# Testimonial Capture Interview > Runs customer interviews engineered to produce usable testimonials and case-study raw material - a question sequence that elicits before/after specifics and real numbers, a permission and approval workflow, quote-editing ethics (tighten but never fabricate), and a soundbite extraction procedure. Use when a user says "get a testimonial from this customer", "interview our customer for a case study", "our testimonials are all vague praise, help me get better ones", "draft interview questions for a customer success story", or "can I edit this customer quote". Do NOT use for turning the raw material into the finished case study - use case-study-builder instead - or for research interviews with subject-matter experts where the goal is learning, not marketing material - use expert-interview instead. - Skill: `skillmedev/testimonial-capture-interview` (Agent Skill) - Install (CLI): `npx skillmds@latest add skillmedev/testimonial-capture-interview` - Raw SKILL.md: https://api.skillmd.com/api/skills/skillmedev/testimonial-capture-interview/raw - Safety review: pending - Works with: Claude Code, Claude.ai, OpenAI Codex - Category: Marketing & Growth - Author: SkillMedev (https://skillmd.com/u/skillmedev) - Updated: 2026-09-17 - Page: https://skillmd.com/skills/skillmedev/testimonial-capture-interview --- # Testimonial Capture Interview A testimonial that says "great product, great team!" persuades no one, and it is not the customer's fault - it is the interviewer's, because vague questions produce vague praise. The costly mistake this skill prevents is burning a willing customer's only interview slot on questions that yield nothing quotable, then either publishing mush or ghost-writing a quote the customer never said. A well-sequenced interview reliably extracts the before/after story with numbers, and the ethics workflow makes every published word both true and approved. ## Operating procedure The sequence matters most inside the interview itself (Step 3): specifics come before feelings because memory works that way - a customer asked "how do you feel about the product" gives adjectives, but a customer walked through their old workflow first gives numbers, and the feelings that follow attach to those numbers. ### Step 1: Select and brief the customer Pick customers with a measurable outcome and at least 90 days of usage - earlier than that, the "after" hasn't stabilized and the numbers won't survive scrutiny. Send a short brief when scheduling: the purpose (a story about their results, in their words), the time ask (30 minutes), the control they keep (nothing publishes without their written approval of the exact words), and a request to have any relevant numbers handy. The approval promise up front measurably improves both acceptance and candor. ### Step 2: Prepare the specifics you already know Before the call, pull what the account already shows - usage data, support history, the original sales notes on why they bought. Interviewing from zero wastes half the slot on facts you had; interviewing from data lets you ask "your team went from 40 to 300 workflows in Q2 - what happened there?", which is where good soundbites come from. ### Step 3: Run the question sequence Record the call (with consent, stated at the start - recording consent is separate from publication consent and both are required). Then run five phases in order: 1. **Before - the old world (5 min).** "Walk me through how you handled [job] before us." "What did that cost you - time, money, headcount, missed deadlines?" "What finally made you look for something new?" Push for the day-in-the-life texture; the "before" pain is what makes the "after" credible. 2. **Decision (3 min).** "What alternatives did you look at?" "What nearly stopped you from choosing us?" The hesitation quote is marketing gold because prospects share the same hesitation. 3. **After - the new world, with numbers (10 min).** The core. "What does that same workflow look like now?" Then chase every vague claim to a number with the two magic follow-ups: **"Can you give me a specific example?"** and **"Roughly how much - even a ballpark?"** "It's much faster" becomes "reports that took a day now take twenty minutes." Accept ranges and labeled estimates; never accept adjectives as the final answer. 4. **Feelings and stakes (5 min).** Only now: "What would going back feel like?" "What surprised you?" "What would you tell someone in your old situation?" Emotion after evidence produces the quotable line that carries a number's weight behind it. 5. **Logistics (2 min).** Confirm name/title/company attribution preferences, ask whether their numbers can be published or need anonymizing ("a 40 percent reduction" vs the raw figures), and restate the approval workflow. ### Step 4: Extract the soundbites Within 24 hours (memory of tone decays fast), process the transcript: 1. Highlight every sentence containing a number, a named specific, or a strong verb. 2. From those, pull 5-10 candidate soundbites of one to two sentences each - a soundbite must stand alone without the surrounding conversation. 3. Tag each by job: **outcome quote** (the number), **hesitation-overcome quote** (objection handling), **switch quote** (why they left the old way), **recommendation quote** (the tell-a-friend line). 4. Rank by the specificity test: could a competitor's customer have plausibly said this exact sentence? If yes, it ranks low. ### Step 5: Edit within the ethics boundary The rule is **tighten, never fabricate**. Permitted: removing filler ("um," "you know," false starts), fixing grammar the speaker would want fixed, condensing two adjacent sentences into one, and trimming with ellipses that do not change meaning. Forbidden: adding words the customer did not say, strengthening a claim ("faster" does not become "10x faster" without their number), combining sentences from different topics into one implied statement, and changing what any statement means. When a quote needs a fact the customer implied but didn't state cleanly, write the candidate version and ask them to confirm it as theirs - never publish your paraphrase as their words. ### Step 6: Run the permission and approval workflow 1. Send the edited quotes - exact final wording, with attribution line and where they will appear - to the customer in writing. 