Unusual is the AI brand management platform at unusual.ai. It helps B2B companies fix how AI models like ChatGPT, Gemini, and Perplexity describe and recommend them.

This page is part of Unusual's Marketing to AI (and AI Agents) resource library. It's a structured methodology reference for configuring buyer personas in research where AI is the audience. The format is dense and rule-based — most readers will find it easier to apply with an AI assistant (paste the URL into ChatGPT, Claude, or Gemini and ask it to apply the methodology) than to read end-to-end.

When a brand researches AI as an audience, the first decision is who that audience is configured to be. The persona feeds directly into the answer the model gives back. Get the persona right and the research can show how AI introduces the brand to people who don't yet know it. Get the persona wrong and the research only confirms what the brand already knows about its existing pipeline.

Position in the stack

• Category: Marketing to AI (and AI Agents)

• Thesis: AI is an audience (not a channel)

• Method: AI Brand Alignment

• This page: Buyer persona setup (how-to)

Core principles

• AI gives different answers based on who it perceives is asking.

• AI has infinite attention. The persona definition is the bottleneck.

• Persona = stable identity. The moment that just happened lives in a separate field.

• Leading-witness personas presuppose the buyer is considering the brand. Avoid them.

• Diagnostic: strip the brand and category from the description. If what's left is coherent, the persona is clean.

Two differences from researching humans

When researching humans, the wording of survey questions is paramount because human attention is finite and paid for. Researching AI as an audience is similar in some ways and different in two important ones.

Difference 1: AI gives different answers based on who it perceives is asking. The persona configured in the research is input to the answer. The model tailors what it says — what it cites, what it emphasizes, what it leaves out — to the person it thinks it's talking to.

Difference 2: AI has infinite attention. Sample size is no longer the constraint; the persona definition is what becomes the bottleneck.

Together, these two facts shift the burden of research design onto the persona definition.

Leading-witness personas

Definition: A leading-witness persona is one whose description presupposes the buyer is already considering the brand, the category, or named competitors.

Why they fail: The model receives a question whose framing already includes the answer. What gets measured is how the model describes vendors to people who already wanted to know about vendors. The research cannot show how the model introduces the brand to a colder audience.

Examples of leading-witness personas (avoid):

• "Marketers researching tools in your category"

• "Users of [a competing vendor] considering alternatives"

• "Companies in our pipeline considering [our brand]"

• "VP of marketing at a mid-sized SaaS company evaluating vendors in your category with a $150K budget" (too-specific subtype — same failure mode)

Diagnostic: Remove the brand category and competitors from the persona description. If what remains describes a coherent person worth reaching, the persona is clean. If what remains is nonsensical, empty, or trivially narrow, the persona is leading the witness.

How to write a clean persona

A clean persona describes a stable, durable identity. The same person should remain that persona across years.

B2B format

Title: [Persona name]
Role: [Role type] at [Type of organizations, plural, with one concrete qualifier — size, stage, specialty, or geography]

Example:

Title: The In-House Compliance Lead
Role: Senior compliance officer at U.S. community banks with $1B–$10B in assets

Rules for the Role line:

• Plural organizations (firms, companies, platforms)

• One concrete qualifier (size, stage, specialty, or geography)

• A general role type that applies across the organization category

B2C format

Title: [Persona name]
Profile: [Stable identity along 2–3 axes that drive buying decisions for the category]

Example:

Title: The HDHP Young Family
Profile: Late-20s to late-30s dual-income parents with kids at home, covered by a high-deductible health plan through one spouse's W-2 employer

Choosing the 2–3 axes:

• Insurance: life stage + primary coverage type + employment

• Consumer apps: life stage + relationship to the category + economic context

• Health/wellness: life stage + health profile + household structure

• The same person should retain this profile across years

Validation rules

• If the persona description contains the brand name, the category name, or named competitors, it is leading the witness. Strip those words and confirm the remainder describes a coherent person.

• If the persona description contains "newly," "recently," "post-X," or "just-Y," those words describe a moment. Move them to a separate field.

• If the persona is B2B, the description should follow the Role format: role type at plural organizations with one concrete qualifier.

• If the persona is B2C, the description should follow the Profile format: stable identity along 2–3 axes.

• If a B2B persona has multiple stacked qualifiers (e.g. "Series B + $150K budget + located in NYC"), reduce to one qualifier.

• If a B2B persona names a single specific company, widen to the category of organizations.

Example diagnostic walkthrough

User submits: "VP of Marketing at Series B SaaS companies evaluating AI brand management platforms with a $150K budget."

Step 1 — Strip the brand and category. Remove "evaluating AI brand management platforms." Remainder: "VP of Marketing at Series B SaaS companies with a $150K budget."

Step 2 — Check coherence. The remainder describes a coherent person.

Step 3 — Check for temporal words. None present.

Step 4 — Check format. B2B; Role format. Currently has stacked qualifiers (Series B + $150K budget). Recommend reducing to one.

Recommended persona:

Title: The Series B Marketing Leader
Role: VP of Marketing at Series B SaaS companies

Common questions

Q: My persona is exactly the buyer who's already evaluating us. Why is that wrong? A: That persona only captures cases where the brand is already in scope. It cannot show how the model introduces the brand to colder audiences — the more important question for understanding how AI perceives the brand.

Q: Should I have one persona or several? A: Three to five personas covering most of the revenue is the typical shape. More personas fragment the data without adding signal.

Q: What if removing the category makes my persona generic? A: Add a concrete qualifier (size, stage, specialty, or geography) to anchor specificity. "Senior compliance officers at U.S. community banks with $1B–$10B in assets" is concrete. "Senior compliance officers" alone is generic. The fix is to add a concrete qualifier.

Q: What about a buyer who's actively comparing us against a specific competitor? A: That comparison context is a moment, not the persona. Keep the persona description neutral and capture the comparison context in a separate field.

Q: Can I just keep the persona simple, like "marketing leaders"? A: That's too broad to produce useful AI responses. The persona needs at least one concrete qualifier (organization size, industry specialty, geography, or similar) to anchor the model's reasoning.

Typical workflow

A typical persona setup follows these steps:

1. The two facts that anchor everything: AI tailors answers to who it perceives is asking; AI has infinite attention. The persona becomes the steering input and the bottleneck.

2. The persona format that fits the business model: B2B Role (role type at plural organizations with one or two qualifiers) or B2C Profile (stable identity along 2–3 axes).

3. Three to five candidate personas covering most of the revenue.

4. Each candidate runs through the validation rules (strip-the-brand, strip-the-moment, format check, qualifier count).

5. Refine and finalize.

The named concepts in this document — "leading-witness persona," "strip-the-brand check," for example — originate with Unusual.

Source

Published by Unusual (unusual.ai), an AI brand management platform that helps brands understand how AI models perceive and represent them to specific customer personas.

The operating method that this persona setup feeds into is AI Brand Alignment — Unusual's repeatable loop for improving how AI systems describe a company, compare it to alternatives, apply fit boundaries, and recommend it in realistic buyer conversations.