AIWorkDex
Marketing Practice

Build an evidence-backed customer-angle matrix

Produce distinct ad-angle hypotheses with traceable customer evidence, safe claim boundaries, and a human-reviewed brief for each test.

Preparation
A focused research and review session, followed by separate creative production and campaign testing.
Steps
8
Difficulty
Moderate
Task
Find evidence-backed ad angles

Requires: Write the audience, product, market, and campaign decision that the research must support. Gather recent reviews, de-identified support excerpts, interview notes, and campaign learning that you may use for this purpose.

Uses: Google Sheets, ChatGPT data analysis, Google Forms

First run: Use one product, one cohort or one customer question. Keep the listed privacy, evidence and review checks.

Back to the Task: Find evidence-backed ad angles

Before you start

Prepare

  • Write the audience, product, market, and campaign decision that the research must support.
  • Gather recent reviews, de-identified support excerpts, interview notes, and campaign learning that you may use for this purpose.
  • Use an organisation-approved ChatGPT workspace and confirm its current data controls before upload.
  • Create a restricted Google Sheet for the working evidence and name the person who will review the analysis.
  • Keep customer evidence inside your own approved tools. AIWorkDex has no intake, file upload, survey hosting, or email capture.

Steps

Work through the method

Use the listed inputs and tools. Check the evidence when you need to verify a step.

  1. Define the decision and evidence boundary

    Write one decision question, the customer group, the market, the date range, and the evidence sources you may use. Record excluded sources and known gaps.

    Step details and sources

    Why it matters: A fixed scope keeps the analysis tied to a real campaign decision and makes its limits visible.

    Input
    The campaign problem, intended audience, product facts, approved evidence sources, and usage permissions.
    Output
    A short research brief with the decision question, cohort, sources, exclusions, owner, and review date.
    Tools in this stepGoogle Sheets

    Evidence for this step

  2. Prepare the minimum useful evidence

    Create one row per review, excerpt, or interview observation. Keep a non-identifying evidence key, source type, date, product context, and exact customer language. Remove names, email addresses, order numbers, and unrelated personal details.

    Step details and sources

    Why it matters: The analysis needs enough context for review. Extra personal data adds avoidable risk and does not improve the angle.

    Input
    Approved reviews, support excerpts, interview notes, campaign learning, and the research brief.
    Output
    A restricted, de-identified evidence sheet with one source item per row.
    Tools in this stepGoogle Sheets

    Evidence for this step

  3. Extract customer tensions with evidence

    Ask ChatGPT to classify only the supplied rows as problems, motivations, objections, desired outcomes, or use cases. Require the evidence key, exact supporting excerpt, confidence, and a short rationale for every label. Keep Unknown as a valid result.

    Step details and sources

    Why it matters: Structured output can reveal repeated customer needs while preserving the row that supports each interpretation.

    Input
    The de-identified evidence sheet, fixed category definitions, and the research brief.
    Output
    A labelled table with source keys, excerpts, categories, confidence, rationales, and Unknown rows.
    Tools in this stepGoogle SheetsChatGPT data analysis

    Evidence for this step

  4. Audit the extracted tensions

    Review every low-confidence and Unknown row. Check a varied sample of the remaining labels against the original evidence. Correct invented, merged, or over-broad themes before counting them.

    Step details and sources

    Why it matters: AI labels can flatten different customer situations or create a pattern that the source material does not support.

    Input
    The labelled table, original de-identified evidence, category definitions, and a named reviewer.
    Output
    A corrected table, review notes, and a record of the rows checked by a person.
    Tools in this stepGoogle SheetsChatGPT data analysis

    Evidence for this step

  5. Build the customer-angle matrix

    Group the reviewed rows without merging distinct situations. For each candidate angle, record the customer situation, tension, desired progress, objection, product relevance, evidence keys, representative excerpts, strength of evidence, gaps, and a creative hypothesis.

    Step details and sources

    Why it matters: The matrix separates a customer angle from the hook, format, or asset used to express it.