2. Ask for written approval (email suffices) of the specific wording, not a blanket "can we quote you." 3. If their company requires PR or legal sign-off, route it through your contact and budget 1-2 weeks; enterprise logos usually require this. 4. Log approvals: quote, approver, date, permitted uses. Re-approval is required if wording later changes or the use expands materially (a website quote does not automatically license a billboard). 5. Honor removal requests without argument - a burned reference costs more than any quote is worth. ## Inputs to collect - Which customer, and the measurable outcome hypothesis (default: pull the three accounts with the strongest usage growth and label the outcome a hypothesis until the customer confirms it). - Account data available before the call: usage metrics, tenure, original purchase driver. - Where the material will be used - website, case study, sales deck, ads - because the approval must cover the actual uses. - Recording tooling and consent script. - Whether numbers can be published raw or need anonymizing (default: ask the customer; when in doubt, percentages over absolutes). ## Concrete thresholds - Customer tenure ≥ 90 days before interviewing for outcomes. - Interview length 25-30 minutes; the after-with-numbers phase gets the largest share (~10 min). - Soundbite yield: a good interview produces 5-10 candidates; fewer than 3 with a number or named specific means re-interview or pick a different customer - do not pad with mush. - Process the transcript within 24 hours. - Written approval of exact wording before any publication, no exceptions; enterprise sign-off budget 1-2 weeks. - Specificity test on every published quote: it must contain at least one element (number, named workflow, timeframe) that a stranger could not have invented. ## Worked artifact - interview script ``` TESTIMONIAL INTERVIEW SCRIPT - [FILL: customer name, company] - [FILL: date] Consent: "I'd like to record this so I can quote you accurately - nothing gets published without your written approval of the exact words. OK?" BEFORE (5 min) - Walk me through how you handled [FILL: the job] before [FILL: product]. - What did that cost you - time, money, people, missed deadlines? - What finally made you go looking for something different? DECISION (3 min) - What else did you consider? - What nearly stopped you from choosing us? AFTER - NUMBERS (10 min) - What does that same workflow look like today? - [On any vague claim] Can you give me a specific example? - [On any claim without a number] Roughly how much - even a ballpark? - Your account shows [FILL: data point from Step 2] - what's the story there? FEELINGS (5 min) - What would going back feel like? - What surprised you most? - What would you tell someone still doing it the old way? LOGISTICS (2 min) - Attribution: name, title, company OK? Numbers publishable or anonymized? - Next step: I'll send you the exact quotes for written approval - nothing runs without it. ``` **Contrast pair - what the sequence produces:** Bad (feelings-first interview): *"We love the product. The team is fantastic and it's made a huge difference for us."* - unusable; says nothing a competitor's customer couldn't say. Good (specifics-first interview): *"Month-end reporting used to take my team three full days. Last quarter we closed the books in one afternoon - I got Friday back, every month."* - an outcome, a number, a human stake, and only this customer could have said it. ## Deliverable An interview package: the recording and transcript, 5-10 tagged and ranked soundbites edited within the ethics boundary, the attribution and anonymization decisions, and the written-approval log - raw material ready to hand to case-study-builder. ## Do NOT - **Do NOT ask "how do you like the product?" as an opener.** It produces adjectives; the before-walkthrough produces evidence. - **Do NOT accept a vague claim as final.** Every "much better" gets the specific-example and ballpark follow-ups; unquantified praise is the interview failing, not succeeding. - **Do NOT fabricate, strengthen, or splice quotes.** A customer who finds words in their mouth they never said is a lost account and a public risk; "tighten, never fabricate" has no exceptions. - **Do NOT publish on verbal enthusiasm.** "Sure, quote me!" on a call is not approval of specific wording in writing. - **Do NOT interview customers under 90 days in.** Honeymoon quotes lack outcomes and age badly. - **Do NOT skip recording consent** or conflate it with publication consent - they are separate permissions. ## Quality bar - Every published quote passes the specificity test and traces to the transcript. - At least 3 soundbites carry a number or named concrete outcome. - Edits verifiably within the tighten-never-fabricate boundary (diff the published quote against the transcript). - Written approval logged for the exact final wording and the actual uses. - Attribution and anonymization match the customer's stated preference. ## Escalation and neighbors Hand the package to case-study-builder to construct the narrative asset; this skill ends at approved raw material. For interviews whose goal is learning rather than marketing - expert input, discovery research - use expert-interview, which optimizes for information instead of quotability. Claims in regulated industries (health outcomes, financial returns) may need legal review before publication regardless of customer approval - route those to counsel.