Build Customer Voice
Skill8 starsChanged 5 months agoMITView on GitHub
Turn the words your customers actually use into a customer-voice doc: the verbatim phrases, the outcomes they want, the worries that stall them, in their language. Keyless from reviews, testimonials, and call notes you paste in; deepens into a full transcript-mined synthesis when your workspace is connected. Foundation for every marketing strategy artefact.
In this repository101 other files it gives its agents.
- templates/CLAUDE.mdCLAUDE.md
- Add a FieldSkill
- Assemble Content PackSkill
- Assemble PackSkill
- Audit My AutomationsSkill
- Audit My DataSkill
- automate-thisSkill
- Brand My WorkspaceSkill
- Break Down A ChannelSkill
- Build Brand StrategySkill
- Build DocumentSkill
- Build FormSkill
- Build Knowledge Base From DocsSkill
- Build My ProofSkill
- Build My VoiceSkill
- Build Social StrategySkill
- Build SpreadsheetSkill
- Build Work Order ProcessSkill
- Cash Flow ForecastSkill
- Coach My CallsSkill
- Compare DocumentsSkill
- Connect A ToolSkill
- Delegate This WorkSkill
- Describe A ProductSkill
- Design My ScenesSkill
- Design Nurture SequenceSkill
- draft-replySkill
- Email Me A ReportSkill
- Estimate My BASSkill
- Extract DocumentSkill
- Find My Next MoveSkill
- Five Day ChallengeSkill
- Follow-up RadarSkill
- Form RadarSkill
- Get Found OnlineSkill
- Grill Me On This DecisionSkill
- import-from-anywhereSkill
- lead-triageSkill
- Learn My BusinessSkill
- Lint DocumentSkill
- Lint FormSkill
- Lint Nurture SequenceSkill
- log-this-callSkill
- make-it-happenSkill
- Make My VideoSkill
- Make Product DemoSkill
- Make Social PostSkill
- Make ThumbnailSkill
- missed-call-recoverySkill
- Nurture HealthSkill
- Onboard Team MemberSkill
- Outstanding DocumentsSkill
- Outstanding InvoicesSkill
- Package My VideoSkill
- Plan My AdsSkill
- Plan My ContentSkill
- Plan My RoadmapSkill
- Plan My YouTubeSkill
- Prep For CallSkill
- Price My WorkSkill
- The first 60 shown.
The file
We split this skill into the rules it gives an agent
11 rules. About 0 are common, 10 are distinctive, and 1 is very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈10
Distinctive · estimate
This author's own choices. Worth a look.
1
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 81
If the converter reports it isn't installed, relay its one-line install hint and continue with whatever was pasted directly.
- Distinctive: line 85
One good prompt when the well is shallow: If there's little to work with, ask the owner for the few things that unlock the most signal — …
- Distinctive: line 92
Read everything end-to-end: Don't skim a single review.
- Distinctive: line 140
Use the actual words customers used.
- Distinctive: line 156
Every named tool, brand, or alternative, in a table:
- Distinctive: line 200
When done, tell the owner:
- Distinctive: line 207
Keep this report under 200 words.
- Distinctive: line 236
Follow that when, and only when, the workspace is connected.
- Distinctive: line 245
Quote VERBATIM: Don't paraphrase a customer's words into your own.
- Distinctive: line 248
Read every source end-to-end: No skimming, in either mode.
- Very specific: line 261
Don't editorialise: If the evidence suggests the owner's current positioning is off, surface that as a finding — don't rewrite the strategy …
≈ “Common” is an estimate: at least 1 in 5 of the 350,495 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Build Customer Voice
You're building the canonical evidence file the owner's marketing strategy gets grounded in. Everything else (brand voice, positioning, value props, nurture sequences) gets authored FROM this file. If it's wrong, every downstream artefact inherits the error. So it must be made of real words real customers used, never copy you invented for them.
The source of truth for the method is
knowledge/marketing-strategy-method.md
— read its "Layer 2 — Customer voice synthesis" section before starting if
you haven't. (Note: that method's transcript-pull and live-queue layers are
the connected-tier deepener, not a requirement for this skill to run.)
