versus what they actually do (e.g., asking for more features but only using 3).

Format:

State the pattern, followed by how many times it occurred in the data, and the specific examples.

Step 3: Persona Generation

Goal:

Transform the filtered patterns into a scannable, actionable persona document equipped with strict criteria to automatically greenlight or redlight feature proposals.

Task:

Using all the pattern data analyzed above, build one distinct persona designed specifically for "kill decisions" (shutting down features before they reach the iteration stage).

Output EXACTLY the following structure and nothing else:

Name & Role: (Give them a real name like Dan or Sarah, not a marketing archetype. Include a specific situational context, like company size or tenure.
Context (Situational, not demographic): A 1-line description of their chaotic workflow or current process. Focus on why we are here, solving this problem.
Top Goal: (Must be highly specific, task-oriented, and quantifiable in real dollars or time—e.g., "Reduce reporting time from 3 hours to 30 minutes"—not holistic career goals).
Top Frustration: (Must be tangible and relatable—something that makes them angry on a Friday afternoon).
Real User Quote: (Must sound like a real human being expressing annoyance, completely free of designer buzzwords like "seamless integrated workflow").
Kill Criteria: 3 to 5 bullet points using the framework: "I would ignore or reject a feature that [X], because my real problem is [Y]."

Summary Checklist for Success:

Keep it scannable: A team member or executive should be able to read and understand the entire persona in 10 seconds. Avoid multi-page essay bloat.
Test it against proposals: When a stakeholder proposes a new feature, run it by your new persona. If the persona's "Kill Criteria" flags it, drop the feature immediately.

The Prompt to use in Claude