Chapter 9
Section Three · Applied
The Research Layer
Audience and Market Research Skills
Five skills that span the research spectrum, from synthesising your own customer data to building grounded personas, extracting jobs-to-be-done, conducting competitive intelligence, and stress-testing segmentation.
Chapter 9
Two skills shown in full
·
No email required
Section 9.1 — Why Research Skills Come First
Every marketing decision, what to say, how to say it, where to say it, to whom, depends on understanding the audience. Not in the abstract, the-customer-is-always-right sense, but in precise, specific, actionable terms. What does this segment actually believe about this problem? What words do they use when they describe it? What would have to be true for them to change their behaviour?
"Most marketing teams do not lack research. They lack research that is synthesised, structured, and made usable at the speed of marketing decisions."
Claude's research skills do not replace primary research. They do not invent data. What they do is make existing research dramatically more useful by synthesising it, structuring it, extracting what matters, and making it available in the format you actually need when you need it.
The Research Workflow · how the five skills connect
↕ Parallel
9.5 — Competitive Intelligence Analyst, supplying the external context positioning operates within.
This chapter develops five skills across that workflow. Each ships with a complete system prompt, worked examples, usage guidance, and the reasoning behind the design choices. Two are shown in full below.
9.2
Skill · Research Layer
Voice of Customer Synthesiser
The Problem This Skill Solves
Your customers are already telling you what matters to them: support tickets, interview transcripts, review sites, NPS comment fields, community forums, sales call notes. The problem is not a lack of signal. The signal is buried in unstructured text, spread across systems, and impossible to process at volume without spending days on it.
VoC synthesis takes that raw qualitative data and draws out the patterns: what customers are trying to accomplish, what frustrates them, what they praise, what language they use, and what gaps exist between their expectations and their experience.
The Voice of Customer Synthesiser takes a batch of raw customer inputs in whatever form you have them, and produces a structured synthesis that is immediately usable for messaging, positioning, and content strategy.
Skill Design Rationale · three decisions
Format-tolerant input (mixed paste of interviews, tickets, reviews, open-ends) · strategic-utility output (language for messaging, jobs for positioning, gaps for content) · verbatim language preserved throughout, since customer phrases outperform marketing-generated copy.
SkillVoice of Customer Synthesiser
Produces
Structured synthesis of customer voice across five dimensions
Input Required
Raw customer text (interviews, reviews, tickets, survey open-ends)
Output Format
Structured report with verbatim language bank
Deployment · Best Model
Claude.ai Project · Sonnet 4.6 (Opus 4.7 for large batches)
The System Prompt · Part 1
System PromptVoice of Customer Synthesiser
Role Definition
You are a senior customer insights analyst with deep expertise in Voice of Customer research. Your role is to analyse batches of raw customer-generated text and produce structured, strategically useful syntheses for marketing teams.
Operating Principle
Every finding must be grounded in actual customer language and repeated evidence patterns. Do not extrapolate beyond the strength of the available signal. If the sample is thin, explicitly state confidence limitations.
Required Inputs
Raw customer text (reviews · interviews · survey responses · support tickets · community posts · sales transcripts) · optional product or service context · optional audience segment context.
Input Confirmation Rule
Begin each response by confirming: type of input data received, approximate volume of records or text processed, and data source type.
1. Jobs To Be Done
Identify primary and secondary customer jobs in the structure: When [situation], I want to [motivation], so I can [outcome]. Include at least 3 primary jobs where signal supports it.
2. Language Bank
Extract verbatim customer phrases only across: problem language · desired outcomes · frustrations · evaluation criteria · product or service experience.
Critical Language Rule
Preserve customer wording exactly, in quotation marks. No paraphrasing inside the Language Bank section.
The System Prompt · Part 2
System Prompt (cont.)Voice of Customer Synthesiser
3. Pain Points & Frustrations
For each pain point: description · frequency (common / occasional / isolated) · most vivid verbatim quote.
4. Delight Drivers
For each delight driver: what customers value · positive surprise factor · frequency · direct quote.
5. Gaps & Unmet Needs
Identify missing expectations, desired features, service gaps, and alternative comparisons relative to competitors or substitutes.
6. Messaging Implications
Convert findings into 4–6 concrete messaging recommendations covering: claims to emphasise · exact customer language to reuse · proof points · objections to pre-empt.
Evidence Rule
If a dimension has weak evidence, state explicitly: INSUFFICIENT SIGNAL FOR RELIABLE FINDING.
Output Style
evidence-led · verbatim-rich · messaging-actionable · insight-confidence aware
What good output looks like
A well-executed synthesis gives you three immediately useful outputs: a language bank you can hand to a copywriter as direct source material, a JTBD framework you can use to structure positioning, and a messaging implications section that translates research into actionable guidance without requiring you to interpret it yourself.
The test of a good synthesis is whether someone who did not read the source data would come away with an accurate and usable understanding of the customer. If the synthesis is so general it could apply to any product, it has not done its job.
Using the Voice of Customer Synthesiser
How to Use This Skill
Prepare your source data before opening the conversation. Paste it in whatever format you have it: interview transcript excerpts, review text from G2 or Trustpilot, NPS open-end exports, support ticket descriptions. The skill handles mixed-format input.
Provide context at the start: what product or service the data relates to, what segment the customers represent if known, and what question you are most trying to answer (reasons for churn, informing a repositioning, building a messaging framework).
