Test and Refine Your Conversation | Docs

Before inviting real participants, take the conversation yourself. Testing from the participant's perspective reveals problems that are invisible when reading the outline — awkward phrasing, missing context, a tone that doesn't fit, or a flow that runs too long.

How to Test

Open your conversation and click Invite Participants.
Copy the participation link.
Open it in a new browser tab (or use incognito mode to simulate a fresh participant).
Complete the conversation as if you were a real participant. Don't rush — give thoughtful answers to see how the agent responds.

Pay attention to:

Does the opening feel right? Is the welcome message clear about what this conversation is and how long it takes?
Are the questions relevant? Does the agent ask things that make sense given your answers, or does it feel robotic and scripted?
Is the tone appropriate? Too formal? Too casual? Not matching your brand?
Does it know when to go deeper? When you give a vague answer, does the agent probe further? When you give a thorough answer, does it move on?
How long does it take? If you are bored or fatigued, your participants will be too.
Does the ending feel complete? Is the wrap-up natural? Does the completion flow work (redirect, thank-you message)?

Providing Feedback to the Design Agent

After testing, go back to the design view and tell the agent what to change. You do not need to edit the outline manually — just describe what you want in natural language.

Tone and style:

"Make the tone more casual and friendly"
"The agent is too formal — it should feel like a colleague, not a survey"
"Use shorter questions"

Content and flow:

"Add a question about how they heard about us"
"Remove the section about pricing — it's not relevant for this audience"
"The agent should spend more time on pain points and less on feature requests"
"Make the question about competitors a mandatory question"

Behavior:

"When someone says they're happy with everything, don't keep probing — wrap up"
"If someone mentions a competitor, follow up on that"
"Limit the conversation to 5-7 minutes"

Settings:

"Change the welcome message to mention this takes 5 minutes"
"Enable quick reply suggestions"
"Require sign-in with Google"

Each change generates a new version of the outline. You can test again immediately.

Outline Versions

Every time you refine the outline, Perspective creates a new version with a version number and a change message describing what was updated. This means you always have a history of how the conversation evolved.

Versions let you:

Track what changed — see the change message for each version to understand the evolution of the conversation.
Compare iterations — understand how the goals, questions, or guidelines shifted over time.
Collaborate with confidence — when multiple people work on a conversation, the version history shows who changed what and why.

The version number increments automatically. The change message is generated by the design agent based on what was modified.

Iterate Until It Feels Right

The best conversations go through two or three rounds of test-and-refine. Each round catches different issues:

First test — catches structural problems: wrong questions, missing topics, broken flows.

Second test — catches tone and pacing: too long, too formal, awkward transitions.

Third test — catches edge cases: what happens with short answers? Unexpected responses? Someone who goes off-topic?

When to Stop Refining

You are ready to launch when:

The conversation feels natural from start to finish
The questions match your actual goals
The length is appropriate for your audience (most conversations should be 5-10 minutes)
The completion flow works correctly (redirect, thank-you email, etc.)
You would be comfortable sending this to a real participant

Do not over-optimize. A good-enough conversation launched today beats a perfect conversation launched next week. You can always refine based on real participant responses later.

Testing Tips

Test on mobile. Many participants will take the conversation on their phone. Make sure the experience works well on a small screen.

Test voice mode. If your conversation supports voice, try it. The experience is different from text — pacing, interruptions, and energy all change.

Have a teammate test. A fresh pair of eyes catches things you have become blind to. Ask them to take the conversation without any context about what it should be.

Review actual transcripts. After your first few real participants complete the conversation, read the full transcripts. This is the most valuable feedback — see how real people respond, where they get confused, and what unexpected things come up. Then refine accordingly.