Using your past WhatsApp conversations to make the agent better: mining them for questions and answers, writing the knowledge, testing against real messages, and what not to upload

The most common question at setup is "can I just give it my chat history?" The honest answer is that the history is gold and the raw export is mud. Customers' personal data, five years of "ok thanks", and answers that were wrong at the time make a poor knowledge base. The questions and your best answers, extracted, make an excellent one.

By Ritchie, ReplyKit · Published 5 October 2026 · 6 min read

Quick answer

Do not upload raw chat exports. Instead, spend an hour reading the last three months of chats, write down every question that recurs with your best answer to it, note the phrases you use that customers respond to, and add those as knowledge lines and tone instructions. Then test the agent with twenty real past messages, including the awkward ones, and fix the gaps. Afterwards the agent's own unanswered-questions list keeps the loop going. ReplyKit accepts the extracted Q&A as text and shows you what the agent could not answer each week.

Why not the raw exportThe one-hour mining routineTest with real messagesKeep it goingFAQ

Why not the raw export

The one-hour mining routine

  1. Open the last three months of chats, or export them to a file you keep private.
  2. For each chat, note the first real question. Tally duplicates. Twenty chats usually give ten distinct questions; a hundred give twenty.
  3. For each question, find your best past answer and rewrite it as a current, plain, one- to three-line answer. Check prices and policies are current today.
  4. Note the phrases you naturally use ("Can", "No problem", "Let me check") and the ones customers respond well to; they go in the persona, as in persona and tone.
  5. Note what you always hand to a person (complaints, custom quotes); those become hand-over rules.
  6. Paste the Q&A list as a knowledge source. Format guidance is in writing a real FAQ.

Test with real messages

Take twenty real past messages, verbatim, typos and all, and run them through the test chat. Compare the agent's reply with what you actually said. Where the agent is wrong, add a line; where it is too formal, adjust the tone; where it should have handed over, add the rule. Two rounds usually settle it. The routine is in testing before going live.

Keep it going

After go-live, the agent lists the questions it could not answer. Each week, add the answers; each month, read the handovers for questions that should not have needed a person. That is the same mining routine, done for you. The method is described in the teach loop.

Frequently asked questions

Can I upload my WhatsApp chat export to train the agent?

You should not. It contains customers' personal data and years of noise and outdated answers. Extract the recurring questions and your best current answers instead.

How long does the extraction take?

About an hour for three months of chats, producing ten to twenty questions with answers.

How do I capture my tone?

Note the phrases you naturally use and put them in the persona instructions as phrases to use and phrases to avoid.

How do I test it?

Run twenty real past messages through the test chat and compare with what you actually replied.

How does it keep improving?

The agent lists unanswered questions weekly; add the answers and the knowledge grows from real demand.

Your best answers, extracted and tested

Paste the Q&A, test with real messages, watch the unanswered list shrink. 7-day free trial, no card.

Start free
Written by Ritchie, ReplyKitPart of the small team in Malaysia that builds and runs ReplyKit. Writes about WhatsApp automation, AI customer service and small business lead response.

How we know this. This is the routine ReplyKit's onboarding recommends when a business has years of chats and no FAQ, and the teach loop in the product is built around the same extraction. Product figures in this article (limits, prices, per-reply costs) are taken from ReplyKit as it runs today and are re-checked when we update the page. Where we cite outside research, the source is linked below. We sell ReplyKit, so read our product claims with that in mind; we say where a different tool or no tool is the better choice. About ReplyKit.

Sources and further reading

  1. ReplyKit Privacy Policy
  2. WhatsApp Business Messaging Policy · opt-in, opt-out and prohibited-use rules every business must follow

ReplyKit is an independent product and is not affiliated with or endorsed by Meta or WhatsApp. This article is general guidance for running a small business, not legal, financial or regulatory advice.