Worked Example · Initial Message
"Below is a batch of customer data for our project management software, specifically from the small business segment (under 20 employees). It includes six interview excerpts, forty NPS verbatim comments from detractors and passives, and fifteen G2 reviews. I'm trying to understand why this segment's retention is lower than our mid-market segment. Please synthesise this."
Follow-up prompts that work
- "Expand on the third job to be done."
- "What are the most copy-ready phrases from the Language Bank?"
- "Based on these gaps, what content topics would address the most common unmet needs?"
◆
The test of a good synthesis: someone who did not read the source data should come away with an accurate, usable understanding of the customer. If the synthesis is so general it could apply to any product, it has not done its job.
9.3
Skill · Research Layer
Persona Builder
The Problem This Skill Solves
Marketing personas have a poor reputation in many organisations, earned by years of fictional character profiles that bear no relationship to actual customer research. The persona with the stock photo, the name, the hobby, and the vague aspiration that could describe half of humanity.
⚠ Watch Out
This kind of persona is worse than useless. It creates the illusion of audience understanding while providing none, and inferred attributes get treated as fact downstream.
A well-constructed persona is built from data, not imagination. It represents a real segment with real purchase behaviour, real decision-making patterns, real language, and real objections. It is specific enough that a content writer can ask "would this person care about this?" and get a reliable answer.
Skill Design Rationale
The critical decision is the explicit separation of evidence from inference. Every attribute is tagged as either grounded in research or inferred from the pattern of evidence. Personas used without this distinction get treated as fact, and inferred attributes become assumptions that distort downstream decisions.
SkillPersona Builder
Produces
One or more grounded marketing personas from research inputs
Input Required
Research data (VoC synthesis, interview notes, segment data, sales observations)
Output Format
Structured persona profile with evidence/inference tagging
Deployment · Best Model
Downstream of VoC Synthesiser · Sonnet 4.6 (Opus 4.7 multi-persona)
The System Prompt · Part 1
System PromptPersona Builder
Role Definition
You are a senior audience strategist specialising in building evidence-grounded marketing personas. Your personas are used by content teams, copywriters, and campaign managers to make targeting and messaging decisions.
Operating Principle
Every persona attribute must be clearly tagged by confidence level. Never invent demographic or behavioural details unless directly supported by evidence.
Required Inputs
Research inputs (interviews · VoC · sales observations · CRM notes · support tickets · survey data) · optional product or category context · optional segment / ICP definition.
Input Confirmation Rule
Begin by confirming: research sources used · approximate evidence volume · evidence strength (strong / moderate / thin).
1. Persona Name & Segment
Use descriptive segment-led labels only (e.g. The Overstretched IT Manager). Include one-sentence segment description.
The System Prompt · Part 2
System Prompt (cont.)Persona Builder
2. Evidence Summary
Briefly note sources used and approximate volume. Explicitly state confidence level of the evidence base.
3. Situational Context
Role & responsibilities · buying / decision position · current pressures & priorities · operating time horizon.
4. Goals & Outcomes
Distinguish: stated goals · underlying goals / true drivers.
5. Frustrations & Obstacles
Blockers · failed previous attempts · perceived causes of failure.
6. Information Landscape
Information sources · trusted voices / communities · preferred content formats · research triggers.
7. Evaluation Criteria
Comparison questions · required proof points · decision confidence signals.
8. Language & Voice
Exact phrases used by this segment · problem framing · desired-outcome framing · messaging language to use / avoid.
9. Objections
For each objection: stated objection · underlying concern it signals.
10. Evidence vs Inference
Tag every major attribute as: [EVIDENCED] · [INFERRED] · [ASSUMED] archetype assumption requiring validation.
11. Differentiation Summary
For multi-persona work, include a comparison table on dimensions that affect targeting and messaging.
Critical Rule
Do not invent demographics, lifestyle traits, or psychographics unless explicitly evidenced.
Thin Evidence Rule
If evidence is limited, label the output as PROVISIONAL PERSONA and list additional research needed.
Output Style
evidence-led · segment-usable · messaging-ready · standalone
Using the Persona Builder
How to Use This Skill
This skill works best downstream of the Voice of Customer Synthesiser. Paste the synthesis as the primary research input and specify how many personas you want, at what level of differentiation, and any distinct segments (by company size, role, use case, or buying behaviour).
If you are building personas without a prior VoC synthesis, provide whatever research you have and be explicit about its nature and quality. Claude will work with what you give it and flag where the evidence base is thin.
Worked Example · Initial Message
"Below is a VoC synthesis from our customer research, plus notes from eight sales calls with prospects who did not convert. I need two distinct personas: one for the in-house marketing manager at a mid-market company, and one for the agency account manager buying on behalf of clients. Please build these as separate profiles and include a differentiation summary."
Follow-up prompts that work
- "What content would this persona most likely engage with at the awareness stage?"
- "Given Persona 2's evaluation criteria, what is the strongest proof point we should lead with?"
- "Show me the messaging that would actively alienate Persona 1 so we can avoid it."
◆
Claude retains the persona context throughout the conversation and can reason about it. The persona becomes a thinking partner: "Would this person care about this content?" returns a reasoned answer, not a guess.
What comes next
Three more research skills and 38 more across the full marketing workflow.
Chapter 9 covers five research skills in total. The rest of the book applies the same methodology across strategy, planning, execution, QA, and analysis — 42 skills across 10 disciplines, each with the full system prompt formatted for deployment.
404 pages. £19. Gumroad instant download.