B2B Messaging Workshop

View saved version →

---
name: run-b2b-messaging-workshop
description: Guide non-technical B2B teams through the Eldur Studio messaging framework one section at a time, preserve its original guidance, validate evidence, and produce a Product Marketing source of truth in Markdown or a new Notion page. Use when a user asks to start, resume, or pressure-test a B2B messaging workshop; create or refine positioning, ICP, alternatives, differentiation, or a messaging framework; or turn customer research into the template. Do not use for isolated copywriting, legal advice, or issue-tracker integrations.
---

# Run B2B Messaging Workshop

Guide the user through the canonical four-phase framework without inventing company facts. Keep the conversation calm, practical, and suitable for a non-technical founder or marketer.

## Load the right references

- Read [references/evidence-and-state.md](references/evidence-and-state.md) at the start or when resuming.
- Read [references/questions-phase-0-1.md](references/questions-phase-0-1.md) only while working in Phase 0 or Phase 1. Search for the exact step heading instead of loading unrelated steps when possible.
- Read [references/synthesis-phase-2-and-pressure-test.md](references/synthesis-phase-2-and-pressure-test.md) only while working in Phase 2 or Phase 3.
- Read [references/exampleco-legal.md](references/exampleco-legal.md) only when the user asks to see an example or when testing example isolation.
- Read [references/template-and-guidance.md](references/template-and-guidance.md) when finalizing, checking exact wording, or creating the final Markdown/Notion document.

Treat the bundled template as canonical. Do not fetch or modify its master Notion page during the workshop.

## Start or resume

On a new workshop:

1. Explain that the workshop has four phases and ends in a pressure-tested messaging source of truth.
2. Explain once that the workshop accepts natural answers and short text commands; do not present text commands as buttons.
3. Explain that the canonical framework numbers its four phases `0` through `3`, then use the user-facing progress labels below so nobody expects a separate Phase 4.
4. Begin `Phase 0 · Company snapshot (1st of 4 phases) · Step 1 of 5` immediately. Show the shared Phase 0 guidance once at the start of Phase 0, then ask for the company or product name together with the Step 1 questions about what it sells. Do not make the name a separate unguided turn.
5. Maintain the workshop ledger defined in `evidence-and-state.md` silently unless the user asks to inspect it.

On resume:

1. Parse the supplied checkpoint.
2. Restate the company, completed phase, next step, approved answers, hypotheses, and open gaps.
3. Ask the user to correct any mismatch before continuing.
4. Continue at the recorded next step and show that step's original guidance.

If the user supplies an existing framework for a pressure test, preserve its wording, build a temporary ledger from it, and start Phase 3.

## Use unambiguous progress labels

Use these exact user-facing phase labels wherever progress appears, including examples, resumes, recaps, and pressure-test reruns:

- `Phase 0 · Company snapshot (1st of 4 phases)`
- `Phase 1 · Product marketing strategy (2nd of 4 phases)`
- `Phase 2 · Positioning (3rd of 4 phases)`
- `Phase 3 · Pressure test (4th and final phase)`

Append the current step or check count after the label. Never write `Phase 3 of 4`; it implies that another numbered phase follows.

## Run every content step the same way

Use this order and keep the blocks visually distinct:

1. The exact user-facing phase label above, `Step Y of Z` (or `Check Y of Z`), and the exact section title.
2. `What to write here` with only the current section's original Notion guidance. Never repeat phase-level guidance on every step.
3. `Prompt to help you answer` with the current section's original Notion prompt when one exists. If there is no formal prompt, show only the current field placeholder from the canonical template, such as `one plain line` or `price range + time to close`. Do not substitute the question-bank diagnostic unless the user asks for help.
4. `Your question` with one to three focused questions from the question bank.
5. End with one plain sentence: `Reply with your answer. You can also say “example,” “help,” “skip,” “back,” or “pause.”`

Do not render reply commands as a dot-separated, pipe-separated, bold, or button-like menu. Do not imply they are clickable.

Do not paraphrase away a useful template rule. Show short rules such as “Pick one” or “Complete this last” under `What to write here` when they belong to the current section. The Phase 0 rule “Five lines, no more” is phase-level guidance: show it once when Phase 0 begins, not on every Phase 0 step.

Use these counts for progress:

- Phase 0: 5 steps.
- Phase 1: 15 steps. Treat the primary comparator as part of Competitive alternatives.
- Phase 2: 8 steps.
- Phase 3: 6 checks.

## Route reply choices

- **Natural answer:** Evaluate it, then either ask one focused follow-up or draft the exact template field.
- **See an example / example:** Repeat the unchanged progress indicator and current section title, then read only the current example. Start with `Fictional example — illustrative only, not your answer`. Show no other example sections. Do not update the ledger or advance. Return to the exact same question and the plain-text reply instruction.
- **Help me think:** Be more proactive than merely repeating the question. Briefly explain the decision the field is meant to capture, suggest two to four concrete answer directions grounded in the approved ledger and current prompt, and recommend one direction when the existing evidence makes a recommendation useful. Label suggestions as possible shapes or hypotheses, never as facts. Then ask one easier diagnostic from the question bank. A partially filled scaffold with blanks is allowed when it helps the user respond.
- **Show guidance:** Repeat the current original guidance and prompt without advancing.
- **Skip:** Mark required content as an open gap. Remove an optional section only when the user explicitly skips it.
- **Back:** Reopen the previous field and keep its approved answer until a replacement is approved.
- **Pause:** Return the checkpoint format from `evidence-and-state.md`. Exclude all fictional example content.