Pick the mode — start with what you have
There are two ways to run this, and the right one depends on what evidence exists right now:
| You have… | Mode | What you produce |
| Reviews, testimonials, call notes, support emails — pasted in or in local files — the default day-one path | Keyless | customer-voice-synthesis.md: the same 10-section doc, built from the words in front of you |
| A connected workspace with recorded call + meeting transcripts | Connected deepener | The same doc, mined from a much larger, richer evidence base |
Default to the keyless mode whenever you don't already have a connected workspace pulling transcripts for you. It needs no accounts and no setup — only the words the owner can hand you. Do NOT stop and demand recorded calls; that's the cold-start failure this skill exists to avoid. The connected transcript mine is the upgrade you offer once it's available, never the price of entry.
Mode A — Keyless (the day-one default)
This runs entirely from material the owner already has: the reviews on their Google or Facebook page, the testimonials on their site, a folder of support emails, notes they jotted after a sales call, a few screenshots of texts from happy customers. No connection, no special access. In a few minutes you give them back the real voice of their market, organised.
A1 — Gather the raw customer words
Pull customer voice from whatever the owner can put in front of you, in this order of richness:
- Anything the owner pastes. Reviews, testimonials, customer emails, support-ticket text, social comments, notes typed up after a call. Treat every pasted block as primary evidence — capture the exact phrasing.
Local files the owner points you at. A folder of saved reviews, an exported reviews CSV, a PDF of testimonials, a Word doc of call notes, a screenshot of a text thread. Convert each one to Markdown first so you're reading clean text, not guessing at a scan:
python ~/.claude/bos-run.py tool markitdown_convert "<path-to-file>"
(The ~/.claude/bos-run.py launcher resolves the install location for you. If it is missing, run python tools/setup.py once from the BOS directory to create it.)
This handles PDF, Word, Excel, PowerPoint, images (OCR), HTML, CSV, JSON. If the converter reports it isn't installed, relay its one-line install hint and continue with whatever was pasted directly. If a file comes back empty (e.g. a scan with no readable text), say so plainly rather than inventing content. 3. One good prompt when the well is shallow. If there's little to work with, ask the owner for the few things that unlock the most signal — for example: "Paste me your last 10 Google reviews, or 3-4 emails from customers who were thrilled, and I'll pull the patterns out." A real doc built from a handful of genuine reviews beats a padded one built from nothing.
Read everything end-to-end. Don't skim a single review. If a pasted block or a converted file is long, read all of it.
Keep the customer's voice, drop the owner's. In call notes and email threads, the owner's / salesperson's own lines are not customer voice — you want what the OTHER person said. Internal team chatter filters out too.
A2 — Write customer-voice-synthesis.md
Offer to write it to marketing-strategy/<BrandName>/customer-voice-synthesis.md
(or alongside the source files the owner pointed you at). Tell the owner
exactly where you wrote it so they can find it.
The file has these 10 sections, in this order. With a smaller keyless evidence base you'll have fewer quotes per theme than a full transcript mine would — that's fine. Write what the evidence genuinely supports; never pad to hit a count.
What downstream reads from each section: §3 jobs-to-be-done captures the Arrival in the buyer's own words (business-method.md §6); §4 worries feed guarantee design — the buyer's biggest fear, inverted (§7.2); §5 competing options are rung 1 of the positioning ladder — real alternatives including "no tool at all" (§14); §6 buying moments feed urgency-as-gain-timing (§7.4, §18); §2 pain themes feed the internal problem list of the Category-of-One build (§7.1), never shipped copy (§18).
1. Who's actually leaving these words
Industries, business sizes, roles, locations — whatever the source material reveals. Quote names + businesses verbatim where they're given. Identify recurring patterns ("repeat customers who'd tried a cheaper option first", "first-timers nervous about cost", "referrals from existing clients").
2. The pain — in their own words
The single most important section. Cluster into themes. Quote VERBATIM
with a footnote-style source tag [Name or "Google review", source-file or
"pasted"]. Aim for 5-15 themes with the quotes you actually have for each.
Examples of theme shapes:
- "Things slipping through the cracks" (leads, follow-ups, info)
- "Doing everything yourself" (overwhelm)
- "Tool fragmentation" (juggling disconnected tools)
- "Tech-light" (explicit non-technical self-description)
- "Cost anxiety" (worry about another monthly bill)
- "Missed calls / enquiries going cold"
- "Marketing money wasted because follow-up doesn't happen"
Use the actual words customers used. Don't invent themes that aren't in the material.
