Abstract
The quality of AI visibility reporting depends on the quality of the prompts being measured. A portfolio built from convenient or favorable questions will produce a favorable but misleading score. This paper presents a protocol for prompt selection, stratification, weighting, and change control.
Definition
First Brand defines a prompt portfolio as a controlled set of natural-language questions used to measure how AI systems represent, cite, and recommend brands. It is not a keyword list. Each prompt includes context such as audience, task, geography, category, and stage of decision. The portfolio is the sampling frame for every visibility metric that follows.
Method
Begin with real buyer language from sales calls, site search, support tickets, community discussions, and search demand. Group prompts by intent: learn, compare, validate, select, and act. Add audience and market dimensions. Remove prompts that differ only cosmetically. Freeze a core portfolio for longitudinal reporting and document every later addition, retirement, or wording change.
Findings
Prompt phrasing can materially change the answer. Brand-neutral prompts reveal category visibility. Comparison prompts reveal differentiation. Constraint prompts reveal whether the system understands use cases, price, geography, or risk. Branded prompts are useful for entity accuracy, but they should not dominate a visibility score because the brand is already supplied in the question.
How to apply the framework
Assign each prompt an owner, rationale, intent, audience, market, and reporting weight. Keep a visible coverage matrix. Run periodic gap reviews against new customer questions and product changes. When a new prompt enters the core portfolio, begin a new baseline instead of pretending the historical series is unchanged. Preserve the old results for auditability.
Limitations
A portfolio is always a sample of a larger and changing question space. It cannot predict every user phrasing. It also does not reveal the complete retrieval process behind an answer. The goal is disciplined comparison, not exhaustive simulation. Discovery prompts should explore change without contaminating the stable trend line.
Frequently asked questions
Should prompts include brand names?
Use a minority of branded prompts to audit entity accuracy. Use brand-neutral prompts to measure discovery and recommendation performance.
When should a prompt be retired?
Retire a prompt when the buyer need, product, market, or wording is no longer material. Record the reason and preserve its historical data.
Primary sources
How to cite this paper
First Brand Research. “Prompt Portfolio Design for Reliable AI Visibility Tracking.” Published August 15, 2026. firstbrand.ai/research/prompt-portfolio-design-ai-visibility.