How Domain Protection became an agent inside GoDaddy Airo

GoDaddy was pivoting Airo from a single AI surface into a platform of product agents. Domain Protection needed one. Here's how I designed the DOP spoke agent from ground up and what it took to teach an AI agent when to speak, what to say, and when to stay quiet.

role

Design Lead

timeline

3 months

scope

Agent UX · Logic design · Copy framework

Agent UX · Logic design ·

Copy framework

impact

+5.8% DOP orders · +3.5K orders/yr

+5.8% DOP orders

+3.5K orders/yr

timeline

3 months

impact

+5.8% DOP orders · +3.5K orders/yr

the initiative.

One AI doing everything
wasn't going to scale

Airo started as a single conversational experience. But one agent can't meaningfully cover Domains, Websites, Email, and Security all at once. So GoDaddy shifted: Airo becomes the orchestrating hub. Each product builds its own spoke agent.

Domains
Agent
Websites
Agent
Email
Agent
DOP
Agent ✦
Compliance
Agent
Payments
Agent
Airo
ORCHESTRATOR
Domains
Agent
Websites
Agent
Email
Agent
DOP
Agent ✦
Compliance
Agent
Payments
Agent
Airo
ORCHESTRATOR
Domains
Agent
Websites
Agent
Email
Agent
DOP
Agent ✦
Compliance
Agent
Payments
Agent
Airo
ORCHESTRATOR

GoDaddy's Airo hub

orchestrates, delegates, holds context

Spoke agents

each product owns its own experience

DOP agent

this project

at a glance.

+

0
0
0
0

%

Relative increase in DOP orders

Relative increase in DOP orders

+

3,200
3,200
3,200
3,200

Extra DOP orders per year

Extra DOP orders per year

1st

Wave of spoke agents shipped inside Airo

1st

Wave of spoke agents shipped inside Airo

*All numbers from controlled A/B experiments run by GoDaddy's analytics team

*All numbers from controlled A/B experiments run by GoDaddy's analytics team

guiding principles.

What guided every
decision in this project

These principles became the shared language between design, eng, and PM.

🧭  Show context first

Before recommending anything, show the user exactly where their domain stands. What's protected, what isn't.

🗣 Get the tone right

Security messaging defaults to fear. This agent had to feel calm and advisory, like a knowledgeable friend, not an alarm.

🎯 Intent over events

Don't trigger on every event. Ask what the user is trying to do right now and only show up when it's genuinely relevant.

the logic.

High-Risk Action flow
when to trigger, what to say

One of many flows was the High Risk Action flow. When a user completes a high-risk action such as changing the domain nameserver, the agent checks DOP status, domain eligibility, and user familiarity before deciding what (if anything) to surface.

the messy middle.

The real work:
what was tried, what was killed, what shipped

01 trigger timing

DOP was triggered at the right moment, but on the wrong surface

The plan was to trigger the DOP recommendation immediately after a high-risk action inside Airo chat.

Problem

Users close chat the moment their job is done. The trigger was firing into an empty conversation.

Solution

Moved the trigger to the GoDaddy overview page. Users go there constantly. The conversation can start where they actually are.

Problem

Users close chat the moment their job is done. The trigger was firing into an empty conversation.

Solution

Moved the trigger to the GoDaddy overview page. Users go there constantly. The conversation can start where they actually are.

Problem

Users close chat the moment their job is done. The trigger was firing into an empty conversation.

Solution

Moved the trigger to the GoDaddy overview page. Users go there constantly. The conversation can start where they actually are.

Problem

Users close chat the moment their job is done. The trigger was firing into an empty conversation.

Solution

Moved the trigger to the GoDaddy overview page. Users go there constantly. The conversation can start where they actually are.

02 copy + tone

From alarming to informative

Trigger timing was solved. Next: what should the agent actually say?

Problem

The risk meter was the first attempt. It looked alarming — users felt something was wrong, not that they were being helped.

