← Back to Blog

AI Chat That Walks Users Through Features After Launch

An AI chat that walks users through features takes a question typed inside your app, like “can I schedule this report?”, moves the user to the page where that feature lives, points out the control, and explains the last step in chat. Most teams file it under onboarding. I think the bigger payoff comes later, with users who are well past their first week: the feature you shipped last month that nobody has found, and the one you moved in a redesign that everybody used to find. In Total Chat, these walkthroughs run for logged-in users on the Starter plan and up, one feature at a time.

Last updated: October 2026

Your launch announcement doesn’t have to teach anymore

Think about how a feature launch usually goes. You ship it, write a changelog entry, send an email with three screenshots, and maybe add a tooltip that fires once. Then you wait. A few power users find it. Most people never open the email, and the ones who did can’t remember where the thing lives three weeks later, when they finally need it. No wonder new features can look unused for months.

That’s a teaching problem dressed up as an awareness problem.

The announcement is doing two jobs at once: telling people the feature exists, and showing them how to get to it. It’s bad at the second job, because the second job only matters at the moment someone needs the feature, and that moment almost never lines up with your send date.

An assistant that can walk users through features changes the split. The announcement only has to do the first job. The second one moves to whenever the user asks.

Two moments where it pays off

New features. An existing user types “can it export to CSV?” or “is there a way to schedule this?” Those “can I…?” questions are the closest you’ll get to a user saying they’d use something if they could find it. A walkthrough answers with the feature itself instead of a paragraph about it.

Moved features. Redesigns break muscle memory. The settings link that lived in the sidebar is now under the avatar menu, and your long-time users are the ones who get stuck, precisely because they knew where it was. “Where did billing go?” is a question a walkthrough answers in one move.

That second case gets less attention than it deserves. New users expect to be a little lost. Your oldest customers don’t, and a redesign that strands them spends goodwill you’ve been building for years.

What a feature walkthrough does in Total Chat

Here’s the sequence, so you know what you’re designing for. When a logged-in user asks about a feature, the walkthrough:

1. Moves them to the page where the feature lives and waits for it to load.
2. Clicks the control that opens the feature if it’s a button or a dialog, or points out the element if it isn’t.
3. Explains the final step in chat, and the user does that part.

It survives the page reload, and if the user clicks or types while it’s running, it pauses. It never fills in forms. Each walkthrough covers one feature; chaining several into a single workflow isn’t built yet, so a longer process runs as separate walkthroughs. The complex workflows guide shows how to split one up, and where an assistant should stop is covered in AI That Navigates Web Apps for Users: How Far It Goes.

All of it runs on a feature map of your app. The map is what tells the assistant a feature exists, which page it’s on and which control opens it. Which brings us to the part teams skip.

A launch checklist for walkable features

A walkthrough can only take users to a feature the assistant knows about. So the work happens before the announcement goes out, not after.

1. Re-scan so the map includes the new feature

Total Chat builds the feature map with its codebase scanner, npx @totalchat/cli scan --upload, which you can run locally or as a GitHub Actions step. The docs include a workflow example that re-runs the scan on every push to main. That’s the setup I’d use, because nobody remembers to re-scan by hand on launch day. The scanner reads Next.js apps only (App Router or Pages Router).

2. Tag the control that opens it

Add a data-guide attribute to the button or menu item that opens the new feature. Pointing at a specific control depends on the scan having captured its selector, and the scanner suggests which elements to tag first. Skip this and the walkthrough can still get the user to the right page; it just may not mark the exact control.

3. Write one knowledge base article in your users’ words

Your team calls it “Scheduled Exports.” Your users will type “send this to me every Monday.” Total Chat’s knowledge base uses semantic search, so the wording doesn’t have to match, but there has to be an article to find. Keep it short: what the feature does, who it’s for, and the one setting people get wrong. Write it before launch. (The URL crawler can draft articles from your live site, but it only seeds the knowledge base. It doesn’t build the feature map.)

4. Point the announcement at the chat

Cut the changelog entry down to what the feature does, plus one line: ask the assistant to show you. That’s it. If you’ve long suspected nobody follows the five-screenshot email anyway, this is the version that doesn’t need them to.

For a redesign, run the scan as part of the same deploy that ships the new layout, so the map and the live pages change together. A renamed button or a moved page that the map doesn’t know about strands the walkthrough, and that’s the exact moment your oldest users are asking for help.

Read what people ask after launch

The questions that follow a launch are better feedback than the launch metrics. Total Chat’s dashboard has conversation analytics, so you can see what people actually ask in the weeks after you ship.

Three patterns are worth watching:

• “Can it…?” about something you already built. The feature exists and people can’t find it. That’s a naming or placement problem, not a demand problem.
• “Where did X go?” after a redesign. If the same one shows up every week, the new location is wrong and the walkthrough is covering for it. Move the thing.
• “X isn’t working.” New features ship with new bugs. When a user says something is broken, Total Chat asks a short follow-up and files a structured report with the page address, browser and recent JavaScript errors, on every plan.

On paid plans, when a conversation exposes a question your knowledge base doesn’t cover, the assistant drafts an article for you to approve, edit or reject. Nothing goes live without your approval. It’s a handy backstop for the article you forgot to write in step 3.

Where this falls short

Fair warning: a chat assistant waits to be asked. A user who has no idea a feature exists won’t ask about it, which is exactly why step 4 still matters. The walkthrough doesn’t replace telling people. It replaces teaching them.

A few more limits to plan around. Walkthroughs are for logged-in users only. The widget on your marketing site answers from your knowledge base and captures leads, but it doesn’t use the page in its answers and doesn’t move visitors around. Navigation isn’t part of the Free plan, where the assistant answers from your knowledge base without moving anyone. And if your app isn’t built on Next.js, the scanner can’t build the map that walkthroughs depend on today.

If most of your feature questions come from people in their first week, the onboarding side is covered in AI User Onboarding Walkthrough: What Actually Works. Setup details and the plans table are in the docs.

Frequently Asked Questions

What does an AI chat that walks users through features actually do?

It answers a “how do I” or “can I” question by taking the user to the feature instead of describing where it is. In Total Chat, a logged-in user’s walkthrough moves them to the right page, opens the control for button or dialog features or points out the element, and explains the final step in chat. It never fills in forms for the user.

Do I need to update anything when I ship a new feature?

Yes. Walkthroughs run on a feature map from Total Chat’s codebase scanner, so re-scan when you ship, ideally as a GitHub Actions step on every push to main. Add a data-guide attribute to the control that opens the feature, and write a short knowledge base article using the words your users will actually type.

Can it walk a user through several features in one go?

No. Each walkthrough covers one feature, and chaining several features into a single workflow isn’t built yet. If a launch touches several features, users ask about each one in turn. Each walkthrough survives the page reload and pauses if the user clicks or types while it’s running, so it doesn’t fight them.

Which Total Chat plan includes feature walkthroughs?

Navigation and walkthroughs are on the Starter plan and every plan above it, per the plans table in Total Chat’s docs. On the Free plan, the assistant answers questions from your knowledge base but doesn’t move users between pages. The codebase scanner and in-chat bug reports are included on every plan, Free as well.

---