---
title: "userflow-knowledge-loop"
slug: "userflow-knowledge-loop"
description: "Knowledge Loop keeps your help center accurate with Audit, Draft, Review and Publish workflows, plus human-reviewed Follow-ups for unanswered Adoption Agent questions."
tags: ["Help Center Automation", "Unanswered Questions"]
updated: 2026-11-04T12:11:20Z
published: 2026-11-04T12:11:20Z
canonical: "help.userflow.com/userflow-knowledge-loop"
---

> ## Documentation Index
> Fetch the complete documentation index at: https://help.userflow.com/llms.txt
> Use this file to discover all available pages before exploring further.

# userflow-knowledge-loop

Keeps your help center accurate using what your users actually ask, and makes sure users your Adoption Agent couldn't help still get an answer from a person. It finds which articles to write, update or retire, drafts them in your team's style, sends them to your reviewer in your team's messaging tool, and publishes only what they approve.

## Download

[](https://cdn.us.document360.io/9697557a-eb5d-476f-9b09-062b0f7bdcbd/Images/Documentation/userflow-knowledge-loop.zip)userflow-knowledge-loop29.37 KB[](https://cdn.us.document360.io/9697557a-eb5d-476f-9b09-062b0f7bdcbd/Images/Documentation/userflow-knowledge-loop.zip)

## What it does

Knowledge Loop runs one loop: **Audit → Draft → Review → Publish**, plus **Export** for a full list of your articles and **Follow-ups** for users whose question went unanswered.

- **Audit**—reads what users asked your Adoption Agent and your support team, threads your team marked for the help center in your team chat, recently announced product changes, and help center searches where your tool shares them. Before calling anything missing, it searches your help center for it. You get a ranked list of articles to create, update or retire, each with the evidence behind it. Every recommendation is specific: when your Adoption Agent missed an answer that's already in an article, you're told the exact fix, such as adding the article as a source, using your users' words in the title, or giving the answer its own section.
- **Draft**—the same report includes drafts of the top three to five new and updated articles, written in your team's style and shown next to their evidence, for your reviewer to approve, edit or skip.
- **Review**—the drafts are sent for approval in your team's messaging tool, in the channel or group you choose. Only a reviewer you name can approve a draft, in the way your team agrees: an emoji reaction, a reply, a task moved to done, or a doc comment. The skill reads the decision back from there.
- **Publish**—approved articles go to your help center tool, or come back ready to paste. Pasted articles are checked on the next run. If your Adoption Agent should use a new article, the skill names the articles to add as sources and reminds you to publish the agent again.
- **Export**—a full list of your articles, for a backup or to find stale ones.
- **Follow-ups**—finds the users who asked your Adoption Agent something it couldn't answer, drafts a personal reply for each, records one follow-up task per user, and sends the replies for review. A person on your team sends every reply.

## What you'll need

- The [Userflow MCP](https://help.userflow.com/docs/mcp) connected, for the Adoption Agent questions, announcement signals and Follow-ups. Without it, the audit still runs on your other sources; Follow-ups needs it.
- Your help center: a connector for your help center tool is best; otherwise its URL
- A named reviewer who approves articles
- Optional: a support inbox, a team messaging tool (to read threads marked for the help center and to send approval requests), and a task or doc tool for the backlog, reviews and follow-ups

## How to use it

- "Audit our help center."
- "What articles are missing?"
- "Update our docs after this release."
- "Publish the approved articles."
- "Export our knowledge base."
- "Follow up on unanswered Adoption Agent questions." or "Run the daily digest."

On the first run, the skill checks which tools you have connected and asks only for what's missing: your reviewer, where approval requests should go (which messaging tool and channel, and whether to mention the reviewer), how they approve, your main product areas, two or three articles you consider good examples, and your company's email domains. The first time you run Follow-ups, it also asks where follow-up tasks go and how often to run it. It saves these as settings so you're not asked again. The default audit window is 7 days, which suits a weekly review.

