Since Amplitude started A/B-testing placements for its Global Agent chat interface in December 2025, weekly message volume is up 13x. The count of users sending at least one message per week has grown 6.52x.

The product team reports those gains did not come from dropping a chat bar on every page. They came from surface-by-surface experiments that produced a 55% lift on the dashboard Home page and a 2x conversion gain in new-user onboarding.

Senior Product Designer Amy Lin led the work and framed the central tension plainly: "Prompts only work when they're anchored to user intentions. It doesn't matter what your model can do. What matters is what a user wants and whether your chat box can get them there."

Why did the first onboarding test fail?

Chethana Kandula and Alex Yoon started by giving new users a chat box at the top of the page plus a strip of clickable suggestions below it. They expected AI-first onboarding to win. It did not. New signups gravitated to suggested content that mirrored what their teammates used.

The follow-up asked for the user's job title first, then served role-specific prompts. New-user conversion doubled versus the original version. Users who asked three questions in their first seven days retained at higher rates than those who didn't.

"New users trust their teammates (and their activity) to recommend valuable content more than they trust AI," Lin wrote. "Trust, like habits, is a muscle that builds up over time."

When does the Home page need a gut renovation?

Home is Amplitude's highest-traffic surface and most navigation-heavy. Adding a small chat box at the top of the existing dashboard layout produced fewer than 10% engagement. The team diagnosed banner blindness.

Lin then stripped nearly everything off the page and centered a single chat box with one button to return to the legacy dashboard view. Global Agent engagement climbed 55%, statistically significant, with under 3% of users opting back to the old Home.

"The chat box wasn't the problem," Lin said. "Users responded to the chat box. What they didn't respond to was the way we originally surfaced it."

Where should an AI entry point live inside chart creation?

Jingshu Zhao owned this round. Existing chart and dashboard builders have deep muscle memory, so the bar to disruption was high. The first placement test: an always-open AI sidebar beat a corner icon by 16.6%, statistically significant. Users complained the sidebar reopened after they closed it, so the team added a guardrail that respects the close action across sessions.

Two treatments then competed inside the chart creation flow. The first embedded a "Build this chart with AI" entry point directly in the existing workflow. The second placed a chat box in the page's empty-state area — visually prominent, but off to the side.

The empty-state treatment underperformed despite its prominence. The embedded version hit significance and was adopted. The same pattern held for dashboards: placing Global Agent at the start of the flow, above traditional "start with" options, won. Many users had the agent generate a full starter chart set on the spot.

What did Amplitude learn about habit formation?

Two patterns held across surfaces. New users need trust earned through personalization anchored to their job-to-be-done. Existing users default to the route they already trust, so the only reliable way to change behavior is to make the new path the only one in front of them for a moment.

What's the takeaway for teams shipping AI interfaces?

Lin recommends two questions before deployment: "Who would go through this flow?" and "How is the chat interface going to change their behavior?" Without a tracking plan tied to those questions, iteration devolves into guessing.

The team's next lever is showing users what similar users are asking Global Agent on Home, betting that social proof from peers feels warmer than a blank box. The impact is still too early to call.

The pattern from Amplitude points toward AI features that earn a place inside established workflows rather than fighting for attention beside them — a shift that will define which AI products feel like infrastructure in 2026 and which feel like chrome.