How to Connect an AI Chatbot with Slack in 2026 (Step-by-Step)

Here's a pattern I see in almost every team I talk to.

Someone drops a question in #general. "What's our refund window again?" Three people ignore it. One person answers it wrong. Twenty minutes later somebody finally digs up the right doc link and pastes it in.

That same question got asked last month. And the month before.

The fix isn't a longer wiki. Nobody reads the wiki. The fix is putting the answer where the question already gets asked - inside Slack.

In this guide I'll show you exactly how to connect a WhisperChat AI chatbot to your Slack workspace. It takes about ten minutes, you don't need a developer, and you don't need to host or run anything.

What you actually get after this setup

Before the steps, let me be clear about what this does - because "Slack chatbot" means about five different things in 2026 and most of them aren't this.

This is not a keyword-triggered bot with pre-written replies. It's your real chatbot - the same one on your website - answering from the same trained knowledge base.

So:

Train once, answer everywhere. Update your knowledge base and the Slack answers update too. No second setup to maintain.
It stays quiet. In a channel it only replies when someone @mentions it. In a DM it answers every message.
Answers land in the thread. A channel mention gets replied to inside its thread by default, so a long answer never buries the conversation. You can switch this off if you'd rather it reply in the channel.
Sources are attached. When the answer came from a crawled page, up to three source links get appended, so people can verify instead of just trusting it.
Everything is logged. Slack questions show up in your dashboard History, Insights and Leads with Slack marked as the source.
Nothing to host. Slack pushes events to WhisperChat over HTTPS. There's no bot process sitting on a server that you have to keep alive.

Before you start

Three things:

A trained chatbot in WhisperChat. Slack answers come from the same knowledge base as your website widget, so if it hasn't been trained yet, do that first - it'll just say it doesn't know.
A Slack workspace where you can install apps. Some workspaces make admins approve every new app. If yours does, give your admin a heads up before you start, otherwise you'll get stuck at step 3.
The Premium plan. The Slack integration is a Premium feature.

That's it. Nothing to install on your machine.

The setup, in eight steps

Everything happens in two browser tabs: api.slack.com/apps and your chatbot's Deploy page in WhisperChat. You'll bounce between them twice - credentials come out of Slack, and a Request URL goes back in.

I'd genuinely keep both tabs open. It makes this much less annoying.

Step 1 - Create the Slack app

Go to api.slack.com/apps and click Create New App → From scratch.

Give it a name and pick your workspace. The name matters a little - this is what your team sees when it replies, so "Support Bot" reads better than "test-app-2".

Step 2 - Add the bot token scopes

In the left sidebar, open OAuth & Permissions, scroll to Scopes → Bot Token Scopes, and add all four of these:

chat:write
im:history
app_mentions:read
users:read

Quick note on why each one is there, because people ask:

chat:write - lets the bot post replies
im:history - lets it read DMs sent to it
app_mentions:read - lets it see when someone @mentions it in a channel
users:read - lets it show who asked by name instead of a raw Slack ID

Miss one and you'll hit a missing_scope error later. Add all four now and save yourself the round trip.

Step 3 - Install the app and copy the bot token

Scroll back to the top of the same page and hit Install to Workspace, then approve it.

Slack now shows a Bot User OAuth Token starting with xoxb-. It looks like this:

xoxb-1234567890-1234567890123-xxxxxxxxxxxxxxxxxxxxxxxx

Copy it. Treat it like a password - don't paste it into a shared doc or a screenshot.

Step 4 - Copy the Signing Secret

Go to Basic Information → App Credentials and copy the Signing Secret.

This one's the security piece. WhisperChat checks every incoming event against this secret, so anything that isn't genuinely from Slack gets rejected before it's even read.

Step 5 - Connect it in WhisperChat

Now switch tabs.

Open your dashboard → your chatbot → Deploy → find the Slack card → press Connect.

Paste in:

the bot token (xoxb-…)
the signing secret

Then choose whether channel answers go into a thread. I leave threading on - it keeps busy channels readable. Press Connect Slack.

WhisperChat validates the token and hands you a Request URL. Keep that dialog open, you need it in a second.

Step 6 - Paste the Request URL back into Slack

Back to Slack. Open Event Subscriptions, flip Enable Events on, and paste the Request URL into the field.

Slack immediately fires a signed challenge at that URL. When it goes through, the field turns green and says Verified.

This is the one step nobody can automate for you - Slack has no API for setting an app's Request URL, so it has to be pasted by hand. The WhisperChat dialog tracks whether it worked, so you're not guessing.

Step 7 - Subscribe to the two bot events

Still on the Event Subscriptions page, open Subscribe to bot events and add these two:

app_mention
message.im

Save. Slack might ask you to reinstall the app - do it. That's normal whenever scopes or events change.

Two events, that's all. app_mention covers channels, message.im covers DMs. The bot never sees anything else.

Step 8 - Invite it and ask something

DM the app first - it answers every message in a DM, no mention needed. Good way to confirm it's alive.

