Answer Common Questions in Slack with AI (From Your Own Docs)

That's the exact pattern this use case kills. With Macha, you put an AI agent inside Slack that you can @mention in any channel — and instead of pinging a teammate, people ping the bot. It reads the thread, searches your connected documentation, and replies in-thread with the answer and a link back to the source. Set it up once; it answers forever.

McKinsey's widely cited social economy research found knowledge workers spend about 1.8 hours a day — 9.3 hours a week — just searching for and gathering information. A Slack answer agent is one of the cheapest ways to claw some of that back, because it meets people where they already are.

This post walks the real build, what it's genuinely good at, and the cases where you should not point it at a channel.

What you're building

The shape of this workflow is deliberately small — three moving parts:

The result, from the team's point of view, is a Slack message and a fast, sourced reply:

@you: @macha what's our refund policy for enterprise customers? @macha: Enterprise refund policy: • Full refund within 30 days of purchase • Pro-rated refund for annual plans after 30 days • Custom enterprise agreements may have different terms — check the customer's contract Source: Enterprise Refund Policy (Notion)

No one left Slack. No one DM'd the ops lead. The answer carries a citation, so the asker can verify it and the agent stays honest.

Why do this inside Macha (and not a single-purpose wiki bot)

If you only ever wanted a Slack-to-wiki Q&A bot, there are dedicated tools for it — Question Base, ClearFeed, Perfect Wiki, Slite, Notion's own AI in Slack. They're good at the one thing.

Macha is different in a way that matters once you've lived with a few of these: the Slack answer agent is one configuration inside the same platform that runs the rest of your support automation. The agents that triage your Zendesk or Freshdesk tickets, the agent that looks up Shopify orders, the chatbot on your storefront — they all live in the same place, share the same knowledge sources, and draw from the same credit balance. You're not buying a fourth point tool; you're adding a Slack surface to the AI layer you already run.

And because these are real agents with tools, not a read-only search box, the same pattern extends past pure Q&A: an internal agent can answer and take an action — look up an order, summarize a ticket, pull a Stripe charge — when the question needs more than a doc lookup. (See the Slack integration and the full use-case library for the action-taking variants.)

How it compares to dedicated Slack wiki bots

To be fair, the single-purpose tools are genuinely good, and if a Slack Q&A bot is all you'll ever need, any of them will serve you. Here's an honest read of where each one sits (third-party prices below are from vendor and roundup pages as of June 2026 and may have changed — check the source before quoting):

The honest trade-off: a tool like ClearFeed is purpose-built for Slack-as-helpdesk and Question Base is sharp at curating an FAQ from your threads. Macha's edge isn't out-bot-ing them on that one surface — it's that the same knowledge and the same credit balance also power the agents resolving your customer tickets and the chatbot on your store. If Slack Q&A is a standalone need, a point tool is fine. If it's the fourth AI surface you're standing up, doing it inside one platform usually beats running four.

The build, step by step

The whole thing takes a few minutes if your docs are already in a connected tool. Start a free trial — 7-day free trial, no credit card required — and follow along.

1. Connect Slack and Notion

In Connectors, add Slack (OAuth — you'll approve the Macha bot into your workspace) and Notion. The Notion connector exposes the read tools your agent needs: Search Pages, Get Page, Search Databases, Get Database Entry.

2. Make your docs a knowledge source

Connecting Notion gives the agent tools to query it live, but for the common-questions case you usually also want Notion synced as a knowledge source — Macha indexes the pages so retrieval is fast and grounded, and it stays current as your docs change.

You're not limited to Notion. The same agent can read a Zendesk Help Center, Confluence, Google Docs, a website, or files you upload — so a single Slack bot can cover product docs and internal policy and the public knowledge base at once. See the Notion integration and data sources docs for the full list.

3. Write the agent

Create an agent and tell it, in plain language, what its job is. Macha agents are configured with natural-language instructions, not flowcharts:

"When someone @mentions me in Slack, read the thread to understand the question, search our Notion wiki and help center for the answer, and reply in-thread. Always include the source. If you can't find a confident answer in the docs, say so and suggest who to ask — never guess."

That last sentence is doing real work — it's your hallucination guardrail. More on that below.

4. Give it the right tools

Grant the agent the tools it actually needs and nothing more: Notion's Search Pages and Get Page, and Slack's Reply in Thread. Scoping tools tightly keeps the agent predictable and the credit cost low.

5. Set the trigger to App Mention

Pick the Slack App Mention trigger so the agent only wakes up when it's actually called. Add Direct Message too if you want people to ask the bot privately. Don't set it to fire on every message in a channel — that's noisy and burns credits for nothing.