3. Jobs to be done
What END STATE are they trying to reach? Quote them describing it.
4. Worries + hesitations
What stalls the decision? Quote the language — cost, a prior bad experience, fear of complexity, "let me think about it". (Naming the worry here, in this internal evidence doc, is fine — this section is for the owner's eyes, not customer-facing copy.)
5. Competing options / what they're escaping
Every named tool, brand, or alternative, in a table:
| Option | Who's escaping it | Verbatim |
| (tool/brand) | (names) | "(what they said)" |
| ... |
6. The "buying moment"
What triggered a real decision? Just outgrew a cheaper option / just got let down by a competitor / just took over a business / a big job landed. Quote the moment.
7. Vocabulary list — the actual words they use
Flat list of recurring phrases. Marketing copy reaches for these first; invented language second.
- "Slipping through the cracks"
- "On the tools"
- "One-man band"
- "Clunky / overkill"
- "Peace of mind"
- etc. — only the phrases that actually appear in the material.
8. Who this is NOT for
Quote any moment where someone was clearly a poor fit — wrong job, wrong size, wrong expectation. Negative-space ICP. If the evidence doesn't show this, say so rather than inventing a misfit.
9. Top representative quotes
The single most pungent verbatim quotes that capture WHO this audience is. Aim for up to 10 — fewer if the evidence is thinner. Each quote: name (or "Google review") + source.
10. Pattern observations
One short paragraph: which customer types are over-represented? The modal customer profile? Quote the evidence behind it.
A3 — Report back
When done, tell the owner:
- Where the file was written
- Roughly how many real quotes it captured, and from how many sources
- The top 2-3 surprising findings — phrases or themes that contradict how their current website talks about the audience
- The 1 modal-customer profile that emerged most strongly
Keep this report under 200 words. The synthesis file IS the deliverable; your report is the headline.
A4 — Offer the deeper mode (don't make it the entry price)
Once the keyless doc is in their hands, offer the upgrade plainly:
This is built from the reviews and notes you handed me. When your customer conversations are being recorded in one place, I can mine every call and meeting end-to-end and turn this into a much richer version — far more quotes, far more themes. No rush; this doc stands on its own today.
Mode B — Connected deepener (the richer upgrade)
This is the deeper path, and it runs only if your workspace is connected
and your customer calls and meetings are being recorded there. It mines that
full transcript library — many more conversations than anyone would paste by
hand — into the same 10-section customer-voice-synthesis.md, just with a
much larger evidence base behind every theme.
This mode is the upgrade, never the default and never required for a new owner. When no connected transcript library is available, run Mode A — do not stop and demand one.
The connected workflow (how the recorded conversations get pulled and read in
bulk) is documented in the "Layer 2" section of
knowledge/marketing-strategy-method.md.
Follow that when, and only when, the workspace is connected. Everything about
the output is identical to Mode A: the same 10 sections, the same verbatim
discipline. The only
difference is the size and richness of the evidence.
Hard rules (both modes)
- Quote VERBATIM. Don't paraphrase a customer's words into your own.
- Keep the customer's voice, drop the owner's. The synthesis is what the CUSTOMER said, not the owner / salesperson.
- Read every source end-to-end. No skimming, in either mode.
- Never invent a quote, a review, or a testimonial. A fabricated quote is worse than no quote. If a theme has only one real quote, write one. If a section has no evidence, say so.
- No padding. Quality beats count. A thinner doc made of real words beats a fat one made of guesses.
- Real names stay in this internal file. This doc is the owner's own evidence. Don't carry real customer names into anything customer-facing (marketing copy, public posts) — generic placeholders at that boundary.
- This doc is internal evidence; downstream copy is the owner's brand. Naming pain and worries here is fine (§2, §4), it's for the owner. The brand strategy and copy authored FROM this doc are written in the owner's brand voice; the framing and marketing psychology are the owner's choice.
- Don't editorialise. If the evidence suggests the owner's current
positioning is off, surface that as a finding — don't rewrite the strategy
inside this file. That's the next skill's job (
build-brand-strategy).
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

