Buyer Persona Generator

The constraints

What made this hard?

Three constraints. The first was time. The whole product had to fit inside three minutes, because anything longer competes with calendar reality and loses. Four steps, no skip-able bloat.

The second was specificity. A persona generator that returns "decision-maker who values ROI and ease of use" doesn't beat what the founder already had. The output needed a specific name, a specific salary, a specific daily reality, specific fears, specific aspirations, and a specific closing sentence the founder could lift directly into copy. Generic kills the tool.

The third was psychology over demographics. Most ICP frameworks are demographic (industry, size, role, seniority) and produce demographic outputs. Buyers don't decide by demographic; they decide by emotional pressure, social proof, and fear of being the one who picked wrong. The tool needed buyer psychology in the prompt scaffold, not as an afterthought.

The approach

How did Tincture frame the problem?

Four steps. Step one captures the category and use case, with enough context for the model to anchor. Step two captures the decision context: who actually buys, why now, what alternatives they're weighing, what's pushing them to act. Step three captures emotional drivers: what they fear (being the one who picked wrong, looking out of touch, missing the window), what they aspire to (looking like the smartest person in the room, getting the budget approved next year, being the one who knew first). Step four reviews and refines.

The output is structured for direct copy lift. Name and surname (psychologically real). Salary (so the founder thinks about budget signals). Daily reality (the kind of week they're having). Fears and aspirations in their actual language. A closing sentence: one line that captures who they are and what they want, ready to paste into a homepage hero or a cold email opener.

The model layer is a Claude API call with a structured prompt that scopes archetype generation against buyer psychology principles, so the output reads like a person rather than a market segment summary.

The build

What was shipped?

A web tool delivered as an overlay, hosted in the existing site framework. Four-step form with progressive disclosure (one section visible at a time, the previous steps' answers visible underneath). Step inputs structured to scope the model's output: category, use case, decision context, emotional drivers.

A single Claude API call per archetype generation, with the system prompt encoding the buyer psychology scaffold and the output schema. Output rendered in the overlay, copy-able to clipboard, downloadable as text.

A library of example archetypes accessible from the tool, so a founder can see what a strong archetype output looks like before generating their own.

The outcome

What were the results?

Live and free. Founders generate their archetype in three minutes and walk away with a specific person they can position copy, outreach, and product decisions against.

The compounding outcome is the language shift. Most founder copy reads as if it was written by someone who's never met the buyer, because it was. The archetype output gives the founder a specific phrase, a specific fear, a specific aspiration in the buyer's voice. That language drops directly into headlines, ad copy, cold email openers, and homepage heroes. The conversion shift from generic to specific copy is one of the largest single moves a founder can make on response rates.

The strategic outcome is what happens to the founder's mental model. After running the tool, the founder doesn't go back to "B2B SaaS, 50-500 employees" because the archetype is louder. The deck slide changes. The outreach changes. The product roadmap conversation changes.

What it took

What tools and methods were used?

Web app delivered as a front-end overlay (Next.js), Claude API for archetype generation, structured form for the four steps, system prompt encoding the buyer psychology scaffold and the output schema. The whole stack is small because the work is in the prompt scaffolding, not the surface.

The methodological underpinning is the practice's pattern for AI tools in research workflows: scope the model's output with structured inputs, encode the framework into the system prompt, and present the output in a format the user can lift directly into their work. Generic AI prompts produce generic output; scoped prompts against a framework produce framework-quality output.

The other move worth naming: psychology, not demographics. Most ICP tools default to demographic fields because they're the easiest to capture. Demographic outputs lose to psychology outputs every time on actual conversion. Building the tool around fear-and-aspiration prompts rather than industry-and-size pulls is what makes the output usable.

The takeaway

What's the transferable principle?

ICP is a person, not a market segment. The unit of analysis is the buyer making the decision, not the company that contains them. Most ICP work fails because it stops at the segment level (the deck slide, the demographic filter, the targeting list) and never produces the specific person whose language ends up in copy that converts.

For the Buyer Persona Generator, that meant designing the tool around the buyer's emotional pressure rather than the company's demographics. Three minutes, four steps, a specific person at the end with specific language the founder can lift directly.

The other transferable principle, broader than ICP: framework-encoded AI tools beat generic AI tools by the size of the framework. The tool's output is good because the prompt scaffolds the model's work against buyer psychology principles. Without that scaffold, Claude produces a confident-sounding archetype that doesn't survive contact with reality. The work in any AI tool is the prompt scaffolding, not the model.

Read more on this in ICP Definition: The 3-Attribute Framework That Cuts Through Vagueness