Email Activation in Airo AI Builder
Designing for AI
Vibe Coding
User Research
Role
UX Design Intern
Team
2 PMs, 2 engineers, 2 designers, 1 content designer
Timeline
June - July 2026 (Shipped in 1 month)
Tools
Figma
Cursor
Lyssna
Antares Design System
Context
Email activation among customers with a email credit sat below 70%, under the team's own target.
Airo is GoDaddy's new AI app builder for small business owners. Type a prompt, build a AI generated website, refine it in chat. Many users already possess a professional email address purchased through GoDaddy - but activation lagged the team's OKR, against a goal of lifting it 3–5% for the year. The gap matters for business owners. A professional email is how a new business signals it's credible, and how its customers know to trust it from day one.
I worked with a newly formed squad of engineers, PMs, and designers, to build email activation entry points into the AI builder over the summer, shipping two of them during my internship.
Shipped Metrics:
Activate Email Button
91.44%
204 activations vs. a 70%+ OKR
Post-Publish Modal
95.36%
226 activations, 70%+ OKR
Combined
430
total activations, both entry points, from 6/20 to 7/31
Ideation
We mapped five different points in the AI site building user journey where someone might encounter email activation.
Working with product, I ideated across the builder's full journey rather than guessing at one placement. Chat-based contextual, a post-publish modal, and a dedicated tab were the strongest candidates; during-publish and during-loading growls were secondary options, each competing with a user's attention while something else was already happening. We selected entry points to explore first based on engineering lift and to boost speed to production.
Explored entry points that didn't make the cut [Loading Screen | Dedicated Email Tab]
Validation
Users understood the email activation flow, but unclear completion states and weak value messaging created uncertainty.
I tested a vibe-coded Cursor prototype of the post-publish modal with 12 users on Lyssna, walking through their comprehension of the flow, their understanding of what a professional email was worth, and whether this was even the right moment to ask.
01
Comprehension was strong
8/12 described the task clearly; 6/12 reported no friction navigating the publish-and-activate flow.
02
Completion state was the top pain point
5/12 were unsure when the flow had finished, or what to expect next.
03
Next steps
Refined the completion state and value messaging with content, work that also fed the email-activation agent experience down the line.
Launch
Ship two entry points, a post-publish modal and an Activate Email button, each solving a different moment.
Working with a PM and two engineers, and vibe-coding directly in Airo's production codebase along the way, both flows shipped within about a month. Airo moved fast; design patterns for it were still being established as we built, alongside other teams working the same surface.
Decision
Ship an Activate Email button and a post-publish modal, not just the stronger of the two
Reasoning
The modal catches the highest-intent moment, right after publishing, with clear next steps action that guides the user into further added value for their business.
The button serves everyone who isn't in that moment: someone returning later, or who dismissed the modal the first time.
What's next
The harder problem isn't where to ask anymore, it's teaching the builder when asking feels helpful instead of intrusive.
Past the two shipped surfaces, I spent the second half of the internship on the next step: surfacing activation inside the AI chat itself while users build their site, and inside a new email-management tab, in close communication with the Airo Inbox team.
An AI assistant can't just interrupt with an offer whenever it's contextually plausible. With engineering, I mapped contextual cues for the agent to notice (eg. finishing a domain connection, mentioning "notifications" in chat, completing a related build step) against what it could detect today versus what needed to be explicitly wired in.
Contextual "next steps" widget prompted by the AI agent
Sharing it
I presented the case study to 40+ designers at GoDaddy's UX Collab Forum.
Taking the work from a shipped feature back into a structured case study, process, data, and open questions included, was a great exercise in translating an internal PLG win into something a design audience could critique.
Internship Review Quote
"He led a confident, well-organized case study presentation on Email Activation to the UX Collab Forum, fielding questions and feedback from design leaders."
Reflection
What I'd carry into the next project
On designing for AI-native moments.
AI is something that many small business owners may not trust. Thus, looking for the moments of highest intent and careful user testing helped us find how to boost metrics while maintaining comprehension.
A chat assistant, a modal, a tab, and a loading state all carry different implicit permission to interrupt someone. The distinction between an interruption that reads as context and one that is disruptive was important to work with and design for.