    Input
    The audited labels, verified product facts, previous campaign learning, and current brand guidance.
    Output
    A matrix of distinct angle hypotheses with traceable evidence and visible gaps.
    Tools in this stepGoogle SheetsChatGPT data analysis

    Evidence for this step

  6. Review claims before creative work

    For each angle, write the likely express and implied claims. Remove or qualify claims that exceed verified product facts. Send regulated, objective, health, safety, performance, price, or effectiveness claims through the organisation’s normal legal and compliance review.

    Step details and sources

    Why it matters: Customer language can inspire an angle without proving an objective claim about the product.

    Input
    The angle matrix, product substantiation, intended creative message, target markets, and review policy.
    Output
    A claim-safe angle row with allowed wording, prohibited wording, evidence links, and required approvals.
    Tools in this stepGoogle Sheets

    Evidence for this step

  7. Fill a defined gap only when needed

    If the existing evidence cannot answer one important question, run a small consented Google Forms survey. Ask only about that gap. Disable email collection, omit names and order numbers, restrict access, hide response summaries, stop collection when the planned sample closes, and review every response before adding it to the matrix.

    Step details and sources

    Why it matters: A bounded survey can test a missing assumption without creating a broad customer-data collection exercise.

    Input
    One documented evidence gap, consent wording, a suitable customer group, access controls, and a retention decision.
    Output
    A de-identified response sheet, review record, and a clear decision to update or leave the angle unchanged.
    Tools in this stepGoogle SheetsGoogle Forms

    Evidence for this step

  8. Choose the next angle tests

    Select a small set of distinct, supported angles. Give each one a separate creative brief with the audience, tension, promise boundary, evidence, concept, format, brand checks, owner, and test measure. Keep execution variants under their parent angle.

    Step details and sources

    Why it matters: Separate briefs let the team test different customer ideas instead of mistaking new wording or crops for new angles.

    Input
    The reviewed matrix, claim decisions, production constraints, brand guidance, and campaign test plan.
    Output
    Human-approved creative briefs that preserve the link from each test to its angle and customer evidence.
    Tools in this stepGoogle Sheets

    Evidence for this step

Success checks

Check the result

  • Every selected angle links to reviewed evidence keys and representative customer language.
  • The matrix keeps distinct customer situations separate and records Unknown or weak evidence openly.
  • A person checked all low-confidence rows and a varied sample of the remaining AI labels.
  • Each creative brief separates the customer angle from its hook, format, and asset variations.
  • Every express and implied product claim stays within verified support and the required market review.
  • Any survey was consented, narrowly scoped, de-identified, access-controlled, and reviewed by a person.
  • The team records the angle tested, the execution used, and the campaign result without treating one result as universal proof.

Failure modes

What could go wrong

  • The output contains generic hooks with no evidence key. Return to the customer rows and require a supporting excerpt for each theme.
  • Several customer situations collapse into one broad angle. Split them and preserve the context that changes the buying decision.
  • Names, email addresses, or order numbers remain in the working file. Stop the analysis and remove them before sharing or upload.
  • The model invents a theme or alters customer wording. Correct the row and audit similar labels against the original evidence.
  • The matrix ranks themes by count alone. Add evidence quality, source coverage, recency, product relevance, and important minority objections.
  • A customer quote becomes an objective product claim. Remove it from the brief until suitable support and review exist.
  • The team launches a broad survey before reading existing evidence. Stop and document the exact unanswered question first.
  • The survey collects direct identifiers or exposes response summaries. Close it, repair the settings, remove unnecessary data, and review access.
  • New crops or headlines are counted as new angles. Group them as executions under the same customer hypothesis.
  • A person approves the creative without checking the evidence trail. Keep the brief in draft until the review is recorded.

Tools

Choose tools

Google Sheets

A shared spreadsheet for source-linked research, decisions, and operating records that people can inspect and correct.

Check fit, limits and pricing

ChatGPT data analysis

Use ChatGPT to organise a checked spreadsheet or research file into tables you can trace back to the source.

Check fit, limits and pricing

Google Forms

A form for collecting consented screening answers or short accounts of a customer’s recent experience.

Check fit, limits and pricing

Sources

Check the sources

Check what each source supports and where the advice has limits.