For a channel, run:

/invite @your-bot

Then mention it:

@your-bot what are our pricing plans?

The answer shows up in that message's thread, with the pages it came from.

Done. Ten minutes, no code.

What's happening under the hood

You don't need this to use it. But if you're the person who has to sign off on installing a new app in the workspace, this is the part your security lead will ask about.

Every event is checked twice. The Request URL carries a per-chatbot secret, so an event can only ever be attributed to the chatbot it was issued for. On top of that, Slack signs each request with your Signing Secret - WhisperChat recomputes the signature over the raw body and rejects anything that doesn't match, or anything with a timestamp more than five minutes old.

Answers are generated off the request. Slack retries any event that isn't acknowledged within three seconds, and retrieval plus generation takes longer than that. So WhisperChat acknowledges instantly and posts the reply from a background worker as a normal message. That's why you never see duplicate answers from a slow response.

Retries never double-answer. Every Slack event ID gets claimed in an idempotency log before the worker starts. A redelivery is acknowledged but never generates - or bills for - a second answer.

The bot doesn't talk to itself. Slack sends the bot's own posts back as ordinary message events, which would otherwise cause an infinite loop. Anything carrying a bot ID, a message subtype, or our own user ID gets discarded before the pipeline is touched.

Long answers don't get cut off. Slack starts truncating past roughly 3,900 characters, so longer answers get split on line breaks and posted as consecutive messages. Markdown links and bold are translated into Slack's mrkdwn, so the reply reads as formatted text instead of raw asterisks.

What the bot can and can't see

This is usually the real blocker for teams, so let's be direct about it.

It only receives events you subscribed it to - mentions of it, and DMs sent to it. Ordinary channel chatter is never delivered to WhisperChat. It is not reading your team's conversations.

Your credentials stay server-side. The bot token and signing secret are never sent back to the browser - the dialog shows a masked placeholder. Leaving a field empty keeps whatever's already saved.

Disconnecting is immediate. It clears the stored credentials and invalidates the Request URL, so the endpoint starts rejecting every delivery right away. If you want the app gone from the workspace too, remove it at api.slack.com separately.

Common problems, and what they actually mean

"Your Request URL could not be verified." Almost always one of two things: you pasted the URL before pressing Connect in WhisperChat, or you copied it with a character missing. Recopy it from the dialog and try again.

missing_scope error. You skipped one of the four scopes in step 2. Add the missing one, reinstall the app, retry.

A conversation shows a Slack user ID instead of a name. That's the users:read scope missing. Same fix - add it, reinstall.

The bot ignores you in a channel. Two possibilities: it wasn't invited to that channel (/invite @your-bot), or you didn't actually @mention it. In channels it only replies to a direct mention. That's deliberate.

Nothing happens at all. Check that both app_mention and message.im are subscribed, and that you reinstalled the app after adding them.

Does this use my message allowance?

Yes. Each answered question costs one message from your plan, exactly like a widget conversation, with the same multiplier for premium models.

When your subscription credits run out there's a small daily grace allowance before the bot starts declining questions - so it degrades gently instead of going dark mid-conversation.

One thing worth knowing: a user's Slack ID is stored as the conversation's display name only. It's an identity, not contact details someone submitted, so a Slack question doesn't turn into a lead in your dashboard.

Should you actually do this?

Honest answer - it depends on where your questions live.

If most repeat questions come from customers on your website, start with the widget. Slack won't help there.

But if your internal team burns time re-answering the same policy, pricing and process questions - and you already have that stuff documented somewhere the bot can read - this is one of the highest-leverage ten minutes you'll spend this quarter. New hires especially. They ask the same fifteen questions, every single time, and now they can ask a bot at 11pm instead of waiting for you.

FAQ

Do I need a developer for this? No. There's no code anywhere in this guide. If you can copy and paste a token, you can do it.

Do I have to host a bot server? No. Slack pushes events to WhisperChat over HTTPS. There's no process to keep running.

Can one Slack app serve several chatbots? Each connection ties a Slack app to one chatbot. If you want a second chatbot in Slack, create a second Slack app for it.

Does the bot read every message in my channels? No. It only receives the events you subscribed it to - mentions and DMs. Regular chatter never reaches WhisperChat.

Can I turn off threading? Yes, in the connect dialog. Turn it off and channel answers post directly in the channel instead of in a thread.

Do Slack conversations show up in my analytics? Yes. They're logged as chat history with Slack as the source, so History, Insights and Leads all include Slack traffic.

What if my workspace admin has to approve apps? Then step 3 will sit in a pending state until they approve it. Everything else works the same once they do.

Wrapping up

Eight steps. Two browser tabs. No server, no code, no ongoing maintenance.

The thing that actually makes this work isn't the integration - it's the knowledge base behind it. A bot connected to thin documentation is just a slower way to say "I don't know." Spend the time on training it properly and Slack becomes the easiest place your team ever gets an answer.

Ready to set it up?Open your dashboard and head to the Deploy page, or read thefull Slack integration docs if you want the reference version.