MenuStart a projectStart

Customer service

Turn LiveChat transcripts into help centre articles

Chat-to-article automation reviews recent LiveChat conversations, groups them by the question behind them and checks each group against your existing help content. Where nothing answers a common question, it drafts an article from your team’s best replies and saves it as a draft for a person to finish.

An example of how the studio would build it. No live services are connected.

From repeated chat questions to draft help articles.

  1. Step 01

    Chats collected

    Fetch last week’s closed chats and their tags from the LiveChat archive, removing names, emails, phone numbers and order numbers.

  2. Step 02

    Questions grouped

    Group chats by the question behind them, such as delivery to Northern Ireland or changing a booking date, and count how often each appears.

  3. Step 03

    Help pages checked

    Compare each group with your existing help articles in Notion to see which questions have no clear answer yet.

  4. Step 04

    Articles drafted

    Use Claude to draft an article for each gap from your agents’ best answers, marking anything that needs a policy decision.

  5. Step 05

    Owner notified

    Save each draft in a Notion database with its status set to draft, and send the list to the person who looks after help content.

The result

Help pages that answer what customers actually ask.

Drafts are never published automatically. Each one links to the chats it was based on and waits for a person to check the facts against current policy and publish it.

Before we connect it.

Where this is useful

Useful for businesses whose chat team answers the same questions every day, such as online retailers, holiday let companies and subscription services. A sensible first version produces a weekly list of the most common unanswered questions, with drafts added once the grouping looks right.

What we need to connect

  1. 01LiveChat access to read archived chats and their tags, through a personal access token or an app authorised for your account.
  2. 02Your current help content, whether it lives in Notion, a help centre or pages on your website, so gaps can be checked against it.
  3. 03A named person who owns help content and decides on policy questions, plus agreed rules on which personal details must be removed before processing.

What needs careful handling

Chat transcripts contain names, emails, addresses and order details, so these are removed before anything is grouped or drafted, and where the AI processing happens is agreed up front. Some repeated questions point to a problem rather than a missing article, such as a confusing checkout step, and those are listed separately for the owner. A draft built from agents’ replies can repeat an outdated policy, so each one links to its source chats and is checked against current policy before publishing.

Questions people ask.

Does LiveChat have something like this built in?

LiveChat reports on chat volume, tags and satisfaction, and the company behind it offers a separate knowledge base product. This approach focuses on finding gaps against the help pages you already have and drafting articles in your tone, wherever those pages live. We check what your current setup covers before building anything.

How much chat history does it need?

A few weeks of chats is usually enough to see the common questions, and the weekly run keeps the list current after that. If you only have a handful of chats a week, a monthly run makes more sense.

Email received → Person sends it
Automation 64

AI support inbox triage and reply drafts

Question received → Conversation kept
Automation 66

WhatsApp AI answers with human handover

All automation examples