How to get Claude to analyze your company's WhatsApp chats: the steps and the prompt

At our company we exported our WhatsApp chats with customers and had Claude read all of them: 3,682 chats with 93,886 messages. We got a report showing where we lose customers, which approach brings more bookings, what annoys customers, the most repeated questions, and practical notes on each employee.

This guide has all the steps, and the prompt is ready to copy with one click.

What do you need?

• An iPhone with WhatsApp or WhatsApp Business, because iMazing only works with Apple devices.
• A Mac or Windows computer.
• iMazing, to export the chats.
• A Claude account. The chat file is usually large, so we prefer Claude Code because it can split the file and distribute it across several agents.

Step 1: Download iMazing

Download it from the official site:

imazing.com/download

Before you rely on it, check the trial version's limits and the license price on their site.

Step 2: Export the chats

1. Connect your iPhone to the computer with a cable, open iMazing, and approve the pairing between the two devices.
2. Choose your device, then click WhatsApp on the Overview screen. If you use WhatsApp Business, the steps are the same.
3. Choose Data Access Only if you don't need a full backup, or Full Backup if you want one.
4. Click Start and enter your iPhone passcode. The first time can take 30 minutes or more, depending on how much data you have.
5. Select all the chats by pressing Command + A on a Mac or Ctrl + A on Windows.
6. Click Export at the bottom of the screen, choose CSV, then save the file.

The official illustrated guide is here:

imazing.com/guides/how-to-export-whatsapp-chats-from-iphone-to-your-computer-as-pdf-excel-csv-or-rsmf

Step 3: Protect your customers' data before you upload

The file contains your customers' names and numbers, and possibly sensitive data inside the message text. Don't upload it as is:

• Open the file in Excel, and replace the chat name, the sender's number and the sender's name with codes such as "Customer 1" and "Customer 2". Don't delete these columns, because Claude needs them to group each chat's messages.
• Keep the file that maps the codes to the names on your own machine only.
• Make sure your company's privacy policy and the laws in your country allow this use.

Step 4: Run the analysis

If the file is large: Install Claude Code, put the CSV file in a folder by itself, open Claude Code on that folder, then paste the prompt.

claude.com/product/claude-code

If the file is small: Upload it to a regular chat with Claude, then paste the prompt.

Click the copy button, and edit the type of business in the parentheses:

You are a sales and customer service analyst. Attached is the WhatsApp chat export for my company. The business is (write your type of business here).
Goal: find out why customers don't complete a purchase, and what brings in sales.

How to work:
1. Inspect the file, the number of chats and messages, and the languages the chats are written in. Separate automated messages from staff replies.
2. Calculate with code: time to the first human reply, messages with no reply, and the purchase rate by message time and by reply speed. If any of this depends on matching words, such as payment or booking words, cover every language in the file.
3. Split the chats into parts by date, and have each agent read a full part and classify every chat: confirmed purchase, asked and didn't continue, got no reply, existing customer, cancellation or refund, not relevant. Record where the customer stopped and why. Read each chat in its original language and classify it by what it means.
4. Merge the results into one clear report in English. Quote examples in their original language, and add an English translation when the language is different. The report includes:
- The reasons for not buying, ranked by count, with a verbatim example for each reason
- What annoys customers
- The replies that ended in a purchase, compared with those that didn't
- The 10 most repeated questions, counting the same question in different languages as one, with the purchase rate for each question and a suggested reply in the language customers asked in
- Notes on each staff member: what they do best and what they need to improve, with examples
- Recommendations ranked by impact against effort, and the first 3 steps for this week

Rules:
- Don't invent numbers. Separate figures calculated with code from estimates made by reading, and state the limits of the analysis, including any language you couldn't read reliably.
- Percentages show correlation, not a confirmed cause.
- Staff notes are for development and training, not a final verdict.
- Don't mention customer names or numbers in the report.

Step 5: Review the report before you make any decision

• Open a sample of the chats yourself, and compare them with what the report says.
• Keep the numbers calculated with code separate from the estimated ones.
• Percentages show correlation and don't prove the cause. For example, someone who asks about your location may already have decided to book.
• Use the staff notes for training, and don't base a decision that affects an employee on them alone.

What did we find?

These are our own numbers, and your results will differ depending on your business:

• The problem wasn't price: Only about 35 chats had a customer say the price was too high, while about 550 customers learned the price and went quiet, and nobody got back to them.
• The approach that brought the most bookings: A specific appointment time, then a payment link within the same few minutes.
• Reply speed: 41% of those who got a first reply within a quarter of an hour booked, compared with 24% when the reply was late.
• The most common question: "When is your earliest appointment?" and not "How much?". 58% of those who asked about the appointment booked, compared with 21% of those who asked about the price.
• What annoys customers most: It comes after payment, when the service is late and nobody replies.
• Staff: Each employee has clear strengths, but each one was giving prices and policies that differed from the others.

Try the steps on your own company, and start with the first three recommendations your report gives you.