6. Test before you ship it

Run the agent against sample questions in the test panel and watch how it searches and what it cites. This is where you catch a too-broad instruction or a missing doc before the whole team sees a wrong answer.

When it reads cleanly, point it at a single low-stakes channel first — #ask-ops or #it-help — then widen.

What it's genuinely good at

• Killing the repeat questions. Policies, process, "where do I find X," onboarding basics — anything that has a documented answer, the agent will field on the first ask.
• Answering where the question lives. No tab-switch to a wiki, no separate search tool. The reply lands in the same thread, so the context is preserved and others see it too.
• Staying current. Because Notion is live-synced, updating the doc updates the answer. There's no separate FAQ to maintain in parallel — the docs are the FAQ.
• Surfacing your knowledge gaps. When the agent regularly can't answer something, that's a signal a doc is missing or stale. Pair this with Macha Studies to analyze where it falls short and fix the source.
• Source links by default. Every answer points back to where it came from, which is what makes people trust it — and what stops it from quietly making things up.

Watch-outs — when not to point it at a channel

This is a sharp tool, but it's not magic, and a few situations call for restraint.

• Garbage in, garbage out. The agent is only as good as your docs. If your Notion wiki is contradictory or three reorgs out of date, the bot will confidently repeat the wrong thing. Clean the source first; an answer agent makes bad documentation more visible, not less harmful.
• Judgment calls and approvals. "Can I expense this?" or "Should we make an exception for this customer?" aren't doc lookups — they need a human with authority. Tell the agent to defer these, don't let it improvise policy.
• Sensitive or access-controlled knowledge. The agent can read whatever you connect. Don't sync HR records, comp data, or security runbooks into a knowledge source that a company-wide Slack bot can reach. Scope the docs to what's safe to broadcast in that channel.
• **It's an internal answer agent, not your customer-facing front line.** For end-customer questions, use a Macha chatbot on your site or a ticket-resolving agent on your Zendesk or Freshdesk helpdesk — those have the right tone, escalation paths, and guardrails. This Slack pattern is for your team.
• Don't fire on every message. Keep it on App Mention / DM. An agent that reacts to all channel chatter is annoying and wasteful.
• Confidence over coverage. Instruct it to say "I don't have a confident answer in the docs — try @name" rather than stretch. A bot that admits the gap keeps its credibility; one that bluffs loses the room after the first bad answer.

What it costs to run

Macha bills in credits, and credits are spent per AI action — not per question, deflection, or "resolution." A typical Slack answer is a couple of actions (a knowledge search, then a reply), and each action's cost depends on the model you choose — roughly 0.5–9 credits by model, with the default GPT-5.4 Mini at 1 credit. Cheaper models for routine FAQ lookups, stronger ones where the question is gnarly.

Because the framing is per-action, cost scales with usage, not with some opaque per-seat or per-resolution meter — a quiet week costs less than a busy one. Plans, credit allowances, and top-up packs are on the pricing page, and the trial is 7-day free trial, no credit card required so you can measure your own channel's volume before committing.

FAQ

Does it work in DMs and private channels, or just public ones? Both. The Slack connector supports the App Mention trigger (any channel the bot is in, public or private) and Direct Message (one-to-one with the bot). For private channels, just invite the Macha bot in.

Where does the agent get its answers? From the knowledge sources and tools you connect — Notion is the common one here, but the same agent can read a Zendesk Help Center, Confluence, Google Docs, a website, or uploaded files. It only knows what you've connected; it doesn't browse the open web unless you give it a search tool.

What happens when it doesn't know the answer? Whatever you tell it to do. The recommended instruction is to admit the gap and point to a human rather than guess. Requiring a source link in every answer is the practical guardrail — if there's no doc to cite, there's no answer to give.

Is this the same as a customer-facing chatbot? No. This pattern is for internal teams in Slack. For customers, use a Macha chatbot on your site or an agent that resolves tickets in your helpdesk — different surface, different tone, different escalation rules.

Can it answer from Confluence or Google Docs, not just Notion? Yes. Notion is the example because it's a clean live-sync source, but Macha indexes multiple knowledge types. Connect what you have; the agent searches across all of it.

How is this different from a dedicated Slack wiki bot? The Slack answer agent is one config inside the same platform that runs your ticket agents, storefront chatbot, and action-taking automations — sharing knowledge sources and one credit balance. It's an additional surface on the AI layer you already operate, not a separate point tool to buy and babysit.

Want to see the rest of the patterns? Browse Macha use cases, the Slack integration, or start a 7-day free trial, no credit card required on the pricing page.