---
title: "Draft a customer story from interview notes and approved facts"
description: "AI prompt and skill that turn interview notes into a customer story draft using only approved facts, with every quote and number flagged."
url: https://atulsingh.co.in/ai-playbooks/case-study-from-interview-notes
skill: https://atulsingh.co.in/ai-playbooks/case-study-from-interview-notes/SKILL.md
function: marketing
roles: ["Marketing manager", "Content marketer", "Customer success manager", "Founder"]
industries: ["saas", "professional-services", "manufacturing"]
published: 2026-11-14T20:49:17.024+00:00
updated: 2026-11-15T15:16:16.494303+00:00
---

# Draft a customer story from interview notes and approved facts

This playbook turns your interview notes or transcript into a first draft of a customer story, with problem, approach and result sections. It uses only the facts and numbers the customer has approved. It leaves placeholders for quotes and lists everything that still needs the customer's sign-off. A marketing manager gets a draft ready for review in one sitting, plus a short, specific approval email to send.

**Time saved:** about one to two hours per story draft

## What to gather

- The interview notes or a transcript of the customer call, even if rough.
- The customer's name, and whether they have agreed to be named in public.
- The results the customer has approved for publication, worded the way they approved them.
- A line or two on the customer's business: what they do, where, and roughly how big.
- What your company supplied to them, in plain words.
- Where the story will appear (website page, PDF, LinkedIn post, sales deck), who should read it and the target length.

## Steps

1. Collect the notes, the permission status and the approved results in one place. If the customer has only agreed verbally, say so in the permission field and add the date.
2. Remove phone numbers, email addresses and other personal details the story does not need. Then open the prompt below, replace each {{field}} with your own material and paste it into your company's approved workspace in ChatGPT, Claude or Gemini.
3. Read every section against your notes, including the Headline, Summary and The result. Strike any sentence that adds a benefit or outcome the customer did not state or approve, such as 'this helps the team' or 'replies are now consistent'.
4. Check the 'Needs customer approval' list line by line against the draft. Models often leave out job titles, descriptions such as 'family-run' and incidents taken from the notes. Add anything that appears in the draft but is missing from the list.
5. Remove anything you cannot confirm. Send the approval list and the 'Gaps' questions to the customer in one short email.
6. Ask the model to revise using the customer's replies. Fill the quote slots only with words the customer has confirmed in writing, check the length against your target, and publish only after written approval of the final text.

## The prompt

```text
You are helping a marketing team write a customer story. Use only the material below. Do not add facts, numbers, dates, job titles or quotes that are not in it.

Customer: {{customer_name}}
Permission status: {{permission_status}}
About the customer: {{customer_background}}
What we supplied: {{our_service}}
Where the story will be used, who will read it and how long it should be: {{format_and_reader}}

Interview notes or transcript:
{{interview_notes}}

Results the customer has approved for publication:
{{approved_results}}

Task: Draft the story with these headings: Headline, Summary (two sentences), The problem, What we did together, The result, Quote placeholders.

Rules:
1. Numbers. Use only figures listed under approved results, worded as approved. If the notes mention any other figure, keep it out of the story and list it under Needs customer approval.
2. Quotes. Do not write quotes. Under Quote placeholders, give three to five slots. Each slot names the topic. Where the notes contain the customer's exact words, show them in quotation marks and label them 'verbatim from notes, needs approval'. If the notes are only a paraphrase, write [Quote needed: topic] and nothing more.
3. Describe the problem from the customer's point of view, in their situation. Keep our role factual and modest. Make no claim that we are the best or the first.
4. If the permission status does not clearly allow naming the customer or publishing a result, write [Customer name pending] or leave the result out, and say so.
5. Mark any assumption you had to make with [ASSUMPTION: ...].

After the draft, add three lists:
- Needs customer approval: every name, job title, figure, claim and quote in the draft that the customer must confirm.
- Gaps: questions to ask the customer to fill missing detail.
- Left out: anything in the notes you did not use, and why.

Write in plain, short sentences. Avoid hype words.
```

### Fill in

- `{{customer_name}}`: Customer's business name and the contact person you interviewed, with job title (for example: Sharma Textile Exports, Surat. Contact: Rakesh Sharma, Managing Director)
- `{{permission_status}}`: What the customer has agreed to, and what is still pending (for example: Rakesh agreed by phone on 6 Nov to be named and to share the results below. Written approval of the final draft is still pending.)
- `{{customer_background}}`: A few lines on what the customer's business does and its size (for example: Family-run garment exporter in Surat, about 60 staff, sells to buyers in the UAE and the UK.)
- `{{our_service}}`: What your company did or supplied for this customer (for example: We set up a shared WhatsApp Business catalogue and a simple enquiry-tracking sheet for the sales support team over four weeks.)
- `{{format_and_reader}}`: Where the story will appear, who it is for and the target length (for example: A page on our website for export business owners considering similar help. About 450 words.)
- `{{interview_notes}}`: Paste your interview notes or transcript as they are, rough is fine (for example: Rakesh said buyer enquiries came in on WhatsApp, email and calls, and the team often lost track. Two staff kept separate lists. He said 'we were embarrassed when a Dubai buyer had to ask twice'. Setup took about a month. The team now logs every enquiry in one sheet. He also mentioned sales may have grown, but he was not sure.)
- `{{approved_results}}`: Only the results the customer has approved, with the wording they approved (for example: Every buyer enquiry is now logged in one place. First reply to new enquiries goes out within one working day. Approved by Rakesh Sharma on 6 Nov.)

