Every Pendo customer starts the same way: by teaching Pendo what their product looks like, page by page and feature by feature, which usually means digging through dev tools for the right CSS selector and then agreeing with your team on what to call the thing you just found.

It works, but never really finishes, because the moment your product ships something new, the whole process starts over. I work on product adoption at Pendo, and I described the feeling on a recent customer webinar as being chased by the boulder from Raiders of the Lost Ark, and anyone who has kept a tagging spreadsheet up to date knows exactly which boulder I mean.

That boulder is the reason Pendo built AI tagging, a feature that watches the DOM on your screen, figures out what's already tagged and what isn't, and drafts the rest for you to accept or reject.

Why tagging is the thing that everything else depends on

Tagging is the input that every other feature depends on in Pendo. Your analytics and the reports you show leadership are only as trustworthy as the tags feeding them, your guides only stay stable when they're pointed at the right page or feature, and your segments only slice visitors correctly when the feature usage behind them is tagged accurately.

Replays work best when you can filter to the moments you actually care about, and now that Leo and Pendo's MCP tooling can query your analytics and hand back answers instead of raw charts, that higher output demands more input to match it. Tag quality is the prerequisite for good analytics and AI, so it’s important to get this right.

With AI tagging, Pendo understands your product as it evolves, automatically

"Our vision with AI tagging is a Pendo that understands your product as well as you do, but eliminates the need for manual tagging," Niamh Jones, lead product manager at Pendo, lead product manager at Pendo and owner of the tagging roadmap, told me.

Three guiding principles drove the work:

  1. Quality and trust come first. Early on, the team chose to surface suggested tags rather than apply them automatically, so AI tagging never adds noise or makes an update you didn't ask for.
  2. A human stays in the loop. AI tagging removes the busywork of finding a selector, drafting a name, and writing a description, but you still approve what gets created. Pendo suggests, and you accept what gets applied.
  3. Every product needs its own configuration. No two DOMs are built the same way, so the team built in room for your naming conventions, your custom HTML attributes, and your specific edge cases, rather than a single one-size-fits-all model.

What it actually looks like inside the Visual Design Studio (VDS)

Open the designer today, and AI tagging is the default option. Before you run it, you can hand it two things that make a real difference in what comes back: your standard naming convention, and any custom HTML attributes your team already uses as unique identifiers, including test IDs like data-cy that double as automation hooks.

Run it on a page, or step through a multi-page flow to capture a modal or a setup wizard, and Pendo returns a list of suggested features, each with a drafted name, an element type, and a confidence score based on how stable the underlying selector is.

High-confidence suggestions are usually built on custom attributes or stable classes, and teams have started using low-confidence tags as a punch list for engineering conversations about which elements still need proper identifiers.

Nothing gets created until you accept it, and updates to tags you've already built never delete the original rule, they add to it, so there's no version of this that quietly breaks your existing analytics.

Tagging is showing up inside Claude, too

In design partner testing right now, Pendo's MCP lets you create and update tags from directly inside Claude. Point a test automation tool like Playwright at your app, tell it which flow to click through, and Pendo processes the DOM, capturing it the same way it does in Visual Design Studio, then returns the suggested tags directly into the chat for you to review and approve.

"We went into design partner phase last week, and we'll probably stay there for the next two to three weeks," Jones told me. Open beta is targeted for mid-to-late September, pending how testing goes.

AI tagging is live now under AI access in your subscription settings, on by default, the next time you open the designer. If you've tried it, Pendo's product team is collecting feedback through a short form shared after each customer session, and they read every response themselves.

Turn it on today. You keep everything you've already tagged.

AI tagging is live under AI access in your subscription settings, ready to go the next time you open the designer.

Every tag suggestion sits in a queue until you approve it, and approving one adds to your rules rather than replacing them. Your existing tags and analytics will carry forward, exactly as they are.

You decide what gets created or updated, tag by tag. Already tried AI tagging? We’re listening to the feedback you drop in the form after each session. We read every response, promise!

P.S. — this blog is based on a User Group. If you prefer to listen vs. read, give it a watch.