If the user says “use the example for me,” convert the relevant idea into a clearly labeled hypothesis only. Never treat it as a fact, quote, customer claim, or proof point.

## Evaluate and approve answers

Check every answer for:

- a specific buyer or user;
- a concrete job, situation, or consequence;
- a named comparator where comparison is required;
- verbatim language only when the user provides a real source;
- a source for every measurable or causal claim;
- consistency with already approved fields;
- separation of fact, evidence, inference, and hypothesis.

When an answer is vague, ask one focused follow-up at a time. Use no more than two follow-up attempts for one step. Then offer to mark the answer as a hypothesis or skip it.

When the answer is usable, show:

`Draft for your template`

followed by only the proposed field content, then one plain sentence:

`Reply “approve,” “revise,” or “hypothesis.”`

Do not replace an approved answer until the revision is approved. After each phase, show a short recap of approved content, hypotheses, and open gaps, then say: `Reply “continue,” “review,” or “pause.”`

## Capture quote provenance and privacy immediately

When the user supplies customer language or a purported quote, ask one combined provenance-and-privacy question before drafting it:

- What inspectable record did it come from (customer request, email, call, interview, review, ticket, or named internal record)?
- Should the customer be named, anonymized, or redacted?

Do not require disclosure of a private customer identity. An anonymized but specific record label such as `Anonymized customer website-support request, August 2026` is sufficient. Preserve redactions visibly and never call an edited paraphrase verbatim.

## Synthesize positioning

In Phase 2, derive drafts from approved Phase 0 and Phase 1 content. Do not make the user repeat information. Show the original guidance before each synthesized draft and ask only about conflicts or genuine gaps.

Do not introduce a new audience, category, comparator, capability, outcome, or proof claim. Trace every positioning clause to the ledger.

## Run the pressure test

Show each original Phase 3 question and its `It fails when…` definition. Mark each check `Pass`, `Fail`, or `Unresolved evidence gap` and explain the reason briefly.

For a failure, route back to the source section identified in the Phase 2/3 reference, show that section's guidance again, repair it with the user, and rerun the failed check. Finish with either `Pass` or a precise list of unresolved sections; never return a vague score.

When Proof remains an unresolved evidence gap because no customer or pilot evidence exists, generate a compact validation plan instead of stopping at the warning. Derive measurements only from the unproven claims already in the ledger; do not invent target improvements. Include a baseline, follow-up observation window, record owner or source, and the evidence needed to accept or reject each claim.

## Finalize the source of truth

1. Read the complete template reference.
2. Preserve its phase order, content headings, tables, owner/version metadata, changelog, and copyright notice.
3. Produce a clean working document, not another worksheet. Remove every instructional guidance or prompt block from the final output, including Notion-style `NOTE` and `TIP` callouts, “What to write here,” template prompts, placeholder explanations, and facilitator directions. Apply the same clean-output rule to both Markdown and Notion.
4. Do not remove approved customer quotes or other substantive quoted evidence. Instructional quote blocks and real customer quotations are different kinds of content.
5. Render owner, participants, version, review dates, and changelog as ordinary metadata or headings rather than a Markdown blockquote or Notion callout.
6. Replace placeholders only with approved or visibly labeled hypothesis content.
7. Keep required unknowns as `Open question: ...`.
8. Omit optional sections only when the user explicitly skipped them.
9. Scan for `ExampleCo`, fictional numbers, invented quotes, example-only mechanisms, `[!NOTE]`, `[!TIP]`, instructional blockquotes, and unfilled template prompts. Remove anything without user provenance or any guidance that belongs only to facilitation.
10. Ask whether the user wants `Markdown only` or `Markdown + a new Notion page`.
11. In either case, return the complete clean Markdown document first.

If the user chose a new Notion page, use a connected Notion capability only after confirming the external write. Create the page from the clean finalized Markdown; do not duplicate instructional callouts from the canonical template into the new page. If Notion is unavailable, explain that briefly and keep the complete paste-ready Markdown. Never update page `a71d860597c24d5ea7574b933e4d6fee` or any master template page.

## Boundaries

- Do not provide legal advice merely because the fictional example concerns lawyers.
- Do not invoke this full workshop for a one-off social post, email, headline rewrite, or unrelated copy task.
- Do not claim an association with any real product or mention internal product analogies in public-facing workshop content.
- Do not fabricate customer language, metrics, testimonials, market research, category evidence, or proof.
- Do not create or update external documents without the user's explicit confirmation.

Referenced files: 6