Turn evidence into a message map

Outcome

Produce an evidence ledger and two problem-based segments without turning guesses into facts.

Start from what you know

AI can rapidly summarize notes, but it can also invent patterns, quotes and customer motivations. Separate raw evidence from interpretation. For real work, use only approved, minimized data. Do not paste customer emails, call transcripts or CRM exports into an unapproved tool. This course uses invented notes labeled R01–R06.

An evidence ledger has source ID, exact observation, interpretation, confidence and limitation. “R03 says no new desk” is an observation. “Budget-aware solo workers may prefer no-purchase steps” is an interpretation. “Most freelancers will buy our service” is unsupported. Repeated mentions in six fictional notes are not market validation.

Segment by relevant need rather than stereotypes: people seeking a no-purchase reset; people navigating a shared or rented space. A person can fit both. Avoid inferring income, gender, health or personal traits. You are organizing hypotheses for better content, not building a covert profile.

Worked message map

Segment hypothesis	Evidence	Need	Message	Proof boundary	Next action
No-purchase starters	R03, R04	A small manageable first step	Try one desk zone before buying storage	Fictional preferences, no time-savings claim	Read sample checklist
Shared/rented spaces	R02, R05	Reversible changes respecting others	Plan a small reset around your space constraints	Not all renters share the need	Use shared-space question list

The message map connects a need to an offer and the evidence supporting it. It also says what we cannot claim. “Double your focus today” fails because neither research nor the service specification proves it.

Prompt template

Use only the supplied synthetic source pack. Return a table with source ID, observation, interpretation, segment hypothesis and limitation. Quote only text present in the pack. Do not infer sensitive traits, invent interviews or describe these notes as real market research. Mark unsupported demand and outcome claims as unknown. Suggest two problem-based messages and cite source IDs.

Exercise

1. Read case-study.md. Write “synthetic training evidence, not real research” above your ledger.
2. Copy all six IDs with exact notes. Add interpretation and limitation columns.
3. Run the template or fill the table manually. Check every quote against the pack.
4. Create two need-based segments. Give each at least two source IDs; overlaps are fine.
5. Write one message per segment, one factual proof boundary and a draft CTA.
6. Red-team the map: highlight any unsupported quantity, guarantee, testimonial or sensitive inference.
7. Replace each issue with a bounded claim or explicit unknown. Save the ledger and map as version 1.

Acceptance check

You should be able to trace every observation to a source ID. The map labels interpretations as hypotheses. No number of fictional notes becomes a “customer success statistic.” A real business would next gather permissioned evidence through an approved research plan; this class does not contact anyone.

Transfer: Before repeating this with real customers, confirm data permissions and tool rules, minimize identifiers and have a human check source fidelity. Keep any opt-in decisions separate from purchase or service records.