## Works with the tools you already use

| Role | Examples |
| --- | --- |
| Help center | Zendesk Guide, Intercom Articles, HubSpot, Document360, Notion, Confluence, Help Scout Docs, GitBook, or any website |
| Support inbox | Intercom, Zendesk, Freshdesk, HubSpot Service, Help Scout, Front, or a CSV export |
| Team messaging (optional) | Slack, Microsoft Teams, Google Chat, Discord, Mattermost |
| Task or doc tool (optional) | Asana, Jira, Linear, Trello, Monday, Notion, Google Docs, Confluence |

These are examples, not a full list. If the skill can't tell which tool your team uses for something, it asks. If that tool isn't connected, it suggests connecting it and, in the meantime, gives you text to paste: for example, an approval request you post yourself, with your reviewer confirming in chat.

## Approval in your messaging tool

Approval requests go wherever your team already talks. You choose the tool and the channel, group or thread once, and whether the post should mention your reviewer. Each run, the skill shows you the exact messages and posts only after you say yes: one message such as "4 help center drafts are ready for review", with a threaded reply per draft giving its title, why it's needed, how many points need confirming, and a link to the full draft. Tools without threads get one message per draft.

Only your named reviewer's agreed signal counts as approval, and a reaction on one draft decides that draft only. The posts never include customer names, emails or company names.

## Team-chat threads marked for the help center

When a colleague answers a question in a team-chat thread and it's worth documenting, anyone can post a short marker message in that thread, such as `kbpin`. The skill picks up those threads on its next run and uses the answer as the starting point for an article.

It reads only threads that carry the marker, never whole channels, and looks in public channels unless you choose to include private ones. A colleague's answer is treated as something to confirm, not a checked fact: the draft marks those points for your reviewer, and internal details such as tool names, private links and customer names are kept out of the article.

## Follow-ups for unanswered questions

Follow-ups run daily by default and cover the time since the last run; the first run looks back 7 days unless you choose otherwise. Each run, the skill:

- finds the questions your Adoption Agent couldn't answer, plus account-specific requests it could only point to support, such as a refund or a password reset (and, if you choose, answers users disliked)
- leaves out your own staff and test accounts, and asks you once about any users that look like test or demo addresses
- drafts one reply per user: an answer with a link to the article when there is one, a note on who will help for account-specific requests, or a holding reply marked for your team to confirm when there's no answer yet
- creates one follow-up task per user in your task tool after you confirm the list, then sends the replies for review the same way as article drafts

A user's name and email appear only inside their task, never in channel posts, reports or the saved settings. Anonymous visitors can't be followed up, but their questions still count in the next audit. Replies not yet sent after two days are flagged on the next run. The skill can't start itself on a schedule; if your AI tool can schedule tasks, it offers to set that up.

## Things worth knowing

**Nothing is published without approval.** "Publish everything" or "go ahead" isn't approval on its own: the skill lists what's approved and asks your reviewer to decide on the rest. A live article is never changed before then.

**The skill never contacts your users.** It drafts follow-up replies; a person on your team reviews and sends each one.

**Nothing is posted without your yes.** Approval requests are visible to your team, so the skill shows the exact messages and asks before posting, every run.

**Drafts flag what they can't confirm.** Anything the skill couldn't verify is marked `[CONFIRM]`, and missing screenshots are marked `[SCREENSHOT]`. Claims about security, privacy, data location or legal terms are always marked for confirmation, with a note on who should check them.

**Customer details stay out.** Reports cite conversation IDs and thread links, and paraphrase questions. Names, emails and company names don't appear in reports, drafts or channel posts. The one exception is a follow-up task, which holds the user's name and email so a person can reply.

**When support and the docs disagree, you decide.** The skill shows both versions rather than picking one.

**Adoption Agent changes need a republish.** New sources and edited Flow purpose text are used only after the agent is published again.

**A small window gives thin evidence.** If no topic has several users behind it, the report says so at the top and suggests a longer window or more sources.

## What to do with it

Run the audit weekly and Follow-ups each morning, and let your reviewer work through the drafts in your messaging tool. If a topic is really a short click path inside your product, a walkthrough may answer it better than an article: see [userflow-create-reactive-flows](https://help.userflow.com/docs/userflow-create-reactive-flows). To see what your Adoption Agent couldn't answer, see [userflow-adoption-agent-topics](https://help.userflow.com/docs/userflow-adoption-agent-topics).

## Related

- [userflow-create-reactive-flows](/userflow-create-reactive-flows.md)
- [Adoption Agent analytics: conversations, ratings, and Noteworthy Topics](/ai-assistant-analytics-insights.md)
- [AI Playbooks and Skills](/ai-skills-library.md)