## Check the result

- Every number in the draft appears in the approved results, with the same wording and unit. Words such as 'twice' or 'about a month' count as figures too.
- The Summary and The result describe results only in the approved wording, with no added benefits.
- The draft contains no quotation marks except words copied exactly from the notes and labelled as needing approval.
- Each sentence in every section can be matched to a line in the notes, the customer background or the approved results.
- The 'Needs customer approval' list includes every name, job title, description, figure, incident and quote that appears in the draft.
- Anything the model assumed is marked [ASSUMPTION: ...] and none of it is presented as fact.
- The customer is named only if the permission status allows it; otherwise the draft shows [Customer name pending].
- The text contains no claims such as best, leading or first unless the customer said so and approved it.
- The story is close to the target length you gave.

## Watch out for

- Invented numbers or quotes: models fill gaps if allowed. The prompt forbids it, but still check each figure and quote yourself before the draft leaves your team.
- Filler claims: models add linking sentences that sound harmless, such as 'this gave the team a shared tool'. Each one is a claim about your work. Delete it or put it on the approval list.
- Incidents from the notes: a moment the customer described, such as a buyer having to ask twice, is still their account. Keep it in a quote slot or on the approval list until they confirm it can be used.
- Permission: a verbal yes is not the same as approval of the final text. Send the finished draft to the customer and wait for written approval before publishing.
- Personal and customer data: interview notes can contain names, phone numbers and commercial details. Remove what the story does not need, use your company's approved AI workspace, and follow your data protection duties under India's DPDP Act or GDPR where they apply.
- Over-claiming: a story can imply that your service caused a result the customer only partly credits to you. Keep the wording to what the customer said and approved.
- Third parties: if the notes mention the customer's own buyers, staff or suppliers, leave them out unless they have agreed to be named.

## Industry notes

- **saas:** Customer stories often rest on usage or time-saving figures. Use only figures the customer has agreed to share, and confirm which product features and plan they are describing.
- **professional-services:** Client confidentiality is often tight. Expect to anonymise the client, and check that the problem description does not identify them by detail.
- **manufacturing:** Stories involve process changes, buyers and sometimes contract terms. Confirm with the customer which plant, product line or buyer details can be shared.

## Before and after

**Before:** Someone on the marketing team listens to the call recording again, copies lines into a document and writes the story from memory. Figures get rounded, quotes get tidied, and the customer finds the mistakes at the approval stage.

**After:** The notes and approved facts go into one prompt, and a structured draft comes back with the quotes left as placeholders. The reviewer checks the approval list against the draft, so the email to the customer is short and specific.

## FAQ

**Can the AI write the customer's quote for me?**

It should not. A quote is the customer's own statement, so the playbook leaves a placeholder or shows words taken exactly from your notes. The customer must confirm the wording before you use it.

**What if the customer has not approved any numbers yet?**

Run the prompt with the approved results field marked 'none yet'. The result section should then describe the outcome in plain words without figures, and any figures from the notes should appear on the approval list. Check that none slipped into the story.

**Is it safe to paste interview notes into ChatGPT, Claude or Gemini?**

Only if your company's rules allow it for this kind of material. Where possible, use a business workspace where your data is not used for training, and remove personal details the story does not need.

**Can I use this when the customer wants to stay anonymous?**

Yes. Write 'anonymous' in the permission status and describe the customer in general terms, such as a Surat garment exporter. Check that no detail in the draft makes them easy to identify.

## Atul's note

When I work with marketing teams on customer stories, the drafts that go smoothly start with facts the customer has already agreed. Writing first and asking later leads to rewrites and awkward emails. I let AI shape the draft, and a person confirms every number, name and quote with the customer before anything goes public.

## Tested on

- Claude Sonnet 5.5, 2026-11-14: pass
- GPT-6, 2026-11-14: pass
- Gemini 4, 2026-11-14: fail

[This page](https://atulsingh.co.in/ai-playbooks/case-study-from-interview-notes) · [Markdown copy](https://atulsingh.co.in/ai-playbooks/case-study-from-interview-notes.md) · [SKILL.md](https://atulsingh.co.in/ai-playbooks/case-study-from-interview-notes/SKILL.md)
