Most companies write messaging from the inside out.

The product team describes what was built. The founder explains the vision. Marketing turns those descriptions into benefits. Sales then discovers that buyers use completely different language for the problem.

The strongest messaging is rarely invented in a workshop. It is recovered from the market.

Customers explain what was happening before they searched for a solution, what triggered urgency, what alternatives they tried, what made them skeptical, and what outcome finally mattered. The difficulty is that this evidence is scattered across call transcripts, emails, support tickets, community channels, reviews, CRM notes, and private conversations.

Claude is particularly useful as a qualitative research layer across that material-provided the team uses a disciplined method rather than asking for a generic summary.

Build the evidence set

Start with material that records actual customer language:

  • discovery and sales-call transcripts;

  • onboarding and implementation notes;

  • churn interviews;

  • support requests;

  • product reviews;

  • community questions;

  • lost-deal explanations;

  • successful customer emails;

  • competitor reviews and public forum discussions.

Remove information the model should not receive, follow the company's data policies, and separate current customers, prospects, churned users, and non-buyers. Those groups may describe the same product very differently.

Do not ask for "key themes" yet

A general theme summary compresses the evidence too early. Begin with structured extraction.

For every conversation, ask Claude to record:

  1. The customer's situation before looking for a solution.

  2. The trigger that created urgency.

  3. The job the customer was trying to complete.

  4. The cost of leaving the problem unresolved.

  5. Alternatives considered or attempted.

  6. Objections, fears, and perceived risks.

  7. The exact words used to describe the problem.

  8. The desired outcome.

  9. Proof required to believe a solution would work.

  10. Stakeholders involved in the decision.

Require short supporting excerpts and source labels. This allows a human reviewer to distinguish a repeated pattern from a polished inference.

Cluster by buying situation, not demographics

Teams often segment evidence using company size or industry because those fields are already present in the CRM. Those attributes can matter, but they do not necessarily explain behavior.

More useful clusters often emerge around the buying situation:

  • a new executive has been hired;

  • a product is approaching launch;

  • an incumbent vendor failed;

  • new regulation changed the risk calculation;

  • a partnership created unexpected demand;

  • a team is overwhelmed by manual work;

  • revenue pressure created a deadline.

These triggers help explain why two otherwise similar companies move at different speeds.

Convert the findings into a message architecture

Once the evidence has been reviewed, translate it into five layers.

Problem: Use the customer's clearest description of what is not working.

Consequence: State the operational, financial, reputational, or long-term cost.

Change: Explain what becomes possible after the problem is resolved.

Mechanism: Describe how the product creates that change without drowning the buyer in features.

Proof: Match evidence to the buyer's perceived risk.

For example, "AI-powered community intelligence" describes a capability. "See emerging customer complaints across Discord, Reddit, and Telegram before they become a launch problem" describes a situation, consequence, and outcome a buyer can recognize.

Create a disagreement report

One of the most valuable Claude workflows is comparing internal messaging with external language.

Ask the model to identify:

  • phrases the company uses that customers never use;

  • problems customers repeat that the website barely mentions;

  • benefits marketing emphasizes that do not influence decisions;

  • objections sales encounters but content never addresses;

  • segments whose language differs enough to require separate messaging.

This is not an instruction to copy every customer phrase literally. Customers understand their pain better than they understand the architecture of the solution. The company still needs judgment. But messaging should begin from recognizable market reality.

Make voice-of-customer research continuous

Do not conduct this exercise once and declare the positioning complete.

Create a monthly workflow in which new calls and community conversations are classified, compared with the previous period, and reviewed for emerging language. Track which phrases appear in successful deals, which objections are growing, and which promised outcomes correlate with retention.

The value compounds when the findings update sales enablement, landing pages, product education, content priorities, and the questions asked in future calls.

Your customers have already written much of your messaging. The job is to listen carefully enough to find it-and exercise enough judgment to know what deserves to become the company's story.

lowob takeaway: Use Claude to increase the surface area of listening, not to replace contact with the market.