I or we? The buyer voice your AI visibility prompt set needs

An AI visibility prompt set needs both individual and team buyer voice. Parse analyzed 14,977 active organic prompts and found 5,399 that begin with I, my, we, or our. Of those first-person prompts, 1,676, or 31.0%, speak for a team using we or our. A set written only as personal questions misses nearly one-third of the first-person buyer framing already present in the public index.

First-person prompts split into two buyers

The two voices ask different kinds of questions. Singular voice usually describes a personal outcome, preference, or risk: "I need a simple way to manage recurring payments" or "My pet has a pre-existing condition." Team voice describes organizational coordination: "We need to integrate with our ERP" or "Our security team requires single sign-on and audit logs."

Both can be commercially important in the same category. A CRM buyer might ask "I need a simple CRM for freelance consulting" and "We need a CRM that routes inbound leads across five territories." The product category is identical. The stakeholders, workflow, and likely shortlist are not.

That difference is why prompt persona cannot be treated as a label added after collection. The voice belongs inside the prompt. It shapes what the engine understands as the buyer, the scope of the problem, and the evidence needed to support a recommendation.

Team voice concentrates in operational categories

Among active categories with at least 10 prompts and at least 3 team-voice prompts, the highest team shares are concentrated in software used across organizations.

These are small category panels, so the exact percentages should not be treated as market estimates. The consistent pattern is the useful part. Products that coordinate teams, systems, compliance, or shared spend are often evaluated in organizational language. A first-person singular prompt set can still find these categories, but it may retrieve a different competitive field from the one a buying committee creates.

The pattern is especially important for the B2B SaaS AI visibility playbook. B2B prompt research should not merely add "for enterprise" to a generic question. It should reflect who is speaking, what system the team already uses, and what the organization must coordinate.

Why persona tags are not enough

Scrunch recommends organizing prompt libraries by persona, funnel stage, country, and other dimensions. That structure is useful, but a tag cannot repair a prompt written in the wrong voice. A row tagged "IT leader" can still contain a consumer-style question.

Compare these two prompts:

• "What is the best identity platform for enterprise security?"
• "We need to replace a legacy identity provider without breaking contractor access. Which platforms support staged migration and complete audit logs?"

The first is a broad enterprise category query. The second gives the engine an organizational change, a migration risk, a user group, and an evidence requirement. Both belong in the set, but they answer different planning questions. The broad prompt measures inclusion. The team prompt measures operational fit.

The production benchmark on what real AI buyer prompts look like shows that 81.1% contain at least 10 words. Buyer voice is one reason. When a person describes who is making the decision and what they need, the prompt naturally becomes more specific than a head term.

If you want to know when AI changes its answer about your brand, start with a free brand check — it takes a minute.

Build an individual and team pair only when the answer can change

The finding does not justify doubling every prompt. Persona variants are useful only when changing the speaker could change the recommendation.

Create an I and we pair when at least one of these changes:

• the number of users or stakeholders;
• the approval, compliance, or security process;
• the integrations or existing systems;
• the contract, billing, or procurement requirement;
• the acceptable implementation effort;
• the source evidence a cautious buyer would expect.

Do not create a pair when the answer is likely identical. "I need a weather app" and "We need a weather app" do not automatically represent two commercial intents. Variants should protect a real decision boundary, not inflate the prompt allowance.

This is the same discipline behind the category prompt-depth benchmark. Count decisions first. Add variants only where a persona, organization, or constraint changes which brand should win.

How team voice changes reporting

A blended score can hide a useful split. If a brand wins personal prompts and loses team prompts, the marketing problem is not general awareness. It may be enterprise proof, integrations, implementation content, security documentation, or third-party validation.

Report at least three fields for first-person prompts:

1. Buyer voice: singular or team.
2. Buyer decision: discovery, fit, trust, price, or displacement.
3. Commercial weight: how closely the question maps to pipeline or retention.

Then compare the brands and evidence by segment. /rankings helps expose category questions. /brands shows where a brand appears across those questions. /sources shows whether team prompts pull more documentation, vendor pages, reviews, or institutional evidence.

The weekly review should turn a split into an action. If team prompts consistently recommend competitors, inspect the sources that justify enterprise fit. If singular prompts lag, inspect usability, price clarity, and consumer proof. A single aggregate score cannot tell the team which evidence layer is missing.

A prompt-persona audit for campaign planning

Export the current set and classify each row as neutral, singular, or team voice. Then ask:

• Do product categories used by teams contain any we or our prompts?
• Do consumer categories contain authentic personal goals instead of brand-written personas?
• Does each team prompt name a coordination, system, risk, or approval requirement?
• Are team prompts overrepresented merely because every sentence was rewritten with we?
• Can the reporting view compare singular and team outcomes without changing the base set?

Use actual sales-call and support language to revise the weak rows. The goal is not conversational decoration. It is to capture the identity of the decision-maker and the unit they are buying for.

The existing prompt-set construction guide explains how to gather and group questions. This benchmark adds a precise audit: among real first-person questions already represented in the index, nearly one in three speaks for an organization.

How we measured this

We queried the production Prompt table through Cosmo under SET default_transaction_read_only = on, a repeatable-read transaction, and a 60-second statement timeout. The snapshot was taken on August 29, 2026. Eligible prompts were organic, active, visible, unpaused, and unarchived.

We lowercased and trimmed each prompt, then extracted its first word. I and my form the singular group; we and our form the team group. The primary query returned 3,723 singular prompts and 1,676 team prompts. An independent query used direct case-insensitive prefix conditions and reproduced both counts exactly.

The category table is limited to active categories with at least 10 prompts and at least 3 team prompts. Category prompt panels are small, so those percentages illustrate where team framing appears, not population-level buyer behavior. The full-panel 31.0% share is the stronger headline.