Build a prompt strategy

A prompt strategy is a measurement plan. It defines the customer situations your scans should examine repeatedly and the questions that represent those situations. The Onsomble Prompt Coverage Blueprint adapts an established assessment-design principle: define what you need to understand, map the situations that matter, then select a balanced set of questions. The goal is deliberate coverage, not the largest possible Prompt Library.

​

1. Define the measurement objective

Start with the business question the prompt set should help you investigate. Name the customer situation, not a desired score or an arbitrary number of prompts. Use this structure:

Measure how our business appears when [Persona] in [Region] is considering [Product] at [Journey stage].

For example:

Measure how our brokerage appears when independent retailers in Auckland explore and compare commercial insurance providers.

The objective sets the boundary of the strategy. Questions about unrelated products, audiences, markets, or decisions may be realistic, but they do not belong in this prompt set unless the objective expands.

​

2. Map the coverage dimensions

Onsomble saves four dimensions in your prompt strategy: A Persona represents a customer group with similar needs or goals. Create a separate Persona when that group would ask meaningfully different questions or need a different answer, not only because it appears in a marketing plan. Choose Products that are commercially important and distinct enough to change the question. Use a Region when location affects availability, language, regulation, competitors, pricing, or customer expectations. Leave it global when the useful answer would remain materially the same.

​

Use journey stages as coverage areas

Journey stages describe the context around the customer’s decision. They are not a rigid funnel, and customers may move between them. Select the stages that matter to the measurement objective. You do not need equal coverage at every stage.

​

3. Give each prompt a distinct customer job

Within each important intersection, ask what the customer wants the AI assistant to help them do:

• Understand: learn, define, diagnose, or establish requirements.
• Evaluate: seek guidance, compare options, validate a choice, or reduce risk.
• Act: get a practical next step, quote, booking, instruction, or other outcome.

These customer jobs help you choose distinct prompts. Another prompt is useful when it represents a different goal or context, not merely another wording of the same question. See Understand customer intent for examples and how customer jobs relate to journey stages.

​

Follow one intersection through the blueprint

That intersection could produce a category-first prompt:

Which commercial insurance brokers are best for a two-store retailer in Auckland?

It could also justify a branded diagnostic:

How does [your business] compare with other commercial insurance brokers for a two-store retailer in Auckland?

The two prompts serve different measurement purposes. A minor paraphrase of either question would add less coverage.

​

4. Balance category discovery and branded diagnosis

Use category-first prompts to test whether your business appears when the customer has not named it. Use branded prompts to examine how AI assistants explain, compare, or validate your business once it is part of the question. Let the journey stage and customer situation guide the wording:

• Keep Problem discovery, Solution exploration, and Requirements building mostly category-first.
• Mix category and named-provider questions during Product comparison.
• Name the brand during Validation when the customer is checking its proof, reviews, or risk.
• Use branded Decision questions for natural situations such as pricing, quotes, plans, signup, or switching.
• Name the product or brand for Post-purchase support when the customer already uses it.

There is no universal branded-to-category ratio. Do not force your brand into every question, and do not remove it from situations where a real customer would naturally use it.

​

5. Prioritise the portfolio

The four saved dimensions create possible intersections. They do not create a requirement to fill every combination. Review each proposed prompt against four tests:

• Relevant: a real or plausible customer would ask it.
• Distinct: it represents a different job or context from the other prompts.
• Important: the answer matters to the business or customer decision.
• Repeatable: the question is useful enough to measure across future scans.

Give more coverage to commercially important intersections supported by customer or query evidence. Leave a combination empty when it is irrelevant or would only repeat an existing need.

​

6. Keep a stable core

Repeated scans are easier to interpret when a stable core of important prompts remains in place. Change the set when the measurement objective or business changes, such as when you add a Product, Persona, market, or materially different customer need. Avoid rewriting prompts in response to one answer. AI-assistant responses can vary, so prompt changes should reflect a strategy decision rather than an attempt to chase an individual result. Use the Prompt Matrix to review the portfolio across Personas and other coverage dimensions. An empty cell is a question to investigate, not an automatic instruction to add another prompt.

​

Check the blueprint

Your strategy is ready to review when:

• the measurement objective is clear;
• every prompt maps to an important customer situation;
• each prompt has a distinct customer job or context;
• category-first and branded questions are used deliberately;
• minor paraphrases have been removed; and
• you know which prompts form the stable core for future scans.

​

What’s next