Solution

The meter was removed. Copy was rewritten to be neutral and informational, explaining the current state of their domain, not dramatising the risk.

Problem

The risk meter was the first attempt. It looked alarming — users felt something was wrong, not that they were being helped.

Solution

The meter was removed. Copy was rewritten to be neutral and informational, explaining the current state of their domain, not dramatising the risk.

Problem

The risk meter was the first attempt. It looked alarming — users felt something was wrong, not that they were being helped.

Solution

The meter was removed. Copy was rewritten to be neutral and informational, explaining the current state of their domain, not dramatising the risk.

Problem

The risk meter was the first attempt. It looked alarming — users felt something was wrong, not that they were being helped.

Solution

The meter was removed. Copy was rewritten to be neutral and informational, explaining the current state of their domain, not dramatising the risk.

03 chat vs panel

The panel wasn't the answer to everything

We couldn't control what Airo said in chat, but we fully owned the bottom-sheet panel. That made it prime real estate.

Problem

Product wanted to use it everywhere. The logic: more surface, more engagement. But showing DOP information did not need a visual.

Solution

Moved general DOP info to the chat. Panel reserved for moments where seeing state actually changes the decision.

Problem

Product wanted to use it everywhere. The logic: more surface, more engagement. But showing DOP information did not need a visual.

Solution

Moved general DOP info to the chat. Panel reserved for moments where seeing state actually changes the decision.

Problem

Product wanted to use it everywhere. The logic: more surface, more engagement. But showing DOP information did not need a visual.

Solution

Moved general DOP info to the chat. Panel reserved for moments where seeing state actually changes the decision.

Problem

Product wanted to use it everywhere. The logic: more surface, more engagement. But showing DOP information did not need a visual.

Solution

Moved general DOP info to the chat. Panel reserved for moments where seeing state actually changes the decision.

04 logic refinement

Added an extra signal to the logic

For domains without a high-value tag, the recommendation was generic. For high-value domains, the agent began surfacing the estimated domain value explicitly.

Generic recommendation for standard domain

Personalized recommendation for high value domain

Users were more likely to add DOP when they understood what they were actually protecting

Generic recommendation for standard domain

Personalized recommendation for high value domain

Users were more likely to add DOP when they understood what they were actually protecting

Generic recommendation for standard domain

Personalized recommendation for high value domain

Users were more likely to add DOP when they understood what they were actually protecting

Generic recommendation for standard domain

Personalized recommendation for high value domain

Users were more likely to add DOP when they understood what they were actually protecting

05 copy framework

Same message, three tones

Messaging adapted based on what triggered the agent. Tone was a deliberate design decision to make the experience feel personal, each variant mapped to a different user mindset and risk context.

SUPPORTIVE

User-led DOP inquiry

"It's always a smart move to add an extra layer of protection. It helps prevent unintended changes and keeps your domain safe."

NEUTRAL

DNS / nameserver change

"You've just completed an update to your domain. Based on its protection level, here's a quick way to boost security."

FEAR-BASED

Autorenew or privacy OFF

"The change you just made is a high-risk action and could affect your domain's security or settings."

the experience.

Conversation to confirmation.
One uninterrupted flow

Domain context surfaced in Airo, protection state made visible, DOP recommended and purchased, without leaving the conversation.

what I learned.

What building in uncharted territory
taught me

01

Restraint is a design decision

Knowing when not to show the agent mattered as much as what it said. Every PM wanted the panel. Saying no and having a principled reason was part of the design work.

02

Build principles before screens

With no reference point, shared principles became the foundation. They aligned design, eng, and the product before a single screen was drawn.

03

Agentic UX is about judgment

When to speak, what to say, when to stay quiet, these are design decisions, not engineering ones. That's the new skill this kind of work demands.

outcome.

Took Domain Protection from a static upsell into an agent-led experience inside Airo, one that shows up at the right moment, speaks in the right tone, and lets users understand and act without ever leaving the conversation.